Claude Code Update v2.1.325 Blocks Disallowed Bedrock Requests

Claude Code Update v2.1.325 Blocks Disallowed Bedrock Requests

  • Claude Code v2.1.325 stops token-counting requests when Bedrock is not allowed.
  • Terminal output from claude -p now allows only text styling control sequences.
  • Background sessions use the claude.ai login they started with, rather than credentials from the shell.

Security fix for decoding compressed responses

Claude Code Update v2.1.325 Blocks Disallowed Bedrock Requests

Claude Code v2.1.325 fixes a security issue in the decoding of responses compressed with zstd, a format used to compress these responses.

Claude Code users can check whether their release includes the fix. The release notes do not describe the issue's impact or the conditions needed to exploit it.

Claude Code v2.1.325 release notes

Bedrock requests stop when the provider is not allowed

Claude Code Update v2.1.325 Blocks Disallowed Bedrock Requests

Amazon Bedrock is a service Claude Code can use to send token-counting requests. In v2.1.325, Claude Code no longer sends a conversation to Bedrock when the managed setting allowedProviders does not allow Bedrock.

Organizations that use Bedrock can check allowedProviders in their managed settings. The update fixes the previous behavior, which sent token-counting requests to Bedrock even when it was missing from the allowlist.

SituationPrevious behaviorv2.1.325 behavior
Bedrock is not on the allowlistSend a token-counting request to BedrockDo not send the conversation to Bedrock
Run claude -p in a terminalControl sequences beyond text could appear in the outputBlock control sequences other than those used for text styling

Changes to terminal output from claude -p

Claude Code Update v2.1.325 Blocks Disallowed Bedrock Requests

When you run claude -p in a terminal, control sequences could appear alongside the text output. Claude Code now allows only control sequences used to style text.

Output sent through a pipe or written to a file is unchanged. This update applies when command output is displayed directly in the terminal.

Login and credentials in background sessions

A background session started with a claude.ai login could use the API key or token from the shell that launched it, and charge usage to those credentials. That issue is fixed: the session now uses the claude.ai login instead of the shell's credentials.

You can no longer switch to a different login during a session. If a login change is rejected, Claude Code displays a message with available options.

  1. Check which login account you used when starting the background session.
  2. If a login change is rejected, follow the instructions in the displayed message.

What to do when an organization policy blocks a request

When an organization policy blocks a request, Claude Code shows next steps beneath the error. You can run /rewind to change the message, or run /clear.

  1. Run /rewind and change the message.
  2. Or run /clear.

FAQ

Are token-counting requests blocked when Bedrock is allowed?

The release notes specify a block when the managed setting allowedProviders does not allow Bedrock.

Does this change how claude -p works with pipes or files?

No. The change applies to text output displayed in the terminal. Output sent through pipes or written to files is unchanged.

Can I continue after an organization policy blocks a request?

Follow the instructions beneath the error. You can use /rewind to change the message or run /clear.

댓글