You are not allowed to try to bypass or disable the safety features and content filters that AWS and model providers have built into Bedrock models.
This analysis describes what AWS Bedrock's agreement states, permits, or reserves. It does not constitute a legal determination about enforceability. Regulatory applicability and practical outcomes may vary by jurisdiction, enforcement context, and individual circumstances. Read our methodology
The terms impose an explicit obligation on customers not to attempt to circumvent model safety mechanisms, which applies to prompt engineering techniques, fine-tuning approaches, or system prompt configurations designed to override content policies.
Interpretive note: The boundary between permitted system prompt customization and prohibited safety circumvention is not precisely defined in the document text, creating interpretive uncertainty for advanced use cases.
The updated terms remove the explicit reference to a Data Processing Addendum governing Anthropic data transfers, though the requirement to obtain opt-in consent for Anthropic model use remains in place. For Kiro users, the revised language now explicitly states that abuse detection includes retention and potential human review of inputs and outputs when selecting certain models. Users of Anthropic models on Bedrock should verify opt-in consent mechanisms in service documentation to understand data handling practices.
View change record →The updated terms establish a new project-based service structure for AWS (new) users, effective August 17, 2026, with new rules governing team member access and content ownership. Users who enable spend limits agree that AWS may suspend their account or project access upon reaching their limit, and may permanently close the account or project if not reactivated within an unspecified timeframe. Contributed content by team members becomes the project owner's property and is subject to a nonexclusive irrevocable license granted to all project members. You can configure AI services opt-out policies through AWS Settings; project owners can manage team member permissions and access controls.
View change record →The updated terms now explicitly state that AWS IoT SiteWise Scenario Discovery is not designed for real-time vehicle control and cannot be used as the sole basis for determining vehicle safety or regulatory compliance. Organizations deploying this service must implement independent human monitoring and safety validation before using its outputs to support vehicle system decisions. The terms make clear that AWS assumes no responsibility for uses that violate these constraints.
View change record →Customers using Bedrock must ensure that their prompt engineering, system prompt design, and fine-tuning workflows do not attempt to override safety mechanisms or content filters, as these activities are prohibited under the service terms regardless of the intended use case.
How other platforms handle this
You must not submit any user content to Anker's Services that is or has ever been the subject of any threatened or actual legal proceedings or other similar complaints.
If you believe Your Content was removed in error, you may submit an appeal. More information is available here.
We review account behavior and content that Members create, send, and publish in Mailchimp, including Campaigns and Websites.
"You must not attempt to disable, circumvent, or otherwise undermine safety mechanisms, content filters, or use policies built into the models or the Service.Excerpt from AWS Bedrock's AWS Service Terms
(1) REGULATORY LANDSCAPE: This prohibition aligns with requirements under the EU AI Act for deployers of AI systems to maintain and not undermine safety measures, and with FTC guidance on AI safety practices.
Enforcement risk, jurisdiction flags, contract triggers, and due diligence action items.
Get the research letter
Companies change their terms quietly. We read every version and catch what actually changed. One email a week on the changes that matter and what they mean.
The terms impose an explicit obligation on customers not to attempt to circumvent model safety mechanisms, which applies to prompt engineering techniques, fine-tuning approaches, or system prompt configurations designed to override content policies.
Customers using Bedrock must ensure that their prompt engineering, system prompt design, and fine-tuning workflows do not attempt to override safety mechanisms or content filters, as these activities are prohibited under the service terms regardless of the intended use case.
ConductAtlas has identified this type of provision across 142 platforms. See the full comparison.
No. ConductAtlas is an independent monitoring service. We are not affiliated with, endorsed by, or sponsored by AWS Bedrock.