Third-Party Model Provider Data Sharing

What it is

When you use an AI model from a company other than Amazon through Bedrock, your data may be sent to that company and governed by their separate rules, not just AWS's rules.

Why it matters (compliance & governance perspective)

This provision creates a meaningful data governance gap: customers who believe they are operating entirely within the AWS compliance framework may unknowingly subject their data to third-party providers with different, potentially less protective terms.

This document changed recently

Consumer impact (what this means for users)

Using third-party foundation models through Bedrock means your prompts, documents, and outputs may leave AWS's direct control and become subject to terms you have not directly negotiated, which could affect your data protection, confidentiality, and compliance posture.

Institutional analysis (Compliance & governance intelligence)

REGULATORY LANDSCAPE: This provision directly engages GDPR Article 28 on sub-processor relationships and the obligation for data controllers to ensure sub-processors provide sufficient guarantees. It also implicates CCPA requirements for businesses relying on service provider exemptions, as onward data flows to third-party model providers may require additional contractual and disclosure steps. EU AI Act obligations for deployers of high-risk AI systems may also be triggered depending on the model and use case. GOVERNANCE EXPOSURE: High. The provision creates a layered data processing chain where the primary customer agreement with AWS does not fully govern the end-to-end data flow. Customers must independently assess each third-party model provider's terms, and the absence of a consolidated compliance framework covering all available models represents a significant due diligence burden for enterprise customers. JURISDICTION FLAGS: EU/EEA customers face the highest exposure, as GDPR requires documented sub-processor agreements and may require updated Records of Processing Activities (RoPA) for each third-party model provider used. California businesses should assess whether these onward transfers satisfy CCPA service provider chain requirements. Healthcare customers handling PHI should confirm whether third-party model providers have executed BAAs and whether Bedrock's BAA coverage extends to these providers. CONTRACT AND VENDOR IMPLICATIONS: Procurement and legal teams should maintain an up-to-date inventory of third-party model providers available through Bedrock and the applicable terms for each. Enterprise contracts with AWS should address whether AWS provides contractual flow-down protections to third-party model providers or whether customers must contract directly. Standard B2B agreements built on the assumption of AWS-only data processing will need review if third-party models are in use. COMPLIANCE CONSIDERATIONS: Data mapping exercises should capture third-party model provider data flows as distinct processing activities. Consent mechanisms and privacy notices may need updating to disclose these onward transfers. Vendor risk assessments should be conducted for each third-party model provider in the customer's Bedrock deployment.

Applicable agencies

Applicable regulations

Provision details

Other risks in this policy

• No-Training Commitment for Customer Content medium
• Customer Responsibility for Output Compliance high
• Acceptable Use Policy Restrictions medium
• Intellectual Property in AI Outputs medium
• Liability Limitation for AI Services high
• Service Modification and Termination Rights medium

Related Analysis

• Meta Removed 438 Sentences From Its Privacy Policy. Here Is What Disappeared.

  ConductAtlas detected a major restructuring of Meta’s privacy policy that removed detailed consumer rights disclosures and relocated them to separate documents.

• 23andMe Is Bankrupt. What Happens to Your DNA Now?

  Your genetic data may be transferred to a new owner as a business asset. Here is what the Terms of Service actually say and what you can do right now.