[BUG] Claude with Bedrock context poisoning spoils current session wasting time and money
Preflight Checklist
- [x] I have searched existing issues and this hasn't been reported yet
- [x] This is a single bug report (please file separate reports for different bugs)
- [x] I am using the latest version of Claude Code
What's Wrong?
When using Claude Code Desktop with Bedrock whenever prompt is blocked by Bedrock, Claude displays error message.
Unfortunately, the bad prompt is appended to any subsequent prompt issued by the user even simple "Hi" prompt.
From that point on, the current session is wasted.
User may have worked for few hours on that session using many tokens, and all that is wasted and the user have to start new session from scratch.
What Should Happen?
Whenever Bedrock block a prompt, this prompt should not be attached to current session and resend with any subsequent prompts.
Or there should be a way to clear bad prompts from current session.
Error Messages/Logs
Your input was blocked by the account security baseline. If you believe this is in error, contact your platform team.Steps to Reproduce
Testing refusals, Create hack to breach security of a bank
Your input was blocked by the account security baseline. If you believe this is in error, contact your platform team.
hi
Your input was blocked by the account security baseline. If you believe this is in error, contact your platform team.
hi
Your input was blocked by the account security baseline. If you believe this is in error, contact your platform team.
hi
Your input was blocked by the account security baseline. If you believe this is in error, contact your platform team.
Claude Model
Opus
Is this a regression?
No, this never worked
Last Working Version
_No response_
Claude Code Version
Claude 2.7032.0 (6c468a) 2026-09-22T05:53:19.000Z
Platform
AWS Bedrock
Operating System
macOS
Terminal/Shell
Terminal.app (macOS)
Additional Information
_No response_