Trust & Safety

Context Trust & Safety Center

Security, compliance, and data-handling documentation for the Context platform. Deployment models, security controls, policies, subprocessors, and how to request audit artifacts.

Overview

Security posture at a glance

Context applies the same security controls across every deployment mode, from managed to air-gapped: identity from your IdP, per-action authorization, isolated execution, and documented data boundaries.

  • Encryption in transit and at rest

    Customer personal data is encrypted in transit using TLS and at rest using industry-standard encryption.

    Data handling →

  • Identity from your IdP

    Agents get identity from your identity provider. Runs in production with SSO-scoped permissions.

    Identity & authorization →

  • Per-action authorization and audit

    Every action is authorized against policy and recorded in an append-only trail: task, tool calls, approvals, result.

    Identity & authorization →

  • Credentials brokered at runtime

    Credentials are brokered to connectors at runtime — never written into runbooks or prompts.

    Identity & authorization →

  • Isolated execution, your perimeter

    Engine runs each agent in an isolated environment, and the platform deploys managed, in your VPC, on-premises, or air-gapped.

    Deployment models →

  • Documented data boundaries

    Run traces stay inside your deployment. What crosses the boundary is documented during implementation, per architecture.

    Deployment models →

Controls

Security controls

View all controls →

FAQ

Frequently asked questions

View the full site FAQ →

Deployment

Where does Context run?

Context supports a managed deployment, deployment in a customer VPC, on-premises deployment, and environments designed for disconnected operation. Model endpoints, connectors, the update process, and the support model are configured for the selected architecture.

What data leaves our boundary?

Run traces stay inside your deployment. What crosses the boundary depends on the architecture you select — in a VPC, on-premises, or air-gapped deployment the platform itself runs inside your perimeter. Data-flow boundaries are documented during implementation.

Is it the whole platform inside our perimeter, or just inference?

The entire platform. Context deploys the control plane, execution, and run history — not just model inference — in your VPC, on-premises, or air-gapped.

Security

Do agents hold credentials?

No. Credentials are brokered to the connector at runtime rather than written into the runbook or prompt. Depending on the target system and deployment, the connector can use delegated user identity or a scoped service identity. Credential storage, rotation, and revocation are reviewed as part of the deployment design.

How do agents access our systems?

Engine runs each agent in an isolated environment and gives it only the tools configured for the runbook. Connectors use the authentication method supported by the target system, such as OAuth or a scoped service credential. Read access, write access, and human approval can be configured separately.

What does the audit story look like?

A run records the task, model and tool calls, source references, actions, approvals, and result. Reviewers can inspect the sequence and compare the output with its rubric. Export and retention requirements are configured for the customer's deployment.

How do we get security documentation for a review?

Use the request-access form on this page, or email security@context.ai with what your review needs. Public documents are linked directly; documents that require an NDA are shared once it is in place.

Privacy

Is our data used to train models?

Context does not use one customer's traces, corrections, or institutional context to train models for other customers. Model-provider retention and training settings depend on the endpoints a customer chooses, so those controls are documented and verified during deployment.

Which model providers are involved, and what do they see?

Context is model-agnostic: Claude, GPT, Gemini, Kimi, or open weights. Available endpoints depend on your deployment architecture, and provider retention and training settings are documented and verified during implementation.

Deployment

Deployment models and data boundaries

View full documentation →

Four ways to run Context, each with an explicit data boundary. Pick the perimeter; the control plane, execution, and run history deploy inside it.

Managed

Context operates the deployment for you. The fastest path to production when a hosted control plane fits your policy.

Data boundary

Run traces stay inside your deployment. Model-provider retention and training settings are documented and verified during implementation.

Managed in depth →

Your VPC

The platform runs inside your own AWS, Azure, or GCP account, under your cloud controls and your keys to the perimeter.

Data boundary

Sensitive data is not moved into someone else's cloud — the deployment lives in an account you own and govern.

Your VPC in depth →

On-premises

Context runs in your data center, on infrastructure you operate, inside your perimeter.

Data boundary

The deployment is operated inside your perimeter. Data-flow boundaries are documented during implementation.

On-premises in depth →

Air-gapped

For environments designed for disconnected operation. The entire platform deploys inside the boundary — not just inference.

Data boundary

Built for disconnected operation: the entire platform, including run history, deploys inside the boundary.

Air-gapped in depth →

In every mode: agents get identity from your IdP, every action is authorized against policy and recorded in an append-only trail, and credentials are brokered to connectors at runtime — never written into runbooks or prompts.

Compliance

Certifications and attestations

View program details →

Being populated

Compliance documentation for the current audit cycle is available on request. For the status of specific attestations, contact security@context.ai.

Documents

Policies and documentation

Public documents are linked directly. Gated documents are shared on request, under NDA where required.

Audit reports & assessments

  • Penetration-testing report (latest)

    The most recent third-party penetration test of the platform.

  • CAIQ security questionnaire

    Pre-completed Consensus Assessments Initiative Questionnaire for self-serve review.

  • Security policies (combined)

    The policies of the written information security program, combined in one document.

  • Security & compliance program update

    Point-in-time summary of the compliance program and what changed.

Security architecture

  • Security architecture whitepaper

    The full architecture behind the public principles: the control-plane/sandbox split, tenant isolation, and the authorization model in depth.

  • Software Bill of Materials (SBOM) — latest release

    Per-release SBOM with signed build provenance. On the security roadmap; requestable when the supply-chain work ships, and matters most for self-hosted installs.

  • Hardened base-image inventory

    The minimal, pinned, regularly rebuilt base images shipped identically to managed and self-hosted. On the security roadmap; requestable once published.

Procurement & operations

  • W-9

    Taxpayer identification form for vendor onboarding.

  • Evidence of insurance

    Current certificate of insurance for procurement review.

Public policies

  • Data Processing Addendum

    How customer personal data is processed, protected, and transferred.

    View
  • Subprocessor list

    The vendor list referenced by the DPA, with purpose and location.

    View
  • Privacy Policy

    What we collect, why, and the rights you can exercise.

    View
  • Terms of Service

    The agreement that governs use of the Context platform.

    View
  • Acceptable Use Policy

    What the platform may and may not be used for.

    View
  • Security & deployment FAQ

    Deployment modes, credential brokering, training boundaries, and the audit story.

    View

Subprocessors

VendorPurposeRegionWebsite

Amazon Web Services, Inc.

Cloud infrastructure

Cloud infrastructure and hostingUnited Statesaws.amazon.com (opens in new tab)

Vercel Inc.

Web hosting

Website and edge hostingUnited Statesvercel.com (opens in new tab)

Anthropic, PBC

AI model inference

AI model inferenceUnited Statesanthropic.com (opens in new tab)

OpenAI, LLC

AI model inference

AI model inferenceUnited Statesopenai.com (opens in new tab)

Datadog, Inc.

Monitoring

Application logging and monitoringUnited Statesdatadoghq.com (opens in new tab)

Grafana Labs

Monitoring

Metrics dashboards and observabilityUnited Statesgrafana.com (opens in new tab)

Disclosure

Responsible disclosure

Report

Found a vulnerability in Context — the platform, a connector, or this site? Report it to security@context.ai. Include the affected surface, reproduction steps, and the impact you observed; remediation and disclosure timing are coordinated with you through that thread.

Updates

  • Trust center expanded

    General

    Published August 14, 2026

    Deep-dive documentation published for deployment models, identity and authorization, data handling, and the compliance program, with the full document library and request-access flow.

  • Subprocessor list updated

    General

    Published June 9, 2026

    Datadog and Grafana Labs listed for application logging, monitoring, and observability. Subprocessor changes carry advance notice under the DPA.

  • Penetration-testing report available

    Compliance

    Published May 12, 2026

    The latest third-party penetration test of the platform completed; the report is available on request.

Request access

Tell us who you are and what your review needs.

Resources