Its theme's rank, 14 days: 2 on the Claude Code sheet
52 people reported
All time, since Sep 15, including 1 thumbs-up on GitHub.
Cyber safeguard pauses sessions
“Opus 5.5 [cyber] safeguard false positive.”
Is this your bug? What gets counted gets fixed. Pile on.
Paste this into your agent. It files your report, with your error and version.
I hit this bug: rapsheet.dev/sheets/claude-code/issues/7298/file Follow it to file my report. Show me the text before you send.
No install. No account. Public and anonymous. The link asks your agent for one report, nothing else. What it sends
Cyber safeguard flag pauses sessions and forces model switch. Sessions stop with "Details: [cyber]" on code review or coding tasks, offering a switch to Opus 4.8 or a prompt edit.
Is this your bug?
File my report- Where
- model safeguards
- Error text
Fable 5's safeguards flagged this message. Our intentionally broad safeguards allow us to deliver more capabilities faster, but can sometimes flag- Versions
- 2.1.220 to 2.1.318
- Platforms
- Linux, macOS, Windows
- How bad
- Breaks a normal workflow.
Part of: Safeguards block session requests34 bugs
- Cyber safeguard blocks sessions377 people
- Reasoning_extraction safeguard block134 people
- Safeguards refuse work requests48 people
- Safeguards block requests47 people
- Classifier halts routine replies12 people
- Requests blocked with a [cyber] flag8 people
- Refusal on own-code hardening11 people
- SSH to own devices refused7 people
- Pentest work blocked6 people
- CTF requests flagged as malicious6 people
Show all 33Show only the largest 10
- Classifier blocks own-project work6 people
- Classifier halts reverse engineering5 people
- Safeguards flag verified cyber work6 people
- Firmware requests flagged as cyber6 people
- Classifier stops benign replies5 people
- reasoning_extraction API error5 people
- Safeguard blocks all messages4 people
- [bio] classifier blocks responses4 people
- Reasoning extraction safeguard block4 people
- Security investigation requests blocked3 people
- Classifier stops DeFi bot work3 people
- Safeguards block CVP members3 people
- "[cyber]" flag stops sessions1 person
- Safety classifier blocks agent3 people
- Fictional hacking content flagged3 people
- Cyber verification prompt recurs2 people
- Cyber safeguards block own devices2 people
- Safety classifier stops auth work2 people
- Phishing research project flagged2 people
- Cyber stop on report tasks2 people
- Cyber safety block on audits2 people
- Biological safeguard triggers in sessions with no biology content2 people
- Unexpected safeguards notice2 people
Original reports52, newest first
- Legitimate cybersecurity questions incorrectly flagged by safety filter#107465
- Bug Report: Claude Code refuses assistance with active network defense during cyber attack#107172
- Classifier flags security coursework as cyber and withholds replies, with no reason shown#107126
- Can't develop my own game, can't sign my own certs, can't protect my own things from exploiters.#106892
- Bug Report: Claude Code unable to perform any actions#106722
- Claude Code blocks legitimate cybersecurity work#106665
- False Positive in Server Cybersecurity Analysis#106626
- Safety filter blocks legitimate security training request#106344
- Claude False Positives: Legitimate Obfuscation Detection Prevents Compatibility With My Obfuscator#106309
- Cyber security requests incorrectly flagged by safeguards, blocking legitimate work#105871
Show all 52Show only the newest 10
- Constant prompts to downgrade to Opus 4.8 citing safety concerns#105708
- False positive [cyber] safeguard block on Opus 5.5 for defensive security work#105584
- Overly broad content filter blocks benign computing content#105533
- False Positive "cyber" Safeguard Blocks Defensive PC Security Check Request#105327
- False-positive safeguard flag force-downgrades Opus 5.5 → Opus 4.8 mid-session on a legitimate code-obfuscation request the model itself proposed#105262
- Opus 5.5 Safety Filter Repeatedly Blocks Legitimate Cybersecurity Work#105232
- Cyber safety classifier falsely flags legitimate security work on own devices#105224
- Overly aggressive safety filter flagging legitimate source code protection requests as cybersecurity violations#105152
- Excessive false positive refusals on legitimate security work#104906
- Opus session repeatedly paused by [cyber] safeguard on defensive review#104883
- Anthropics Misunderstanding of How Security Works#104266
- Involuntary model downgrade and recurring safeguard flag during long sessions#104066
- Classifier Blocks Defensive Security Work on Own System#103899
- Anthropic API Error: Opus 5.5 Safeguard False Positive on Firmware Code Review#103561
- Opus 5.5 [cyber] classifier false-positives on coordinated vuln disclosure, and the flag persists across benign follow-up turns#102795
- Cyber safeguard false positive on defensive RASP code (root/Frida/Xposed detection) forces mid-session model switch#102663
- Cyber Safety Filter False Positive Blocks Defensive Access Control Fix#102407
- Opus 5.5 cybersecurity safeguards repeatedly block defensive infrastructure work#101999
- Cyber classifier blocks YouTube transcript extraction and downgrades model to Opus 4.8#101608
- Opus 4.5 sessions flagged for compiler obfuscation work (trading marketplace)#101320
- Over-refusal of LLVM obfuscation requests for red team work#101109
- Fable 5 safeguards repeatedly flag legitimate cybersecurity code in session context#100839
- False-positive cyber safeguard stops benign analytics/maintenance review#100758
- Cyber classifier forces model switch to Opus 4.8 on legitimate product-protection work#100408
- Code obfuscation requests incorrectly flagged as policy violation#99462
- Fable 5 safeguard false positive: [cyber] flag on session editing its own permission hooks (req_011CffmtDtTRX4NrwePWGLfh)#99295
- False positive [cyber] flagging for defensive Git command guards in user repositories#99205
- Safety-classifier fallback silently switches whole session to Opus 4.8; switchModelsOnFlag:false is ignored#98875
- Safety classifier over-blocks defensive work on repository's own agent-safety hooks#98596
- [cyber] False positive safety block during cloud IAM source repository recovery (req_011CfTq4r5u9p4SjGEgzcfzC)#97608
- [cyber] False positive on desktop screen capture during cross-device widget development (req_011CfT9YfeiAy3y1mruk6UfJ)#97554
- False positive security flag on non-security iOS development queries#96785
- Anthropic API Error: Overly aggressive content filtering blocking legitimate RPA refactoring tasks#96242
- Anthropic API Error: Unexpected model downgrade to Opus 4.8 with safeguards flag#95884
- How is a company smart enough to build Opus dumb enough to guard it with something my grandma with Alzheimer's could have coded better?#95758
- Fable 5 safeguards chain-flag entire sessions: children's song lyrics and routine Korean e-commerce instructions all flagged (6 request IDs)#95500
- [cyber] safety classifier stopped 7+ turns across 3 sessions while building my own home camera system on my own LAN (Opus 5 and Fable 5.1; Opus 4.8 unaffected)#95483
- Model switch to Opus triggered during security analysis task with Claude 3.5 Sonnet#95299
- False positive cyber classification during legitimate local plugin development inspection#95285
- Keep getting flagged wrongly for [cyber]#94747
- Opus 5 cyber safeguard incorrectly blocks defensive security review of user's own code#94706
- Claude 5.1 Fable Cyber Safeguards incorrectly flagging legitimate coding requests - /Low-priority mode switch#94517