anthropics / anthropics/claude-code

[Bug] CVP-approved organization still receives cyber blocks and model fallbacks in Claude Code

Open
#93,822 0 comments 0 reactions 0 assignees View on GitHub
api:anthropic area:auth area:model bug duplicate platform:macos
Dominant language
Python
Stars
145k
Forks
23.1k
PR merge metrics
PR metrics pending

Description

**Bug Description**
CVP-approved org (Active) still gets [cyber] blocks and model fallbacks.

Org f076b231-2eeb-43cb-9b5c-cf9b0379d644 — Claude Pro, first-party OAuth,
no API key. Verification Portal shows Cyber Verification Program = Active,
both controls satisfied. Org ID confirmed matching against the approval and
oauthAccount.organizationUuid.

Two symptoms: hard [cyber] API errors (req_011CeYT6a7VqCRYQjD8hJNb5,
req_011CeYYtLsPa96NxrApezP4f), and 25 distinct Opus 5 -> Opus 4.8
model_refusal_fallback events (all apiRefusalCategory=cyber, scope=session)
between 2026-08-29 and 08-31. Once a session is flagged even content-free
messages are blocked, and the 4.8 fallback inherits the same context and
fails too.

Work: C2 framework development in my own isolated lab — the category my CVP
application covers, and the one the help center calls high-risk dual use
"adjustable through the CVP". Not claiming a false positive on benign
content; reporting that CVP Active has no observable effect.

Question: does CVP approval reach the classifier for consumer (Pro/Max)
claude.ai OAuth sessions in Claude Code at all? Behavior on an Active org
is indistinguishable from no approval.

Also filed: anthropics/claude-code#84689, #93224.

**Environment Info**
- Platform: darwin
- Terminal: Orca
- Version: 2.1.269
- Feedback ID: 78e83aa6-136c-4748-af3d-e99c10614314

**Errors**
```json
[]
```

Contributor guide

No contributing guide indexed for this repository

Research direction

No source files or tests are named. Start by correlating the two request IDs, the 25 model_refusal_fallback events, feedback ID, and OAuth organization UUID across the Claude Code CVP and classifier path; done means establishing whether Active approval reaches consumer OAuth sessions and explaining the blocks and fallbacks.

Written by the indexing model from the issue text.

Assessment

Domain
api, authentication, security
Issue type
Bug
Difficulty
4/5
Estimated time
3-5 days
Activity status
Active
Clarity
Mostly clear
Newbie friendliness
30/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.