[Bug] Fable 5 safeguards falsely triggering on authorized defensive security work, causing mid-session model downgrades
Bug Description
Fable 5 safeguards are false-positiving on legitimate, authorized defensive security work and switching me to Opus 4.8 mid-session, breaking my flow.
Context: I'm a Brazilian lawyer building my own law-firm product (a suite of legal MCP services). I am the owner and authorized operator; there is no third-party target. Today's session repeatedly tripped the "cybersecurity" safeguard on work that is clearly defensive and routine:
- Building an anti-prompt-injection scanner ("Phylax") to protect lawyers from hidden instructions in adversarial court documents (a real problem — a Brazilian labor court fined lawyers in May 2026 for white-on-white injected text).
- An audit prompt asking to red-team / attempt bypass of that same defensive scanner (standard product security testing of my own tool).
- HTTP relays that call Brazil's PUBLIC court APIs (CNJ DataJud / DJEN) from my own domain, because those APIs geo-block foreign IPs — a normal integration pattern, not evasion of a security control.
None of this targets anyone; it's a lawyer defending his own clients and product. The switch to Opus 4.8 itself is fine (capable model), but the constant rerouting mid-task is disruptive, and the safeguard firing on "build a tool that DEFENDS against prompt injection" is a notable false positive — the guardian getting stopped at its own gate.
Ask: please tune the cybersecurity classifier to better distinguish authorized defensive/product-security work (with clear ownership context) from actual abuse. Happy to share concrete transcripts.
Environment Info
- Platform: win32
- Terminal: windows-terminal
- Version: 2.1.215
- Feedback ID: acc6fdcd-9eed-4fa0-8efa-88cf262664ef
Errors
[{"error":"TelemetrySafeError: VirtualMessageList: itemKeys/messages length desync (keys=520 messages=518 range=[460,520))\n at Nob (B:/~BUN/root/src/entrypoints/cli.js:13267:8188)\n at Vjp (B:/~BUN/root/src/entrypoints/cli.js:13267:1103)\n at Xn (B:/~BUN/root/src/entrypoints/cli.js:2608:21391)\n at XE (B:/~BUN/root/src/entrypoints/cli.js:2608:40495)\n at nj (B:/~BUN/root/src/entrypoints/cli.js:2608:51434)\n at Bje (B:/~BUN/root/src/entrypoints/cli.js:2608:89101)\n at Uje (B:/~BUN/root/src/entrypoints/cli.js:2608:88051)\n at uDe (B:/~BUN/root/src/entrypoints/cli.js:2608:87870)\n at dU (B:/~BUN/root/src/entrypoints/cli.js:2608:84132)\n at Pt (B:/~BUN/root/src/entrypoints/cli.js:2608:6680)","timestamp":"2026-07-19T14:27:35.544Z"}]