Security, privacy and compliance

Your customers trust you. Keepara is built to keep it that way.

Keepara holds the most sensitive thing a business owns: the full history of its customers. So security and privacy are not a layer on top. They are the shape of the system, from the database up.

Walled from every other businessA person sends every messageNever used to train anyone elseStop everything in one click
Your customers' data at the centre, inside four protections: walled from other businesses, seen only by your team by role, every look recorded, and a person sends every message. A PERSON SENDS EVERY MESSAGE EVERY LOOK RECORDED YOUR TEAM, BY ROLE Your customers'history WALLED FROM EVERY OTHER BUSINESS
Designed for the standards that matter

Built for the audits and laws your security team will ask about.

Keepara's architecture, data handling and customer contracts are designed from the start to meet each of these.

To be clear: Keepara does not hold any of these certifications today, and no audit has been carried out yet. We have designed the architecture, the data handling and our contracts to meet them from the start, so each audit confirms what is already built rather than forcing a rebuild. We will begin each one when the time is right, in the order set out under How claims are proven.
In plain words

Six promises, the same for every business.

These are not settings. No customer can switch them off and no price changes them.

Your data stays yours

Your customers' history is walled off from every other business on Keepara. It is never sold, never shared, and never used to teach anyone else's system.

How it works

The wall is enforced by the database itself, not only by application code. Learning stays inside your business forever.

GDPR Art. 28 Processor obligationsCCPA §1798.140(ag) Service provider duties, including no combining across businessesSOC 2 CC6.1 Logical access security

A person sends every message

Keepara drafts and briefs. Your team reads, edits, approves and sends. Nothing reaches a customer on its own.

How it works

Every action type starts at Propose. Any wider autonomy is granted by you, one action type at a time, and nothing can widen its own permissions.

GDPR Art. 22 No solely automated decisions with legal effectEU AI Act Art. 14 Human oversight (high-risk duty, followed by choice)OWASP LLM LLM06 Excessive agency

Only what is needed comes in

Keepara keeps data only about the customers you choose to keep. Everything else is dropped unread.

How it works

Connections are read-only and use the narrowest permission each tool offers. Card numbers and unrelated records are never stored.

GDPR Art. 5(1)(c) Data minimisationCCPA §1798.100(c) Collection reasonably necessary and proportionateSOC 2 P3.1 Collection limited to the purpose

Every look is recorded

Each time anyone or anything reads customer content, it is written to a log that cannot be edited.

How it works

If the log line cannot be written, the read itself fails. Every draft also carries its own record: the facts used, the rules applied, the model and prompt version.

SOC 2 CC7.2 Monitoring for anomaliesISO 27001 A.8.15 LoggingGDPR Art. 5(2) Accountability

One click stops everything

Any user in your business can halt Keepara in one action. It takes effect within 10 seconds and holds through restarts.

How it works

Waiting drafts are frozen, then checked again from scratch after the halt ends.

NIST AI RMF MANAGE 2.4 Means to stop or disengage the systemEU AI Act Art. 14 Human oversight (high-risk duty, followed by choice)SOC 2 CC7.4 Responding to incidents

When you leave, it is gone

At exit, all of your data is returned or deleted, including every backup copy, and you receive written confirmation.

How it works

A deletion register sits outside the database, so a restore from backup can never bring deleted data back.

GDPR Art. 28(3)(g) Delete or return all data at the endCCPA §1798.105 Right to deleteSOC 2 C1.2 Disposing of confidential information
Follow your data

One customer record, from your systems to your exit.

Eight stops. At each one, see what happens to the data and what protects it.

Step 1 of 8

It starts in your own systems

Your calendar, meeting notes, CRM or booking system holds the record. Keepara connects with read-only permission.

What protects it here

  • Read-only access with the narrowest scopes each tool offers
  • Connection tokens held by the connector service, revocable by you at the source
  • Inbound events verified by signature; content always fetched from the source, never trusted from the event

Designed for

SOC 2 CC6.1 Logical access securitySOC 2 CC6.6 Protection at the system boundaryISO 27001 A.5.15 Access controlOWASP LLM LLM06 Excessive agency
Step 2 of 8

Only kept customers get in

Before anything is stored, Keepara checks whether the record is about a customer you keep. If not, it is dropped unread and only the count is kept.

What protects it here

  • Least data in, by design
  • Private customers excluded entirely
  • Drops are counted, so you can see what was refused

Designed for

GDPR Art. 5(1)(c) Data minimisationGDPR Art. 25 Data protection by design and by defaultCCPA §1798.100(c) Collection reasonably necessary and proportionateSOC 2 P3.1 Collection limited to the purpose
Step 3 of 8

It lands inside your wall

The record is stored in your business's own walled space, encrypted at rest and in transit, in the region you choose.

What protects it here

  • Row-level security in the database: a query without your business attached raises an error
  • Encryption at rest; connections only over TLS
  • Secrets held in encrypted settings, one per process

Designed for

GDPR Art. 32(1)(a) Encryption and pseudonymisationISO 27001 A.8.24 Use of cryptographyISO 27001 A.8.3 Information access restrictionCCPA §1798.100(e) Reasonable security
Step 4 of 8

The model sees one customer, for one task

When a draft is needed, the model receives one customer's relevant facts, in a labelled untrusted region. Text from any source is treated as data, never as orders.

What protects it here

  • One customer per model call, refused at the gateway otherwise
  • Only model routes with recorded region, retention and no-training terms
  • The reasoning step has no tools and no network access

Designed for

OWASP LLM LLM01 Prompt injectionOWASP LLM LLM02 Sensitive information disclosureGDPR Art. 28(2) Prior authorisation of sub-processorsNIST AI RMF MEASURE 2.7 Security and resilience evaluated
Step 5 of 8

Every draft passes seven checks, in order

Nothing invented. Nothing from another customer. Nothing stale. No false familiarity. Within contact limits. No harm to a sensitive matter. Your own rules. There is no shortcut and no privileged caller.

What protects it here

  • Every sentence must cite a current, sourced fact
  • Another customer's information is a hard block that no one can override
  • A changed fact or rule voids the pass and the draft is checked again

Designed for

OWASP LLM LLM09 MisinformationOWASP LLM LLM02 Sensitive information disclosureSOC 2 PI1.3 Accurate, complete processingEU AI Act Art. 15 Accuracy, robustness, cybersecurity (high-risk duty, followed by choice)
Step 6 of 8

A person decides

The draft waits for a person on your team. They approve, edit or reject it, and send it from your own tools. By default the message says it was drafted with AI assistance.

What protects it here

  • Approval required for every action type unless you choose otherwise
  • Edits are checked again before they can be sent
  • AI-assistance disclosure on by default

Designed for

GDPR Art. 22 No solely automated decisions with legal effectEU AI Act Art. 14 Human oversight (high-risk duty, followed by choice)EU AI Act Art. 50 Transparency for AI-generated contentNIST AI RMF MAP 3.5 Human oversight defined
Step 7 of 8

Everything is written down

The decision trail keeps what woke the agent, the facts it used, the checks it passed, who approved, and what was sent.

What protects it here

  • Logs cannot be edited
  • No customer content in system logs, alerts or monitoring
  • Every content read logged in the same transaction as the read

Designed for

SOC 2 CC7.2 Monitoring for anomaliesGDPR Art. 5(2) AccountabilityGDPR Art. 30(2) Processor records of processingEU AI Act Art. 12 Record-keeping (high-risk duty, followed by choice)
Step 8 of 8

And it leaves when you do

On your request or at exit, Keepara exports your data in an open format, then deletes it from the live system and every backup, and confirms in writing.

What protects it here

  • Single-customer erasure for individual requests
  • Deletion register re-applied after any restore
  • Written confirmation of deletion

Designed for

GDPR Art. 17 Right to erasureGDPR Art. 20 Right to data portabilityGDPR Art. 28(3)(g) Delete or return all data at the endCCPA §1798.105 Right to deleteISO 27001 A.8.10 Information deletion
Who sees what

Pick who is looking.

The question every security review starts with. Choose a viewer and see exactly what is visible to them.

Customer history
Drafts and briefs
Decision trail
Access log
IDs and counts
The sent message

Can seeLimitedNever
For security, privacy and legal teams

The detail behind every promise.

Every control below names the requirement it is designed for: the regulation article, the audit criterion or the standard's clause. Designed for means the control is built into Keepara's architecture to meet that requirement. Audits and certifications follow in the order set out under How claims are proven.

42SOC 2Security standard
38ISO 27001Security standard
24NIST CSFSecurity standard
30GDPRPrivacy law
12CCPAPrivacy law
10EU AI ActAI law
17NIST AI RMFAI standard
19ISO 42001AI standard
17OWASP LLMAI security practice

Number of controls on this page mapped to each framework, across 163 distinct requirements.

The architecture

Every security question gets one of three answers.

It is a platform invariant, it is your policy, or it is assurance. Nothing is an exception. A concession for one customer is that customer's own policy setting, recorded like any other.

Layer 1
Platform invariantsTen commitments true for every business, forever.

Not configurable, not negotiable at any price. A setting that would breach one cannot exist.

Layer 2
Your policyYour choices, inside the invariants.

Every setting starts at its most protective. Your admins widen it, and every change is logged where nobody can edit it.

Layer 3
AssuranceHow every claim is proven, and to whom.

Each requirement is traced to a named test, evaluation, runtime guard or review. Outside proof follows in a fixed order.

Layer 1

Ten platform invariants.

True for every business, forever. Choose one to see how it is enforced and which requirements it is designed for.

Invariant 1 of 10

Tenant wall

A customer exists inside one business only. Businesses are walled apart by the database, not only by the code.

How it is enforced

Row-level security on every table; a transaction with no business attached raises an error. The isolation suite checks every pair of 50 simulated businesses and confirms that a wall opened on purpose is caught.

Designed for

SOC 2
CC6.1 Logical access securityC1.1 Protecting confidential information
ISO 27001
A.8.3 Information access restrictionA.5.15 Access control
NIST CSF
PR.AA Protect: identity, authentication, accessPR.DS Protect: data security
GDPR
Art. 32(1)(b) Ongoing confidentiality, integrity, availability, resilience
CCPA
§1798.140(ag) Service provider duties, including no combining across businesses
OWASP LLM
LLM02 Sensitive information disclosureLLM08 Vector and embedding weaknesses

Where it lives in the architecture

Invariant 2 of 10

Tenant-local learning

Nothing learned from one business's data ever helps another business or any shared model.

How it is enforced

Learning comes only from explicit feedback inside the business: approvals, edits, rejections. Each lesson keeps its author and can be reversed. Model routes that train on inputs cannot be used.

Designed for

GDPR
Art. 5(1)(b) Purpose limitationArt. 29 Processing only on instructions
CCPA
§1798.140(ag) Service provider duties, including no combining across businesses
ISO 42001
A.7 Data for AI systems
NIST AI RMF
MEASURE 2.10 Privacy risk examined
OWASP LLM
LLM04 Data and model poisoning

Where it lives in the architecture

Invariant 3 of 10

Untrusted content boundary

Text from any source is data, never orders. It enters a model only inside a labelled untrusted region, and only schema-valid, sourced facts come out.

How it is enforced

Extraction returns a fixed schema and nothing else. The reasoning step has no tools and no network. Drafting sees facts, never raw source text. A red-team set runs on every prompt or model change, with a pass mark of 100%.

Designed for

OWASP LLM
LLM01 Prompt injectionLLM05 Improper output handlingLLM07 System prompt leakage
NIST AI RMF
MEASURE 2.7 Security and resilience evaluated
EU AI Act
Art. 15 Accuracy, robustness, cybersecurity (high-risk duty, followed by choice)
ISO 27001
A.8.26 Application security requirements
ISO 42001
A.6 AI system life cycle

Where it lives in the architecture

Invariant 4 of 10

One checkpoint

Every outward action passes the seven checks, in order, with no shortcut and no privileged caller.

How it is enforced

A structural test fails the build if any route reaches a user's screen without the check chain. A pass counts only for the exact draft, facts, rules and settings it ran against.

Designed for

SOC 2
PI1.3 Accurate, complete processingCC5.2 Technology controls chosen and built
OWASP LLM
LLM09 MisinformationLLM02 Sensitive information disclosure
NIST AI RMF
MEASURE 2.5 Validity and reliability shown
EU AI Act
Art. 15 Accuracy, robustness, cybersecurity (high-risk duty, followed by choice)
ISO 42001
A.6 AI system life cycle

Where it lives in the architecture

Invariant 5 of 10

A person sends by default

Every action type starts at Propose. Autonomy comes one action type at a time, after a track record, with a recall window. Nothing widens its own permissions.

How it is enforced

Approval-free sending, if a business ever allows it, is a state of the last check, never a separate route. Learning cannot touch check settings, caps or action types.

Designed for

GDPR
Art. 22 No solely automated decisions with legal effect
EU AI Act
Art. 14 Human oversight (high-risk duty, followed by choice)
NIST AI RMF
MAP 3.5 Human oversight defined
ISO 42001
A.9 Responsible use of AI systems
OWASP LLM
LLM06 Excessive agency

Where it lives in the architecture

Invariant 6 of 10

Total visibility

Every content read is logged. Every wake, draft and check has a decision trail. The logs cannot be edited.

How it is enforced

The access log is written in the same database transaction as the read; if it cannot be written, the read fails. Each decision records its trigger, the facts used, the rules fired, and the model and prompt version.

Designed for

SOC 2
CC7.2 Monitoring for anomalies
ISO 27001
A.8.15 Logging
NIST CSF
DE.CM Detect: continuous monitoring
GDPR
Art. 5(2) AccountabilityArt. 30(2) Processor records of processing
EU AI Act
Art. 12 Record-keeping (high-risk duty, followed by choice)
NIST AI RMF
MEASURE 2.8 Transparency and accountability examined

Where it lives in the architecture

Invariant 7 of 10

No content outside the wall

System logs, alerts and monitoring carry IDs and counts only. People in the operator role see IDs only.

How it is enforced

Logging uses an allow-list of fields; error text is stripped of content; notifications name no customer. Planted test values prove nothing escapes.

Designed for

ISO 27001
A.8.11 Data maskingA.8.12 Data leakage prevention
SOC 2
C1.1 Protecting confidential informationCC6.3 Role-based access and least privilege
GDPR
Art. 25 Data protection by design and by defaultArt. 32(4) Staff act only on instructions
OWASP LLM
LLM02 Sensitive information disclosure

Where it lives in the architecture

Invariant 8 of 10

Exit means gone

At exit all data is returned or deleted, including every backup copy, with written confirmation.

How it is enforced

An append-only deletion register outside the database records each deletion before and after it runs, and is re-applied first after any restore. A single customer's data can be erased on its own.

Designed for

GDPR
Art. 17 Right to erasureArt. 28(3)(g) Delete or return all data at the end
CCPA
§1798.105 Right to delete
SOC 2
C1.2 Disposing of confidential informationP4.3 Secure disposal of personal information
ISO 27001
A.8.10 Information deletion

Where it lives in the architecture

Invariant 9 of 10

Least data in

Only data about customers the business keeps is stored. Everything else is dropped unread, and the drop is counted.

How it is enforced

A keep-list filter runs before storage. Connectors hold read-only permissions. Card data and unrelated records never enter.

Designed for

GDPR
Art. 5(1)(c) Data minimisationArt. 25 Data protection by design and by default
CCPA
§1798.100(c) Collection reasonably necessary and proportionate
SOC 2
P3.1 Collection limited to the purpose
ISO 27001
A.5.34 Privacy and protection of personal data

Where it lives in the architecture

Invariant 10 of 10

One-step halt

Any user in a business stops waking, drafting and model calls in one action, within 10 seconds. It survives a restart. Held drafts are checked again afterwards.

How it is enforced

Halt state lives in the database and is checked at job claim, before a model call is sent, and before its result is saved. A database notification cancels open model requests.

Designed for

NIST AI RMF
MANAGE 2.4 Means to stop or disengage the system
EU AI Act
Art. 14 Human oversight (high-risk duty, followed by choice)
SOC 2
CC7.4 Responding to incidents
ISO 42001
A.9 Responsible use of AI systems
OWASP LLM
LLM06 Excessive agency

Where it lives in the architecture

Layer 2

Your policy, starting at the safest setting.

Your business makes these choices. Each starts at its most protective, can be widened only by an admin, is logged where nobody can edit it, and can never go below an invariant.

AreaWhat you controlThe default
Who sees whatRole-based access: which roles see which customers and which kinds of content.Least privilege. Each person sees only what their role needs.
Contact rulesCaps and spacing per customer, private customers, sensitive-matter rules, and which action types are Propose, Automatic or Never.Every action type at Propose. Opt-outs are absolute and cannot be widened.
Data rulesHow far back history is imported, how long it is kept, and which AI model routes may be used.Shortest useful history; only routes with recorded no-training terms.
Data residencyWhere your data lives and which regions may process it, including several locations with their own residency terms.Your chosen region.
AI disclosureWhether outgoing messages carry a line saying they were drafted with AI assistance.On. You may change it where your own law allows.
Your own rulesPlain-language rules every draft must follow, enforced as the seventh check.None until you add them.
The control map

51 controls, each mapped to the requirements it serves.

Filter by framework to see every control designed for SOC 2, ISO 27001, GDPR, the EU AI Act or any other. Or search for the control you need.

ControlHow Keepara does itDesigned for
Governance
Three-layer security architectureEvery security question has one of three answers: a platform invariant, your policy, or assurance. Nothing is an exception, and pilot choices are recorded as that customer's policy.
SOC 2
CC5.2 Technology controls chosen and built
ISO 27001
A.5.1 Information security policies
NIST CSF
GV.PO Govern: policy
GDPR
Art. 25 Data protection by design and by default
ISO 42001
A.2 Policies for AI
Threat model with named defencesNineteen named threats, each with the defence that stops it and the signal that detects it, from prompt injection to a stolen phone to a supplier breach. Reviewed when the system changes.
SOC 2
CC3.2 Risk identification and analysis
ISO 27001
6.1.2 Information security risk assessment (clause 6.1.2)
NIST CSF
ID.RA Identify: risk assessment
NIST AI RMF
MAP 5.1 Likelihood and size of impacts identified
ISO 42001
A.5 Assessing impacts of AI systems
Every requirement traced to a checkAll 247 requirements map to a named check of five kinds: test, safety test, evaluation, runtime guard or review.
SOC 2
CC4.1 Ongoing evaluation that controls work
ISO 27001
A.8.29 Security testing before release
NIST CSF
ID.IM Identify: improvement
NIST AI RMF
MEASURE 2.5 Validity and reliability shown
ISO 42001
A.6 AI system life cycle
Security review before go-liveA review of sign-in, roles, the tenant wall, secrets, public addresses, event receivers, content security policy, process privileges and red-team results. Every finding is fixed or accepted in writing, with its date.
SOC 2
CC4.1 Ongoing evaluation that controls workCC8.1 Change management
ISO 27001
A.8.29 Security testing before release
GDPR
Art. 32(1)(d) Regular testing of security measures
NIST CSF
ID.IM Identify: improvement
Excluded high-risk usesKeepara is not offered for credit, insurance pricing, employment, essential-service eligibility, or any decision with legal effect on a person. Contracts prohibit these uses.
EU AI Act
Art. 5 Prohibited AI practicesArt. 6, Annex III High-risk classification
GDPR
Art. 22 No solely automated decisions with legal effect
CCPA
ADMT regulations California rules on automated decisionmaking technology
NIST AI RMF
MAP 1.1 Intended purpose and context documented
ISO 42001
A.9 Responsible use of AI systems
AI literacy and role trainingThe team building Keepara and the people approving drafts are trained on what the system does, what it cannot do, and how to stop it.
EU AI Act
Art. 4 AI literacy
SOC 2
CC1.4 Competent, trained people
ISO 27001
A.6.3 Security awareness and training
NIST CSF
PR.AT Protect: awareness and training
NIST AI RMF
GOVERN 2.2 AI risk training for staff and partners
ISO 42001
A.4 Resources, including people and skills
Isolation and ownership
Database-enforced tenant wallRow-level security on every table. A transaction with no business attached raises an error. The isolation suite runs every pair of 50 simulated businesses.
SOC 2
CC6.1 Logical access securityC1.1 Protecting confidential information
ISO 27001
A.8.3 Information access restriction
NIST CSF
PR.DS Protect: data security
GDPR
Art. 32(1)(b) Ongoing confidentiality, integrity, availability, resilience
CCPA
§1798.140(ag) Service provider duties, including no combining across businesses
OWASP LLM
LLM08 Vector and embedding weaknesses
Tenant-local learningLessons come only from your own team's explicit feedback, carry their author, can be reversed, and never leave your business. No model route that trains on inputs can be used.
GDPR
Art. 5(1)(b) Purpose limitationArt. 29 Processing only on instructions
CCPA
§1798.140(ag) Service provider duties, including no combining across businesses
ISO 42001
A.7 Data for AI systems
NIST AI RMF
MEASURE 2.10 Privacy risk examined
OWASP LLM
LLM04 Data and model poisoning
One customer per model callEach model call carries one customer's content, or one joint meeting's, and the gateway refuses anything else. A fact said by one customer about another is never used for either unless marked shared.
OWASP LLM
LLM02 Sensitive information disclosure
SOC 2
C1.1 Protecting confidential information
ISO 27001
A.8.12 Data leakage prevention
GDPR
Art. 32 Security of processing
You own your data and what is derived from itYour customers' records, the agent's memory of them and every insight derived from them belong to your business. Keepara processes them only to serve you.
GDPR
Art. 28 Processor obligationsArt. 29 Processing only on instructions
CCPA
§1798.140(ag) Service provider duties, including no combining across businesses
ISO 42001
A.10 Third-party and customer relationships
Identity and access
Sign-in with two-step verificationOnly accounts your admin adds can sign in. Two-step verification is checked and recorded for every user. Repeated refused sign-ins raise an alert.
SOC 2
CC6.1 Logical access securityCC6.2 Adding and removing users
ISO 27001
A.8.5 Secure authenticationA.5.17 Authentication information
NIST CSF
PR.AA Protect: identity, authentication, access
Least privilege, by roleEvery role starts with the least it needs. Widening is an admin action, logged where nobody can edit it.
SOC 2
CC6.3 Role-based access and least privilege
ISO 27001
A.5.15 Access controlA.5.18 Access rightsA.8.2 Privileged access rights
NIST CSF
PR.AA Protect: identity, authentication, access
GDPR
Art. 25 Data protection by design and by default
Operators see IDs onlyPeople running the service work from IDs and counts. Customer content is outside the operator role, and the system-admin role sees no business's content.
SOC 2
CC6.3 Role-based access and least privilegeC1.1 Protecting confidential information
ISO 27001
A.8.2 Privileged access rightsA.8.11 Data masking
GDPR
Art. 32(4) Staff act only on instructions
Sessions ended in one stepSession limits on every sign-in. Removing a user ends their sessions at once. An admin ends all of one user's sessions in a single action, for a lost phone or a suspected stolen login.
SOC 2
CC6.2 Adding and removing users
ISO 27001
A.5.18 Access rightsA.8.5 Secure authentication
NIST CSF
PR.AA Protect: identity, authentication, access
Process privilege separationEach running process holds only its own keys and its own database role. The learning worker cannot write facts or check settings. The backup job cannot read the backup private key.
SOC 2
CC6.3 Role-based access and least privilege
ISO 27001
A.5.3 Segregation of dutiesA.8.2 Privileged access rights
NIST CSF
PR.AA Protect: identity, authentication, access
OWASP LLM
LLM06 Excessive agency
Access reviewsThe list of users, roles, named powers and system admins is reviewed at each release and before go-live. Changes go to logs nobody can edit.
SOC 2
CC6.2 Adding and removing users
ISO 27001
A.5.18 Access rights
NIST CSF
GV.RR Govern: roles and responsibilities
Data protection
Encryption in transit and at restConnections only over TLS. Databases and backups encrypted at rest. Backups are encrypted on the server before they leave, with the private key held outside every provider.
SOC 2
CC6.1 Logical access securityCC6.7 Protecting data in transmission
ISO 27001
A.8.24 Use of cryptography
NIST CSF
PR.DS Protect: data security
GDPR
Art. 32(1)(a) Encryption and pseudonymisation
CCPA
§1798.100(e) Reasonable security§1798.150 Liability for breaches from unreasonable security
Secrets managementSecrets live in encrypted environment settings, one set per process, never in code or its history. A secret scan runs on every change. Any exposed secret is rotated at once.
SOC 2
CC6.1 Logical access security
ISO 27001
A.5.17 Authentication informationA.8.24 Use of cryptography
NIST CSF
PR.DS Protect: data security
OWASP LLM
LLM07 System prompt leakage
Least data inA keep-list filter runs before storage. Anything not about a kept customer is dropped unread and counted. Card numbers and unrelated records never enter.
GDPR
Art. 5(1)(c) Data minimisationArt. 25 Data protection by design and by default
CCPA
§1798.100(c) Collection reasonably necessary and proportionate
SOC 2
P3.1 Collection limited to the purpose
ISO 27001
A.5.34 Privacy and protection of personal data
Read-only connections, narrowest scopesEvery connection to your tools is read-only, using the narrowest permission each provider offers. A stolen token could read, never change or send.
SOC 2
CC6.1 Logical access securityCC6.3 Role-based access and least privilege
ISO 27001
A.5.15 Access control
NIST CSF
PR.AA Protect: identity, authentication, access
OWASP LLM
LLM06 Excessive agency
No content in logs, alerts or monitoringSystem logs, alerts, error reports and notifications carry IDs and counts only, enforced by a field allow-list and tested with planted values.
ISO 27001
A.8.11 Data maskingA.8.12 Data leakage preventionA.8.15 Logging
SOC 2
C1.1 Protecting confidential information
GDPR
Art. 25 Data protection by design and by default
OWASP LLM
LLM02 Sensitive information disclosure
Residency in your chosen regionYour data lives in the region you choose, including several locations with their own residency terms. Transfers outside it rely on recognised safeguards: EU Standard Contractual Clauses, the EU-US Data Privacy Framework where the provider is certified, and the UK transfer addendum.
GDPR
Art. 44-46 International transfers with safeguards
ISO 27001
A.5.23 Security of cloud services
NIST CSF
GV.SC Govern: supply chain risk
Retention you setImport depth and retention windows are your settings. Facts weigh less as they age, and an old fact is never presented as current.
GDPR
Art. 5(1)(e) Storage limitation
SOC 2
P4.2 Retention of personal information
ISO 27001
A.8.10 Information deletion
Sensitive matters handled with careSensitive facts are labelled and restricted. They reach a customer only in that customer's own words. A mixed session's raw text is stored restricted and shown on no screen.
GDPR
Art. 9 Special categories of personal data
CCPA
§1798.121 Right to limit sensitive personal information
SOC 2
C1.1 Protecting confidential information
AI safety
Untrusted content boundarySource text enters a model only inside a labelled untrusted region. Extraction returns a fixed schema. The reasoning step has no tools and no network. Drafting never sees raw source text.
OWASP LLM
LLM01 Prompt injectionLLM05 Improper output handling
NIST AI RMF
MEASURE 2.7 Security and resilience evaluated
EU AI Act
Art. 15 Accuracy, robustness, cybersecurity (high-risk duty, followed by choice)
ISO 27001
A.8.26 Application security requirements
ISO 42001
A.6 AI system life cycle
Seven checks on every draftNothing invented; nothing from another customer; nothing stale; no false familiarity; within contact limits; no harm to a sensitive matter; your own rules. In that order, with no shortcut.
OWASP LLM
LLM09 MisinformationLLM02 Sensitive information disclosure
SOC 2
PI1.3 Accurate, complete processing
NIST AI RMF
MEASURE 2.5 Validity and reliability shown
EU AI Act
Art. 15 Accuracy, robustness, cybersecurity (high-risk duty, followed by choice)
ISO 42001
A.6 AI system life cycle
A person sends by defaultEvery action type starts at Propose. Autonomy is granted by the business one action type at a time, after a track record, with a recall window. Nothing widens its own permissions.
GDPR
Art. 22 No solely automated decisions with legal effect
EU AI Act
Art. 14 Human oversight (high-risk duty, followed by choice)
NIST AI RMF
MAP 3.5 Human oversight defined
ISO 42001
A.9 Responsible use of AI systems
OWASP LLM
LLM06 Excessive agency
AI-assistance disclosure on by defaultMessages drafted with Keepara carry a line saying so, by default, and the business may change it where its own law allows.
EU AI Act
Art. 50 Transparency for AI-generated content
NIST AI RMF
MEASURE 2.8 Transparency and accountability examined
ISO 42001
A.8 Information for interested parties
Approved model routes onlyA model route can be called only once its provider, region, retention and training terms are recorded. Routes that keep nothing are preferred. An answer from an unlisted provider raises an alert.
GDPR
Art. 28(2) Prior authorisation of sub-processorsArt. 28(4) Same duties flow down to sub-processors
SOC 2
CC9.2 Vendor and partner risk
ISO 27001
A.5.19 Security in supplier relationships
NIST AI RMF
GOVERN 6.1 Third-party AI risk policies
OWASP LLM
LLM03 Supply chain
ISO 42001
A.10 Third-party and customer relationships
Spend bounded per decisionEach wake has a fixed budget and ends with what it has. Jobs are safe to run twice and make one draft per reason. Spending is watched hourly, with an alert at five times the expected rate.
OWASP LLM
LLM10 Unbounded consumption
SOC 2
A1.1 Capacity management
ISO 27001
A.8.6 Capacity management
Identity matching with a person in the loopUncertain matches wait for a person. Nothing is drafted for an unconfirmed customer. Every merge can be split.
GDPR
Art. 5(1)(d) Accuracy
SOC 2
PI1.3 Accurate, complete processing
NIST AI RMF
MEASURE 2.5 Validity and reliability shown
OWASP LLM
LLM09 Misinformation
Evaluations and red-teaming on every changeLeak, fabrication, tone and prompt-injection sets run on every prompt or model change. Leak and injection sets must pass at 100%.
NIST AI RMF
MEASURE 2.7 Security and resilience evaluatedMANAGE 4.1 Monitoring after deployment
EU AI Act
Art. 15 Accuracy, robustness, cybersecurity (high-risk duty, followed by choice)
SOC 2
CC8.1 Change management
ISO 27001
A.8.29 Security testing before release
OWASP LLM
LLM01 Prompt injectionLLM02 Sensitive information disclosure
ISO 42001
A.6 AI system life cycle
Decision trail for every draftWhat woke the agent, the facts it used, the rules that fired, each check's result, the model and prompt version, who approved, and what was sent.
EU AI Act
Art. 12 Record-keeping (high-risk duty, followed by choice)Art. 13 Information for deployers (high-risk duty, followed by choice)
GDPR
Art. 5(2) Accountability
NIST AI RMF
MEASURE 2.8 Transparency and accountability examined
SOC 2
PI1.3 Accurate, complete processing
ISO 42001
A.8 Information for interested parties
Logging and monitoring
Every content read loggedWritten in the same database transaction as the read. A read whose log line cannot be written fails. Logs cannot be edited.
SOC 2
CC7.2 Monitoring for anomalies
ISO 27001
A.8.15 Logging
NIST CSF
DE.CM Detect: continuous monitoring
GDPR
Art. 30(2) Processor records of processingArt. 32 Security of processing
Alerts on the signals that matterBursts of refused sign-ins, an unlisted model provider answering, runaway spend, and any leak or invented fact reported by a user.
SOC 2
CC7.2 Monitoring for anomaliesCC7.3 Evaluating security events
ISO 27001
A.8.16 Monitoring activities
NIST CSF
DE.AE Detect: adverse event analysis
NIST AI RMF
MANAGE 4.1 Monitoring after deployment
Verified inbound events, listed hosts onlyEvery pushed event is verified by signature or channel token before anything else. Keepara fetches only from each source's own listed hosts, so an altered link cannot reach an internal service.
SOC 2
CC6.6 Protection at the system boundary
ISO 27001
A.8.20 Network security
NIST CSF
PR.IR Protect: infrastructure resilience
Content shown as text, never as codeAll customer content is rendered as plain text under a strict content security policy, so a note cannot run script in the app.
OWASP LLM
LLM05 Improper output handling
ISO 27001
A.8.26 Application security requirementsA.8.28 Secure coding
SOC 2
CC6.8 Preventing malicious code
Secure development
Pinned and scanned dependenciesExact pinned versions, scanned on every change and nightly. A high or critical flaw fails the build.
SOC 2
CC7.1 Detecting vulnerabilities and configuration changeCC8.1 Change management
ISO 27001
A.8.8 Technical vulnerability managementA.8.28 Secure coding
NIST CSF
PR.PS Protect: platform security
OWASP LLM
LLM03 Supply chain
Separate environments, fabricated test dataDevelopment, staging and production are separate. Production data stays in production. Tests run on fabricated businesses and customers.
ISO 27001
A.8.31 Separate development, test and productionA.8.33 Test information
SOC 2
CC8.1 Change management
GDPR
Art. 25 Data protection by design and by default
Reviewed, tested changesEvery change passes the test suite and safety tests in continuous integration before release. Prompt and model changes also pass the evaluation and red-team sets.
SOC 2
CC8.1 Change management
ISO 27001
A.8.25 Secure development life cycleA.8.32 Change management
NIST CSF
PR.PS Protect: platform security
ISO 42001
A.6 AI system life cycle
Resilience
Encrypted backups and a practised restoreNightly encrypted backups, checked by size and checksum. A weekly test restore compares row counts. A practised restore targets a working system within 4 hours, losing at most one day of data.
SOC 2
A1.2 Backup and recoveryA1.3 Recovery testingCC7.5 Recovering from incidents
ISO 27001
A.8.13 Information backupA.5.30 ICT readiness for business continuity
NIST CSF
RC.RP Recover: recovery plan execution
GDPR
Art. 32(1)(b) Ongoing confidentiality, integrity, availability, resilienceArt. 32(1)(c) Timely restore after an incident
One-step haltAny user stops waking, drafting and model calls in one action, within 10 seconds, surviving restarts. Held drafts are checked again after.
NIST AI RMF
MANAGE 2.4 Means to stop or disengage the system
EU AI Act
Art. 14 Human oversight (high-risk duty, followed by choice)
SOC 2
CC7.4 Responding to incidents
ISO 42001
A.9 Responsible use of AI systems
OWASP LLM
LLM06 Excessive agency
Continuity for your dataYour data can be exported in an open format at any time on request, and at exit. If Keepara stops operating, your data is returned and then deleted.
SOC 2
CC9.1 Business disruption risk
ISO 27001
A.5.30 ICT readiness for business continuity
GDPR
Art. 20 Right to data portabilityArt. 28(3)(g) Delete or return all data at the end
Privacy rights
Exit means goneAll data returned or deleted at exit, including every backup copy, with written confirmation. A deletion register outside the database is re-applied after any restore.
GDPR
Art. 17 Right to erasureArt. 28(3)(g) Delete or return all data at the end
CCPA
§1798.105 Right to delete
SOC 2
C1.2 Disposing of confidential informationP4.3 Secure disposal of personal information
ISO 27001
A.8.10 Information deletion
Help with every individual requestTools to find, export, correct and erase one customer's data, so your business can answer access, correction, deletion and portability requests.
GDPR
Art. 15 Right of accessArt. 16 Right to rectificationArt. 17 Right to erasureArt. 20 Right to data portabilityArt. 28(3)(e) Processor helps with data subject requests
CCPA
§1798.105 Right to delete§1798.106 Right to correct§1798.110 Right to know
SOC 2
P5.1 Access to personal information
Opt-outs are absoluteA customer who opts out is never proposed for contact again. No setting can override it. Contact caps and spacing limit how often anyone is approached.
GDPR
Art. 21(3) Objection to direct marketing is final
CCPA
§1798.120 Right to opt out of sale and sharing
SOC 2
P3.1 Collection limited to the purpose
Support for impact assessmentsArchitecture, data flows, this control map and the threat model are provided to support your data protection impact assessment and risk assessments.
GDPR
Art. 35 Data protection impact assessmentArt. 28(3)(h) Information and audits
EU AI Act
Art. 13 Information for deployers (high-risk duty, followed by choice)
ISO 42001
A.5 Assessing impacts of AI systems
Suppliers and incidents
Published subprocessor registerEvery outside service that touches customer data is named with its role, what it sees and where. Customers are told before a new one is added, and the same duties flow down by contract.
GDPR
Art. 28(2) Prior authorisation of sub-processorsArt. 28(4) Same duties flow down to sub-processors
CCPA
§1798.140(ag) Service provider duties, including no combining across businesses
SOC 2
CC9.2 Vendor and partner risk
ISO 27001
A.5.19 Security in supplier relationshipsA.5.20 Security in supplier agreementsA.5.21 ICT supply chain security
NIST CSF
GV.SC Govern: supply chain risk
ISO 42001
A.10 Third-party and customer relationships
Incident responseRunbooks for incidents, leaks and account compromise. Contain first, then assess, then tell you without undue delay, with the facts you need for your own legal deadlines.
GDPR
Art. 33(2) Processor reports breaches without undue delay
CCPA
§1798.82 California breach notice: service providers tell the data owner
SOC 2
CC7.3 Evaluating security eventsCC7.4 Responding to incidents
ISO 27001
A.5.24 Incident management planningA.5.26 Response to incidentsA.5.27 Learning from incidents
NIST CSF
RS.MA Respond: incident managementRS.CO Respond: incident communication
NIST AI RMF
MANAGE 4.3 Incidents and errors communicated
Vulnerability disclosureA security contact at [email protected], a security.txt file, and good-faith safe harbour for researchers who report a flaw responsibly.
ISO 27001
A.6.8 Security event reportingA.8.8 Technical vulnerability management
NIST CSF
ID.RA-08 Identify: receiving vulnerability disclosures
SOC 2
CC2.3 Communication with customers and other outside parties
Regulations and standards, one by one

What each one requires, and how Keepara is designed to meet it.

Privacy law, AI law, security and AI standards, and the sector rules for the industries Keepara serves.

Privacy law

GDPR and UK GDPR

Regulation (EU) 2016/679; UK GDPR and the Data Protection Act 2018

What it is

The European and UK law on personal data. It applies whenever a business processes data about people in the EU or UK.

Keepara's role

Your business is the controller. Effectv, which makes Keepara, is your processor for customer data, and a controller only for its own account and billing records.

What it requiresHow Keepara is designed to meet it
Art. 28: processor contractA Data Processing Agreement with every customer: instructions, confidentiality, security, sub-processors, help with requests and assessments, deletion at the end, and audit information.
Art. 29: only on instructionsCustomer data is processed only to run your workflows. No use for Keepara's own purposes, no sale, no learning across businesses.
Art. 25: by design and by defaultLeast privilege by default, least data in, private customers excluded, every setting starting at its most protective.
Art. 32: security of processingEncryption in transit and at rest, the database-enforced wall, total logging, a practised restore, and testing on every change.
Art. 28(2) and (4): sub-processorsA published register, advance notice of changes, and the same duties passed down by contract.
Art. 15 to 20: individual rightsFind, export, correct and erase one customer's data. Corrections supersede old facts; erasure removes the customer entirely.
Art. 21(3): objection to marketingOpt-outs are absolute and no setting can override them.
Art. 22: automated decisionsKeepara makes no decision with legal or similarly significant effect. A person sends every message by default.
Art. 9: special categoriesSensitive matters are labelled, restricted, and reach a customer only in that customer's own words.
Art. 33(2): breach noticeYou are told without undue delay, with the facts you need to meet your own 72-hour deadline.
Art. 35: impact assessmentsArchitecture, data flows, threat model and this control map provided for your assessment.
Art. 44 to 46: transfersData stays in your chosen region. Any transfer relies on EU Standard Contractual Clauses (Decision 2021/914), the EU-US Data Privacy Framework where the provider is certified, or the UK International Data Transfer Addendum.
ePrivacy Directive Art. 13 and UK PECR reg. 22Your consent and soft opt-in records are respected as contact rules; Keepara never contacts anyone your records exclude.
Privacy law

CCPA, CPRA and US state laws

California Consumer Privacy Act as amended by the California Privacy Rights Act; Virginia, Colorado, Connecticut, Texas, Oregon and other state privacy laws

What it is

California's privacy law gives consumers rights over their data and sets duties for businesses and their service providers. Most other US states with comprehensive privacy laws follow a similar controller and processor model.

Keepara's role

Keepara is your service provider under the CCPA and your processor under the other state laws.

What it requiresHow Keepara is designed to meet it
§1798.140(ag): service provider contractTerms that name the business purpose; forbid selling or sharing; forbid using the data outside your relationship; forbid combining it with other businesses' data; and give you audit rights.
No combining across businessesMade structural by the database-enforced wall and tenant-local learning, not only promised in a contract.
§1798.105, .106, .110: delete, correct, knowTools to find, export, correct and delete one consumer's data, so you can answer each request in time.
§1798.120: no sale or sharingKeepara never sells or shares personal information, and contracts say so.
§1798.121: sensitive personal informationUsed only to perform the service; labelled, restricted and kept out of every screen that does not need it.
§1798.100(e) and §1798.150: reasonable securityThe full control map on this page is the reasonable-security programme.
§1798.82: breach noticeAs a service provider, Keepara tells you immediately after discovering a breach of your data.
Automated decisionmaking regulationsKeepara makes no significant decision about a consumer, and the excluded uses below keep it that way, so pre-use notice and opt-out duties for such decisions are not triggered.
Other state laws: processor dutiesConfidentiality, deletion or return at the end, sub-processor flow-down, and help with your assessments, all covered by the same agreement.
Privacy law

India DPDP Act

Digital Personal Data Protection Act, 2023, and its Rules

What it is

India's personal data law. It applies to digital personal data processed in India, and to processing outside India for offering goods or services there.

Keepara's role

Your business is the Data Fiduciary. Keepara is your Data Processor.

What it requiresHow Keepara is designed to meet it
§8(2): processing under a contractKeepara processes only under a written contract with you.
§8(5): reasonable security safeguardsThe control map on this page.
§8(6): breach intimationYou are told promptly, with what you need to inform the Board and affected people.
§8(7): erasure when the purpose endsExit means gone, including backups.
§11 to 13: rights of individualsTools to find, correct and erase one person's data.
§16: cross-border transferData lives in your chosen region, India included, and is never sent to a restricted country.
AI law

EU AI Act

Regulation (EU) 2024/1689 on artificial intelligence

What it is

The European Union's law on AI systems. It bans some practices, sets heavy duties for high-risk uses, and requires transparency for AI-generated content.

Keepara's role

Effectv is the provider of the Keepara AI system. Your business is the deployer. The makers of the underlying general-purpose models are model providers with their own duties.

What it requiresHow Keepara is designed to meet it
Art. 5: prohibited practicesNone of Keepara's functions manipulate people, exploit vulnerabilities, score people socially, recognise emotions at work, or categorise people by biometrics.
Art. 6 and Annex III: high-riskKeepara is designed for uses outside every high-risk area. The classification of a deployment depends on how the business uses it. See the excluded uses below.
Art. 50: AI-generated contentAn AI-assistance line on by default for every drafted message.
Art. 4: AI literacyTraining for the Keepara team, and guidance for every person who approves drafts: what the system does, its limits, and how to stop it.
Art. 25: value chainCustomer terms will prohibit repurposing Keepara for a high-risk use, which would make the deployer a provider.
Art. 53: model providersModel routes are chosen from providers that publish their documentation; each decision records the exact model used.
Art. 12 to 15: high-risk duties, followed by choiceLogging of every decision, clear information for deployers, human oversight with a one-step stop, and testing for accuracy, robustness and security on every change.
AI law

US AI and bot laws

Colorado AI Act (SB 24-205); Utah Artificial Intelligence Policy Act; California bot disclosure law (Bus. & Prof. Code §17940 onward)

What it is

State laws on AI that makes consequential decisions about consumers, and on disclosing AI use in communications.

Keepara's role

Keepara is a developer under Colorado's law; your business is the deployer.

What it requiresHow Keepara is designed to meet it
Colorado: high-risk consequential decisionsKeepara is not used for decisions on employment, credit, housing, insurance, education, health care, legal or essential government services. The excluded uses keep it out of scope.
Colorado and Utah: telling consumers about AIThe AI-assistance line is on by default, and your team can confirm AI use whenever a customer asks.
California: bot disclosureA person reviews and sends every message, and the disclosure line is on by default.
Security standard

SOC 2

AICPA SOC 2 examination against the 2017 Trust Services Criteria (revised points of focus, 2022)

What it is

An independent auditor's report on a service organisation's controls. It is the report most US security teams ask for.

Keepara's role

Keepara's controls are designed against all five Trust Services categories. The audit order is set out under How claims are proven.

What it requiresHow Keepara is designed to meet it
Security: CC1 to CC5, governance and riskThree-layer architecture, named threat model, traced requirements, role training.
Security: CC6, accessDatabase wall, least privilege by role, two-step verification, session control, operators see IDs only, encryption in transit.
Security: CC7, operationsDependency scanning, alerting on the signals that matter, incident and compromise runbooks, practised recovery.
Security: CC8 and CC9, change and vendorsEvery change tested in continuous integration; every model route and supplier recorded before use.
Availability: A1Nightly encrypted backups, weekly restore checks, practised restore within 4 hours.
Confidentiality: C1Content kept out of logs and alerts; deletion at exit including backups.
Processing integrity: PI1Seven checks on every draft and a decision trail for each one.
Privacy: P1 to P8Least data in, retention you set, secure disposal, help with every individual request.
Security standard

ISO 27001 family

ISO/IEC 27001:2022, with 27701 (privacy), 27017 (cloud) and 27018 (personal data in cloud)

What it is

The international standard for an information security management system, with extensions for privacy and cloud services.

Keepara's role

Keepara's controls map to Annex A themes. Physical security is inherited from the hosting provider, which holds its own ISO 27001 certification and SOC 2 reports.

What it requiresHow Keepara is designed to meet it
Organisational controls (A.5)Policies, segregation of duties, access control, supplier security, cloud services, incident management, continuity, legal register, privacy.
People controls (A.6)Role training and a route for reporting security events.
Physical controls (A.7)Inherited from the cloud provider's certified data centres.
Technological controls (A.8)Privileged access, secure authentication, vulnerability management, deletion, masking, leak prevention, backup, logging, monitoring, cryptography, secure development, environment separation, change management.
27701 and 27018: personal dataProcessor duties: instructions only, sub-processor disclosure, help with rights, deletion, no use for the provider's own purposes.
Security standard

NIST CSF 2.0

NIST Cybersecurity Framework 2.0

What it is

The US framework that organises security into six functions. Many US security teams use it to structure vendor reviews.

Keepara's role

Keepara's controls cover all six functions.

What it requiresHow Keepara is designed to meet it
GovernThree-layer architecture, roles, supplier register, legal register.
IdentifyNineteen-threat model, traced requirements, vulnerability disclosure intake.
ProtectWall, least privilege, two-step verification, encryption, secrets, least data in, platform hardening.
DetectRead logging, anomaly alerts, spend and provider alerts.
RespondIncident, leak and account-compromise runbooks; notice to you without undue delay.
RecoverPractised restore within 4 hours, deletion register re-applied before anything else.
AI standard

NIST AI RMF and ISO 42001

NIST AI Risk Management Framework 1.0 (with the Generative AI Profile, NIST AI 600-1); ISO/IEC 42001:2023

What it is

The leading frameworks for managing AI risk: the US government framework and the international standard for an AI management system.

Keepara's role

Keepara's AI controls map to all four NIST functions and to the ISO 42001 Annex A areas.

What it requiresHow Keepara is designed to meet it
Govern (NIST) and A.2 to A.4 (ISO)Excluded uses, legal register, AI literacy, third-party model route policy.
Map (NIST) and A.5 (ISO)Intended purpose documented; human oversight defined; threat model with likelihood and impact.
Measure (NIST) and A.6 to A.7 (ISO)Evaluation sets for fabrication, leaks, tone and injection on every change; privacy risk examined; data provenance on every fact.
Manage (NIST) and A.8 to A.10 (ISO)One-step halt, monitoring after release, incident communication, information for deployers, supplier terms recorded.
AI standard

OWASP LLM Top 10

OWASP Top 10 for Large Language Model Applications, 2025

What it is

The security community's list of the ten most important risks in applications built on language models.

Keepara's role

Each of the ten has a named defence in Keepara's design.

What it requiresHow Keepara is designed to meet it
LLM01 Prompt injectionUntrusted region, schema-only extraction, no tools or network in reasoning, red-team set at 100%.
LLM02 Sensitive information disclosureOne customer per call, check 2 hard block, no content in logs.
LLM03 Supply chainPinned and scanned dependencies; model routes recorded before use; unlisted provider alert.
LLM04 Data and model poisoningLearning only from your team's explicit feedback, attributed and reversible, never shared.
LLM05 Improper output handlingSchema-validated outputs; content rendered as plain text under a strict content security policy.
LLM06 Excessive agencyRead-only connectors; a person sends; nothing widens its own permissions; one-step halt.
LLM07 System prompt leakagePrompts hold no keys and no other customer's data; secrets live only in process settings.
LLM08 Vector and embedding weaknessesSemantic indexes sit inside the same database wall as every other table.
LLM09 MisinformationEvery sentence must cite a current, sourced fact; stale facts blocked.
LLM10 Unbounded consumptionFixed budget per decision; hourly spend alert; one draft per reason.
Sector rules

Customer communications

TCPA (47 U.S.C. §227); CAN-SPAM Act (15 U.S.C. §7701 onward); EU ePrivacy Directive; UK PECR

What it is

Laws on contacting customers by phone, text and email: consent, identification and opting out.

Keepara's role

Your business sends every message through its own tools and remains the sender. Keepara drafts, and enforces your contact rules before a draft is shown.

What it requiresHow Keepara is designed to meet it
TCPA: calls and textsKeepara places no calls and sends no texts itself. Your consent records and opt-outs are contact rules every draft must pass.
CAN-SPAM: emailSender identity and unsubscribe live in your email tool; Keepara never proposes contact to anyone who has opted out.
ePrivacy and PECR: electronic marketingSoft opt-in and consent records respected; contact caps and spacing limit frequency.
Sector rules

Financial services

GLBA Safeguards Rule (16 CFR Part 314); NYDFS Cybersecurity Regulation (23 NYCRR Part 500); EU DORA (Regulation (EU) 2022/2554); PCI DSS v4.0.1; ECOA and FCRA

What it is

Rules for banks, insurers and other financial firms, including how they oversee technology suppliers.

Keepara's role

Keepara is a third-party service provider to the financial institution.

What it requiresHow Keepara is designed to meet it
GLBA §314.4(f): oversee service providersSecurity documentation, this control map and contract terms that require the safeguards described here.
NYDFS §500.11: third-party providersTwo-step verification, encryption in transit and at rest, and prompt notice of cybersecurity events affecting you.
DORA Art. 28 to 30: ICT third-party riskContract terms covering data location, access, audit, incident help and an exit strategy with full return and deletion.
PCI DSS: cardholder dataCard numbers are never collected; least data in keeps Keepara outside your cardholder data environment.
ECOA and FCRA: credit and eligibilityKeepara makes no credit, eligibility or adverse-action decision and is not a consumer reporting agency.
Sector rules

Telecommunications

Customer Proprietary Network Information rules (47 U.S.C. §222; 47 CFR §64.2001 onward)

What it is

US rules protecting what a telecom carrier knows about a customer's service and usage.

Keepara's role

Keepara processes such information only as the carrier's agent, under contract.

What it requiresHow Keepara is designed to meet it
Use and disclosure limitsNetwork information is used only for the carrier's own customer care, never disclosed, never combined with another business's data.
Customer approval statusThe carrier's records of each customer's approval are respected as contact rules.
Safeguards and breach noticeThe control map on this page; notice to the carrier without undue delay.
Sector rules

Health information

HIPAA Privacy and Security Rules (45 CFR Parts 160 and 164)

What it is

US rules for protected health information held by health plans, providers and their business associates.

Keepara's role

Keepara is not offered for protected health information today.

What it requiresHow Keepara is designed to meet it
§164.504(e): business associate termsRequired before any protected health information could be connected. Not offered yet.
§164.312: technical safeguardsUnique sign-in, session control, audit logging of every read, integrity checks, and encryption in transit, all part of the standard design.
Minimum necessaryLeast data in: only facts about kept customers, sensitive matters restricted.
Privacy law

Other privacy laws

Canada PIPEDA and Quebec Law 25; Brazil LGPD; Australia Privacy Act 1988; Singapore PDPA

What it is

Other national privacy laws that follow the same pattern: accountability, security, limits on use, and rights for individuals.

Keepara's role

Keepara acts as processor or service provider under each.

What it requiresHow Keepara is designed to meet it
Contract and accountabilityThe same processing agreement, adapted to local terms.
Security and breach noticeThe control map on this page and notice without undue delay.
TransfersYour chosen region, with recognised transfer safeguards where data must move.
EU AI Act

Where Keepara sits, and where it never will.

The EU AI Act sorts AI systems by risk. Keepara drafts customer communication that a person reviews and sends. It is designed for the transparency tier, and the uses that would make it high-risk are excluded by design. How a deployment is classified depends on how the business uses it.

Effectv is the provider. Your business is the deployer. Duties for high-risk systems, such as decision logging, human oversight and robustness testing, are followed by choice even though Keepara is designed for uses outside the high-risk class.

ProhibitedManipulation, social scoring, emotion recognition at work, biometric categorisation.
Not Keepara
High-riskCredit, insurance pricing, employment, essential services, education.
Excluded by design
Transparency dutiesAI-generated content must be disclosed and marked.
Keepara sits here
Minimal riskEverything else.

Excluded uses

Keepara is not offered, and may not be used, for any of the following.

  1. Credit decisionsDeciding creditworthiness, credit scores or credit terms.EU AI Act Annex III 5(b); ECOA; FCRA
  2. Insurance pricingAssessing risk or setting prices for life or health insurance.EU AI Act Annex III 5(c)
  3. Employment decisionsHiring, firing, promotion, task allocation or monitoring of workers.EU AI Act Annex III 4; Colorado AI Act
  4. Access to essential servicesEligibility for benefits, housing, education, health care or essential services.EU AI Act Annex III 3 and 5(a); Colorado AI Act
  5. Decisions without a personAny decision with legal or similarly significant effect made without human review.GDPR Art. 22; California automated decisionmaking rules
  6. Biometrics and emotionIdentifying people by biometrics, recognising emotions, or categorising people by sensitive traits.EU AI Act Art. 5 and Annex III 1
  7. ChildrenContacting or profiling children.COPPA; GDPR Art. 8
  8. Selling dataSelling or sharing personal information, or using it to build profiles for anyone else.CCPA §1798.120
Subprocessors

Every outside service that touches customer data.

Named by role, with what each one sees and where. Customers are told before a new subprocessor is added, and the same duties flow down by contract.

ServiceRoleWhat it seesWhere
Amazon Web ServicesHosting, database and encrypted backup storageAll stored customer data, encrypted at restThe AWS region you choose
OpenRouterRoutes each AI model call to an approved model providerOne customer's relevant facts for one taskUnited States and EU routes; each route's region and retention recorded before use
Model providers reached through OpenRouterRun the language model for a single taskOne customer's relevant facts for one task, never used for trainingListed per route in the register before any route is used
NangoHolds the read-only connections to your calendar, meeting and notes toolsConnection tokens.Recorded per deployment
GoogleStaff sign-in with two-step verificationSign-in identity only, no customer contentGoogle's global infrastructure

Register last reviewed 14 November 2026.

How claims are proven

Proof inside, then proof from outside, in a fixed order.

Every requirement is traced to a named check. Outside assurance is added step by step, each step started by a clear trigger.

247requirements each traced to a named test, evaluation, runtime guard or review
19named threats each with the defence that stops it and the signal that detects it
100%pass mark for the leak and prompt-injection sets, on every prompt or model change
50simulated businesses in the isolation suite, every pair checked against every other
  1. Internal security reviewBefore any customer's real data enters production.

    Sign-in, roles, the wall, secrets, public addresses, event receivers, content security policy, process privileges, red-team results. Every finding fixed or accepted in writing.

  2. Independent penetration testBefore the first enterprise rollout.

    An outside firm attacks the running service: the web app, the API, the wall between businesses, and prompt injection through real connectors.

  3. SOC 2 Type IWhen an enterprise customer requires it.

    An independent auditor confirms the controls on this page are designed and in place.

  4. SOC 2 Type IIFollows Type I and continues every year.

    The auditor tests that the controls operated over a period of months.

  5. ISO 27001 and ISO 42001When European or enterprise customers require them.

    Certification of the security management system and the AI management system.

For security reviewers. We walk your team through the architecture, the threat model, this control map and the code itself, under a confidentiality agreement. Write to [email protected].
Incidents and disclosure

If something goes wrong, you hear it from us.

Contain first, then assess, then tell you without undue delay, with the facts you need for your own legal deadlines. Under GDPR, that includes your 72-hour deadline to report a breach to the regulator.

Report a security issue

Email [email protected]. Our security contact is also published in security.txt.

  • We acknowledge every report and keep you informed until it is resolved.
  • Good-faith research that avoids customer data, service disruption and social engineering is welcome, and we will not pursue legal action against it.
  • Please give us reasonable time to fix an issue before you disclose it.
Questions security teams ask

Straight answers.

Do you train AI models on our data?

No. Learning stays inside your business, comes only from your own team's feedback, and can be reversed. Model routes are allowed only when their terms forbid training on inputs.

Can people at Keepara read our customers' data?

The operator role sees IDs and counts only. Customer content is outside every Keepara staff role, and any read of content by anyone is logged where it cannot be edited.

Where is our data stored?

In the region you choose, on Amazon Web Services, encrypted at rest. Each model route's processing region is recorded before it can be used.

What happens if a model provider is breached?

A provider only ever receives one customer's relevant facts for one task, and only through routes with recorded retention terms. Routes that keep nothing are preferred.

One of our customers wants their data deleted. Can you do it?

Yes. A single customer's data can be found, exported or erased on its own, and the erasure survives any later restore from backup.

Do you have SOC 2?

Keepara's controls are designed against all five SOC 2 Trust Services categories today. The audit follows the order under How claims are proven, starting when an enterprise customer requires it.

Can our security team review the code and architecture?

Yes. We walk your team through the architecture, the threat model, this control map and the code itself, under a confidentiality agreement.

What if Keepara stops operating?

Your data is exported to you in an open format, then deleted from the live system and every backup, with written confirmation.

Can Keepara send messages on its own?

Not by default. Every action type starts at Propose. Your business decides whether any type may ever act alone, one type at a time, and can stop everything in one action.

Security review

Bring your hardest questions.

I will walk your security, privacy or legal team through the architecture, the control map and the code.