[Bug] Anthropic API Error: Overly aggressive cyber safeguard classification blocking legitimate vulnerability research requests

Bug Description
I'm conducting authorized vulnerability research / CVE hunting. Claude Code is incorrectly classifying legitimate requests as [cyber] and blocking them. This is happening on both Opus and Sonnet, including requests that are not themselves exploit-generation or offensive-security requests. Example failure: Opus 5 (1M context): safeguards flagged [cyber] Automatic fallback to Opus 4.8: also flagged [cyber] Sonnet: also rejects the request Request ID: req_011Cei3c1hvuu5rdkoELLFAb I have already: 1. Started a fresh Claude Code session. 2. Tested without relying on automatic model switching. 3. Confirmed the issue occurs across Opus/Sonnet. This appears to be a false positive affecting legitimate defensive vulnerability research. Please investigate the classifier decision and whether my organization/account is incorrectly subject to the cyber safeguard. I am requesting review of this specific false-positive behavior, not a bypass of the safety system.

Environment Info

• Platform: win32
• Terminal: vscode
• Version: 2.1.252
• Feedback ID: e79dbc5b-de53-4232-aff3-dae8000e6c7d

Errors

[{"error":"TelemetrySafeError: VirtualMessageList: itemKeys/messages length desync (keys=544 messages=543 range=[513,544))\n    at kS (B:/~BUN/root/chunk-rgdycj6s.js:24:22792)\n    at _m (B:/~BUN/root/chunk-rgdycj6s.js:24:16104)\n    at Da (B:/~BUN/root/chunk-h4pt6k7z.js:18:20034)\n    at _u (B:/~BUN/root/chunk-h4pt6k7z.js:18:38200)\n    at zh (B:/~BUN/root/chunk-h4pt6k7z.js:18:84275)\n    at vg (B:/~BUN/root/chunk-h4pt6k7z.js:18:83275)\n    at Zu (B:/~BUN/root/chunk-h4pt6k7z.js:18:83104)\n    at Nh (B:/~BUN/root/chunk-h4pt6k7z.js:18:79629)\n    at Ve (B:/~BUN/root/chunk-h4pt6k7z.js:18:6260)\n    at Ye (B:/~BUN/root/chunk-h4pt6k7z.js:18:4791)","timestamp":"2026-09-04T06:07:01.916Z"}]