AI Governance & Security Vendor Wiki

7 min read

Amazon Q Business

One-liner — AWS’s managed enterprise AI assistant: a generative-AI chat/RAG layer over your enterprise content, with retrieval that honors each user’s existing source-system permissions and a contractual no-training commitment, delivered as a SaaS service inside your AWS account boundary.

Categories — enterprise-ai-assistant (primary), entitlement-aware-rag

What it does

Amazon Q Business is a fully managed generative-AI assistant for the workforce. Employees ask questions in natural language and get synthesized answers, summaries, and actions grounded in the company’s own documents and applications — wikis, file shares, ticketing, CRM, email, and 40+ connectors. It is AWS’s answer to Microsoft 365 Copilot, gemini-enterprise, glean, and openai-chatgpt-enterprise. The draw for an AWS-centric shop is that the index, the retrieval, and the model inference all run as an AWS service governed by IAM, CloudTrail, and AWS’s compliance envelope, rather than a third-party SaaS you have to onboard through vendor risk from scratch.

Where it sits in the stack

UX layer — the chatbot people actually use (enterprise-ai-assistant) — but its defining governance feature is entitlement-aware retrieval (entitlement-aware-rag), so it spans both. Its main risk exposure is sensitive data — it ingests and surfaces internal documents, so the controls that matter are (a) not leaking content to a user who shouldn’t see it, and (b) not leaking content to a model trainer. Q Business’s permission-aware retrieval and no-training stance are aimed squarely at both. It is not a network/egress control and not an untrusted-input firewall; pair it with ai-access-governance / dlp for shadow-AI and exfiltration concerns.

Deployment & architecture

  • SaaS, in-account. A managed AWS service (no servers to run); you create a Q Business application, attach data sources, and expose a web experience or embed via API.
  • Identity / SSO. Authenticates workforce users through AWS IAM Identity Center, which federates to an external IdP (Okta, Microsoft Entra ID, Ping) via SAML/OIDC and supports SCIM user/group provisioning. Newer setups use trusted identity propagation so the end user’s identity flows through to the data layer for authorization. An IAM-federation path also exists for apps not using Identity Center.
  • Entitlement-aware retrieval (ACL crawling). Connectors crawl each source’s access-control lists at the document level and store principal info (users, local and federated groups) in the Amazon Q Business User Store. At query time the assistant filters responses to documents the asking user is actually permitted to read. ACL/identity crawling is on by default for supported connectors; once enabled it cannot be turned off (a deliberate safety default). This is the feature that lets it be deployed over sensitive shares without becoming a leak engine — the same problem Copilot’s “oversharing” controversy raised.
  • Admin controls / guardrails. Admin console with global and topic-level guardrails (blocked words/topics, response controls), data-source scoping, and Amazon Q Apps governance.
  • Encryption. TLS 1.2+ in transit (1.3 recommended); encryption at rest with AWS-owned or customer-managed KMS keys.
  • Audit / logging. API and user-activity logging via AWS CloudTrail; integrates with the broader AWS security tooling (Macie, CloudWatch). Conversation data is retrievable via the Q Business API, which is the hook for any archival/export pipeline.

Positioning & differentiators

  • No-training commitment (the headline for a CTO). Per AWS docs: “Amazon Q Business does not use customer data for service improvement or for improving underlying LLMs.” Unambiguous and primary-sourced.
  • Native permission-aware RAG. Unlike a bare ChatGPT/Claude deployment where you bolt on retrieval, Q Business’s connectors enforce source ACLs out of the box — its closest peers here are glean and microsoft-365-copilot (Graph), not the raw frontier-model assistants.
  • AWS compliance envelope. Q Business itself is attested for HIPAA (HIPAA-eligible since Oct 2024), SOC 1/2/3, PCI, and ISO 42001 (the AI-management-system standard — relatively rare and notable). It rides AWS’s broader SOC/ISO 27001/FedRAMP posture.
  • Tradeoff vs. frontier assistants. Buyers generally rate its raw answer quality and model choice below ChatGPT Enterprise / Claude Enterprise; the value is governance and AWS-native integration, not best-in-class reasoning. Model is AWS-selected (Bedrock-family), not user-pickable to the same degree.

Ownership, funding & M&A

Amazon Q Business is a product of Amazon Web Services, a segment of Amazon.com, Inc. (NASDAQ: AMZN), a public company. Launched at AWS re:Invent (announced Nov 2023, general availability 2024). No standalone funding or M&A — corrected from the stub, which incorrectly listed independent. Ownership confidence: high.

CTO / hedge-fund lens

  • Day-1 if the firm is already AWS-centric and wants an internal assistant grounded in its own content without standing up third-party SaaS or a self-built RAG stack. The IAM/CloudTrail/KMS familiarity shortens the security review.
  • No-training + in-account + permission-aware retrieval is a clean story for an investment firm worried about MNPI/research leaking into a model or across Chinese walls. The ACL-honoring retrieval is the control that maps to information-barrier requirements.
  • SR 11-7 / model risk: Q Business is a productivity assistant, not a model that prices or trades, so it’s largely outside core model-risk scope — but its outputs informing analysts may fall under your AI governance policy; log usage via CloudTrail and govern it through ai-governance-platform / enterprise-grc.
  • Comms surveillance / eDiscovery gap (verify). Conversations are accessible via API, but Q Business is not a purpose-built comms-surveillance archive. A regulated fund treating the assistant as a monitored communications channel (MAR/MNPI, SEC/FINRA books-and-records) would need to build export into an archive and feed a comms-surveillance tool (behavox, steeleye, theta-lake); native retention/legal-hold controls for chat are limited. Treat this as an open item, not a solved one.
  • Fit: strongest for mid/large AWS shops; a 50-person fund with no AWS footprint gets a simpler path from ChatGPT Enterprise, Claude Enterprise, or Glean. Hence medium fit overall.

Competitors / alternatives

microsoft-365-copilot, gemini-enterprise, glean, openai-chatgpt-enterprise, anthropic-claude-enterprise, perplexity-enterprise. For the permission-aware-RAG dimension specifically: glean, microsoft-graph, knostic.

Open questions / to verify

  • Exact list of AWS Regions where Amazon Q Business is GA (check the AWS Regional Services List); commercial vs GovCloud (US) scope for FedRAMP.
  • Native conversation retention / legal-hold / eDiscovery controls and whether a supported export path to comms-surveillance archives exists — current read is “API access yes, purpose-built retention no.”
  • Whether GDPR DPA terms and data-residency guarantees pin processing to the selected Region with no cross-region inference fan-out by default (cross-region inference is a documented feature — confirm default and opt-out).
  • Degree of model choice / Bedrock model selection exposed to admins.

Sources

History

  • [2026-06-28] Stub created from seed registry.
  • [2026-06-28] Researched; corrected ownership independent→public (AWS / Amazon.com, NASDAQ: AMZN, high confidence). Established no-training commitment (primary doc), entitlement-aware ACL-crawling retrieval via connectors + IAM Identity Center/SCIM/trusted identity propagation, CloudTrail audit logging, TLS/KMS encryption, and compliance attestations (HIPAA-eligible, SOC 1/2/3, PCI, ISO 42001). Added entitlement-aware-rag as secondary category. Flagged comms-surveillance/eDiscovery retention as an open gap for regulated funds. Set hedge_fund_fit medium, confidence medium.

Graph View

Backlinks