[Bug] False positive cyber safeguard blocks agentic workflow, causing token loss without recovery path

Resolved 💬 3 comments Opened Jun 5, 2026 by mevvmevvcodinggarg Closed Jul 11, 2026

Bug Description
Subject: False positive on cyber safeguard — token loss during agentic workflow

Request ID: req_011CbjamdwURFvMLxhwasjsT

While running an autonomous (agentic) workflow in Claude Code, one request was blocked with the message that it "appears to violate the Usage Policy" and "triggered cyber-related safeguards." My task was legitimate and contained nothing harmful or cyber-offensive.

The core problems I'm reporting:

  1. False positive: The blocked request was part of normal development work and did not involve creating malware, exploits, or any malicious activity. The cyber safeguard appears to have misclassified it.
  1. Token loss: Before the final request was blocked, the agent had already consumed roughly 1,000,000 tokens across sub-tasks. Those tokens were billed to me even though the workflow was halted and could not be completed.
  1. No recovery path: Once the request was blocked, the main thread could not continue. There was no way to resume the workflow, so all the work and tokens up to that point were lost.

What I'm requesting:

  • A review of the block on Request ID req_011CbjamdwURFvMLxhwasjsT to confirm whether it was a false positive.
  • Consideration of a token refund/credit for the ~1M tokens consumed by the interrupted workflow.
  • Guidance on how to avoid this happening again, and whether the Cyber Verification Program applies to my use case.

Please let me know what additional details you need. Thank you.

Environment Info

  • Platform: darwin
  • Terminal: ghostty
  • Version: 2.1.162
  • Feedback ID: b7b223ab-417f-481b-801c-269a387c936a

Errors

[{"error":"Error: 400 {\"type\":\"error\",\"error\":{\"type\":\"invalid_request_error\",\"message\":\"The following domains are not accessible to our user agent: ['reddit.com']. Read more: https://support.anthropic.com/en/articles/8896518-does-anthropic-crawl-data-from-the-web-and-how-can-site-owners-block-the-crawler\"},\"request_id\":\"req_011CbitUE6dZHFraD2X67wij\"}\n    at generate (/$bunfs/root/src/entrypoints/cli.js:11:65489)\n    at makeRequest (/$bunfs/root/src/entrypoints/cli.js:51:7694)\n    at processTicksAndRejections (native:7:39)","timestamp":"2026-06-04T20:56:53.963Z"}]

View original on GitHub ↗

This issue has 3 comments on GitHub. Read the full discussion on GitHub ↗