Secure AI agents: control and audit.
CortexHub is the governance, control, and security layer for AI agents. Policy runs at runtime before any action executes, credentials never reach the model, sensitive data is redacted, and every action leaves a tamper-evident record your compliance team can trust, mapped to SOC 2, ISO 42001, the EU AI Act, and NIST AI RMF.
How your data is protected.
Designed from the ground up for privacy and data minimization.
Policies enforced on the gateway
All policy decisions happen server-side on the CortexHub gateway, not in your environment. Every tool call is evaluated before it is forwarded to the external API. There is no bypass mode.
Credentials stored in a secure vault
OAuth tokens and API keys are stored in CortexHub's secure credential vault and injected at call time, scoped to the operation. Your agents never see credentials. They only receive tool results.
Audit records are hash-chained
Every tool call produces a tamper-evident enforcement record linked to its predecessor. Deleting or modifying any record breaks the chain, detectable offline from an export alone, without trusting CortexHub.
Encrypted at rest and in transit
All data in transit is encrypted (TLS 1.2+, with TLS 1.3 preferred). All data at rest is encrypted (AES-256). Credentials live in a hardened, encrypted secrets vault and never appear in application logs. Enterprise customers can discuss deployment and hosting options with our team.
Isolated per organization
Each organization is a hard boundary for policy, credentials, audit, and usage. Tokens and keys are namespaced per organization in the vault, so one team's data is never reachable from another's context.
Every action attributed to a person
Each tool call is bound to the end user the agent is acting for, not just the agent. Policy, approvals, and audit all resolve to a named person, so "who did this?" always has a clear answer.
What CortexHub records
CortexHub evaluates every tool call on the gateway. You control how much payload data is persisted; governance metadata is always recorded.
Always recorded
- ✓ Policy decision (permit, deny, require approval, modify, or step-up) and the policy that triggered it
- ✓ Tool name and toolkit context
- ✓ Agent identity and end-user identity (user_id) if provided
- ✓ Timestamps and call duration
- ✓ Approval ID and outcome if an approval was required
- ✓ Cryptographic hashes of request and response payloads for integrity verification
Tool call payloads - your choice
Request and response bodies are not stored by default unless you enable payload storage. Configure this at organization or agent level.
- • When payload storage is off, CortexHub keeps metadata only - no request or response bodies are written to the audit log.
- • When payload storage is on, bodies are never stored raw. They pass through a redaction pipeline before persistence.
Redaction you control
When payloads are stored, sensitive values are detected and replaced before anything is written. You choose what CortexHub should never retain.
- • Built-in PII and secret detection (emails, tokens, card numbers, and similar patterns)
- • Custom guardrail regex patterns at organization or agent scope for fields you define
- • Defined once and enforced on every tool call, not configured integration by integration
- • Matched values replaced with redaction markers - originals are not written to audit logs or exports
Example stored record - with guardrails on email and phone
{ "tool": "crm.update_contact", "args": { "name": "Alice", "email": "[REDACTED]", "phone": "[REDACTED]" } }
Never stored
- ✗ OAuth tokens, API keys, or any credentials - injected at call time, never logged
- ✗ LLM prompts or model responses - the enforcement audit records governed tool calls, not model-inference content (AI content is handled per the Privacy Policy)
- ✗ Raw unredacted request or response payloads at rest
OAuth tokens and API keys live in the CortexHub vault, are injected at call time, and are never written to audit logs, dashboards, or exports.
Infrastructure security
Encryption in transit: TLS 1.2+ (TLS 1.3 preferred)
Encryption at rest: AES-256 for all stored data
Secrets management: Hardened encrypted vault, isolated per organization
Access controls: Role-based access with least-privilege principles
Organization isolation: Policy, credentials, and audit scoped to each organization
Internal audit logging: All access to customer data is logged
Telemetry export: Signed outbound webhooks stream activity to your SIEM or warehouse
Deployment options: Enterprise hosting and region preferences available on request
Availability: High-availability infrastructure with SLA for Enterprise
Your audit records cannot be quietly altered.
A dashboard score isn't evidence. CortexHub makes every AI decision record cryptographically verifiable, so your auditor can confirm nothing was changed after the fact, without taking your word for it.
Recorded on the gateway
Every tool call produces an enforcement record on the CortexHub gateway with tool name, policy decision, agent identity, and timestamps, written immediately when the decision is made.
Hash-chained to its predecessor
Each record includes a hash of the previous record for the same agent. Any insertion, deletion, or modification of a record breaks the chain at that exact point.
Verifiable offline
Export your audit log and run the chain verifier independently. Any broken link identifies the exact record that was tampered with. No CortexHub account required.
Tampered records surface immediately
If any record is altered after being written, the chain breaks. The exact record is identified, not just a count. Evidence for your auditor, not just a dashboard score.
Our compliance posture.
We hold ourselves to the same standards we help our customers meet.
Current practices
Our security controls are built to enterprise standards today, and several can be verified independently without trusting us. Formal certifications are in progress:
- ✓ Access controls and least-privilege principles across all systems
- ✓ Encryption at rest and in transit for all customer data
- ✓ Audit logging for all access to customer data
- ✓ Regular security reviews
- ✓ Incident response procedures
- ✓ Secure credential vault for OAuth tokens
Frameworks and certifications
- → SOC 2 Type II: In progress, targeted for 2026
- → ISO 27001: On the roadmap
- → ISO 42001 (AI management): Controls mapped for evidence today
- → EU AI Act: Audit trail and evidence mapped for obligations
- → NIST AI RMF: Controls mapped to the risk management functions
- → GDPR alignment: In practice today. DPAs available on request
Privacy & GDPR.
Deployment options
CortexHub Cloud runs on enterprise-grade infrastructure. Enterprise customers can discuss hosting and region preferences with our team based on compliance requirements.
GDPR posture
We are GDPR-aligned. We store full tool call requests and responses, with fields redacted according to the guardrails you configure. Credentials are never stored in audit logs. Data Processing Agreements (DPAs) are available on request.
Customer agent data
CortexHub stores the full request and response for every tool call, subject to your guardrail redaction rules. OAuth credentials are stored in a secure vault and never appear in audit records or exports. For AI features (co-workers, chat, and meetings), CortexHub processes prompts, retrieved context, and model responses to generate results and may retain them as part of your co-worker's memory and meeting records, as described in the Privacy Policy; your content is never used to train third-party models.
Data retention
Governed invocation spans (tool calls, policy decisions, redacted arguments) are retained for 365 days. Audit records and cryptographic receipts are kept for 7 years. Retention terms can be extended by contract. Data deletion requests are honoured within 30 days.
Found a vulnerability? Tell us first.
We value the security research community. If you find a vulnerability, please tell us before making it public.
In scope
- ✓ CortexHub platform (app.cortexhub.ai)
- ✓ Python SDK (cortexhub on PyPI)
- ✓ Marketing website (cortexhub.ai)
- ✓ CortexHub-owned infrastructure
Out of scope
- ✗ Social engineering or phishing
- ✗ Physical attacks
- ✗ Third-party services without direct impact
- ✗ Rate limiting or version disclosure without exploit
How to report
Email security@cortexhub.ai with a clear description, affected system, steps to reproduce, and your assessment of impact.
Initial response: Within 4 business days
Safe harbour: Good-faith research following this policy is authorised. We will not pursue legal action.