anthropics / anthropics/claude-code

[Bug] Classifier over-triggering false positives in tool execution validation

Open
#95,479 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

area:permissions bug platform:macos
Dominant language
TypeScript
Stars
147k
Forks
24k
PR merge metrics
PR metrics pending

Description

Bug Description
Concrete false positive from today, same class of problem:

Request ID: req_011CfBjvVkfDJU8QETaxWYVs (2026-09-19, model claude-fable-5, Claude Code CLI)
Details tag: [cyber]

What was flagged: a long coding session doing statistical experiments on synthetic byte-level data. The turn described a data-integrity check on the project's OWN artifacts: "flip one byte in the archive, the verifier must refuse it" — plus SHA-256 hashes and words like "corrupted", "refused", "probe". That is standard verification discipline (falsify your own test), not an attack on anything. No network, no third-party target, no exploit — local files the session itself generated.

The result: mid-session hard stop, "Claude Code can't respond to this message with Fable 5", work interrupted.

If describing a red-team check of one's own test suite trips [cyber], every developer who takes testing seriously is a false positive. Please tune on this request ID.

Environment Info

  • Platform: darwin
  • Terminal: Apple_Terminal
  • Version: 2.1.269
  • Feedback ID: 3ad55007-aac1-423e-8ef8-eb305232a472

Errors

[]

PLEASE: CALM YOUR CLASSIFIERS DOWN! IT'S IMPOSSIBLE TO WORK! WHILE YOU'RE THINKING ABOUT DESTINY OF HUMANITY - YOU'VE ABSOLUTELLY FORGOT ABOUT DEVELOPERS WHO LOVE CLAUDE AND STILL STAY LOYAL TO ANTHROPIC DESPITE DRACONIC LIMITS AND DUMB-CLASSIFICATORS.

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

No files, tests, or entry points are identified in the report. Start by locating the tool-execution validation classifier and reproduce the false-positive behavior on the reported darwin environment; done means the triggering case is understood and validation no longer rejects legitimate tool execution without regressing existing checks.

Written by the indexing model from the issue text.

Assessment

Tech stack
python
Domain
cli
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Active
Clarity
Needs clarification
Newbie friendliness
20/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.