RAP Sheet home
Install the bug reporting skill
Its theme's rank, 14 days: 2 on the Claude Code sheet
377 people reported

All time, since Sep 14, including 23 thumbs-up on GitHub.

Cyber safeguard blocks sessions

“Claude Code repeatedly blocked my request with “safeguards flagged this session” and Details: [cyber].”

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/1695/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

Fading. 27 reports and thumbs-up in the last 7 days and 40 the week before, down from 85 a week at the late-September peak. Not fixed, but a workaround is known.

Cyber safeguard blocks sessions with [cyber] flag on varied requests. Sessions are blocked with "safeguards flagged this session" on tasks from code packaging and log analysis to a greeting like "hii".

Is this your bug?

File my report
Where
API client
Error text
safeguards flagged this session
API Error: Sonnet 5's safeguards flagged this message. Our intentionally broad safeguards allow us to deliver more capabilities faster, but can sometimes flag…Pasted in 4 reports
API Error: Mythos 5.1's safeguards flagged this message. Our intentionally broad safeguards allow us to deliver more capabilities faster, but can sometimes flag…Pasted in 2 reports
API Error: Opus 4.8's safeguards flagged this message. Our intentionally broad safeguards allow us to deliver more capabilities faster, but can sometimes flag…Pasted in 2 reports
Versions
2.1.207 to 2.1.319. Still reported on the newest version, 2.1.319.
Platforms
Android, Linux, macOS, Windows

Part of: Safeguards block session requests34 bugs

  1. Reasoning_extraction safeguard block134 people
  2. Cyber safeguard pauses sessions52 people
  3. Safeguards refuse work requests48 people
  4. Safeguards block requests47 people
  5. Classifier halts routine replies12 people
  6. Requests blocked with a [cyber] flag8 people
  7. Refusal on own-code hardening11 people
  8. SSH to own devices refused7 people
  9. Pentest work blocked6 people
  10. CTF requests flagged as malicious6 people
Show all 33Show only the largest 10
  1. Classifier blocks own-project work6 people
  2. Classifier halts reverse engineering5 people
  3. Safeguards flag verified cyber work6 people
  4. Firmware requests flagged as cyber6 people
  5. Classifier stops benign replies5 people
  6. reasoning_extraction API error5 people
  7. Safeguard blocks all messages4 people
  8. [bio] classifier blocks responses4 people
  9. Reasoning extraction safeguard block4 people
  10. Security investigation requests blocked3 people
  11. Classifier stops DeFi bot work3 people
  12. Safeguards block CVP members3 people
  13. "[cyber]" flag stops sessions1 person
  14. Safety classifier blocks agent3 people
  15. Fictional hacking content flagged3 people
  16. Cyber verification prompt recurs2 people
  17. Cyber safeguards block own devices2 people
  18. Safety classifier stops auth work2 people
  19. Phishing research project flagged2 people
  20. Cyber stop on report tasks2 people
  21. Cyber safety block on audits2 people
  22. Biological safeguard triggers in sessions with no biology content2 people
  23. Unexpected safeguards notice2 people

Original reports387, newest first

  1. Anthropic API Error: False positive security block on legitimate code audit request#108204
  2. Cyber safeguard blocks benign follow-ups in session after pentest comman#108138
  3. Claude Code pauses with "[cyber]" flag on benign prompts, Claude.ai works#108096
  4. Opus 5.5 Safeguards Flag Session with [cyber] False Positive#108067
  5. Session incorrectly flagged as [cyber] for JWT security work#107864
  6. Opus 5.5 safeguards incorrectly flag non-cybersecurity work (Arc time series DB fix)#107762
  7. False Positive Flag on Defensive Security Lab Scenario#107492
  8. API Error: Cyber Safeguards Falsely Flag Non-Security Build Task#107474
  9. Cyber Safeguard Blocks Local Security Lab Request Without Internet Access#107446
  10. Opus 5.5 safety classifier false positives on authorized defensive security audits#107166
Show the newest 100 of 387Show only the newest 10
  1. [cyber] false positive: safeguards block benign wait-time question in isolated security lab#107131
  2. Cyber classifier false positive on authorized web pentest prompt#107095
  3. Claude Code refuses legitimate infrastructure security analysis as misuse#107004
  4. Anthropic API Error: Content flagged by safeguards policy#106956
  5. Over-refusal: Claude Code declines authorized in-house security test suite development#106950
  6. Fable 5 Cybersecurity Safeguard Falsely Blocks Korean Greeting "안녕"#106927
  7. Opus 5.5 safeguards falsely flag non-cyber request with [cyber] error#106869
  8. Safeguards flag legitimate bug-analysis session despite CVP enrollment#106755
  9. Session Flagged for Security Content During OSCP Study#106724
  10. Opus 5.5 refuses authorized bug bounty work despite CVP approval#106700
  11. Overly Broad Refusal on Nmap Wrapper Script Refactoring for Intranet Use#106608
  12. Legitimate bug bounty request incorrectly flagged by cyber safeguard#106589
  13. Cyber safeguard false positive blocks smart contract security audits#106508
  14. Cybersecurity safeguard incorrectly triggers on defensive security questions about user's own app#106425
  15. Claude Code refuses security/VAPT review of local application#106423
  16. Failing to security review https://github.com/H1z3-br0/CVE-2026-98432#106285
  17. Bug Report: Security flaw patch review request#106230
  18. Claude Code refuses security review of user's own project#106215
  19. Claude Code repeatedly interrupted by cyber guardrails during wind turbine site systems work#106201
  20. Model classifier over-triggers on legitimate bug bounty work#106189
  21. Mythos 5.1 Repeatedly Flags Bug Bounty Work Under CVP Defensive Tier#106181
  22. Safety Classifier Interrupts Authorized Defensive Security Test Generation#106169
  23. Safeguards Flag Legitimate API VAPT Request as Harmful#106141
  24. Session incorrectly flagged as unsafe on user's own project#106133
  25. False positive from the cyber safeguard. An Opus subagent...#106129
  26. Constant safeguard security flagging blocks basic questions#106084
  27. Session paused by Opus 5.5 cyber safeguard during legitimate security work#105881
  28. Cyber Classifier False Positive Blocks Hardware-in-the-Loop Test Prompts#105820
  29. Work stops unexpectedly with no error message#105695
  30. Overly restrictive refusals on legitimate security hardening and pentest-style requests#105637
  31. Cyber Error Returned for "hi" Message with Opus 5.5#105630
  32. Anthropic API Error: Overly aggressive [cyber] blocking for authorized security testing#105544
  33. Cyber safeguards incorrectly refuse defensive security investigation#105425
  34. Bug Report: False positive flagged on cybersecurity class assignment#105400
  35. Cyber safeguard incorrectly blocks legitimate request#105391
  36. Opus 5.5 flags security work with no explanation of trigger, location, or reason#105377
  37. API Error: Request Blocked for Security Testing on Own Project#105306
  38. Cyber Safeguard False Positive Blocks Authorized Bug Bounty Testing#105287
  39. Opus 5.5 safeguards incorrectly flag non-cybersecurity session#105246
  40. Session incorrectly flags game anti-cheat implementation as cyber#105214
  41. Claude Code refuses PHP bug fixes on client site, falsely flagged as cyber misuse#105172
  42. ok [Bug] Opus 5.5 refuses authorized security testing of own apps#105127
  43. Cyber safeguard false positive blocks defensive testing of own code#105116
  44. Fable Cyber requests downmodeled despite CVP enrollment#105109
  45. Overly Restrictive Refusal for Benign Stack Overflow Cybersecurity Video Request#105069
  46. Security flags incorrectly triggered for historical/educational ROM analysis#105068
  47. False Positive Cyber Safeguard Blocks Python-to-C++ Build Refactor#105067
  48. Safeguards Block Legitimate Defensive Security Work#105027
  49. Bug Report: Security audit of libsmb2 blocked by safeguard issues#104962
  50. Opus 5.5 over-restrictive safety blocking makes model unusable for basic local tasks#104948
  51. Business question incorrectly flagged as cyber-related#104934
  52. Claude Code blocks legitimate database hardening request in internal dev environment#104834
  53. Safety classifier repeatedly stops responses during authorized smart-contract bug bounty work#104787
  54. API Error: Sonnet 5 Cyber Safeguards Flag Legitimate Request#104782
  55. Opus 5.5 cyber safeguards flag legitimate access-control device integration work#104705
  56. API Error: Opus 4.8 Safeguards Flag Academic Cybersecurity Research (BladeRunner FYP)#104701
  57. Security Audit Fix Request Flagged for Cyber Misuse and Downgraded to Model 4.8#104699
  58. Claude Code Refuses Legitimate Security Research on a library#104673
  59. Unexpected downgrade to Opus 4.8 when discussing Anthropic cybersecurity reports#104656
  60. Over-refusal of legitimate defensive security work#104638
  61. Opus 4.5 safeguards flag unreleased project as cyber security risk#104571
  62. Cyber safeguards block legitimate security audit work on Opus 5.5/4.8#104448
  63. Opus 4.5 safeguards incorrectly flag benign session#104396
  64. Opus 5.5 safeguards falsely flag non-cyber request as [cyber]#104368
  65. Opus 5.5 Flags Defensive Security Work as Unauthorized Pentest#104351
  66. Sessions Repeatedly Flagged as Cyber During Fuzzing Framework Work#104244
  67. Sonnet 5.5 safeguards falsely flag non-cyber request with [cyber] block#104185
  68. Cyber Safeguards Block Explanation of Vulnerability Impact#104126
  69. Opus flags legitimate Cloudflare security skill as malicious#104078
  70. Real-time cyber safeguard false-positives on defensive audit work, blocking subagents#104073
  71. Opus flags benign cybersecurity-adjacent conversations as policy violations#103997
  72. Safeguards Force Fallback to Opus 4.8 During Defensive CVE Mitigation Testing#103958
  73. Safety classifier stops defensive work on own code without reason#103937
  74. Cyber Safeguard False Positive Interrupts Agent Integration Work#103933
  75. Safety classifier falsely blocks defensive security review of user's own repo#103932
  76. Opus 5.5 falsely flags legitimate connection audit task as misuse#103903
  77. False positive safety refusal during legitimate internal fuzz testing#103770
  78. Frequent Fable 5.1 Safeguard False Positives on Non-Security Projects#103768
  79. Over-Triggered Cyber Safety Warning on Benign Requests (Camera Setup, Underwater Bubble Detection)#103740
  80. Anthropic API Error: Real-Time Cyber Safeguards Flag Legitimate Firewall Config Request#103660
  81. False Positive Cybersecurity Flag Triggered Without Cause#103611
  82. False Positive AUP Cyber Safeguard Block on Defensive Pentest Work#103571
  83. Overly Strict Safeguards Block Malware Analysis on WordPress Exploit#103545
  84. Unexpected Cyber Safety Flag on Minimal Input#103520
  85. Cyber-safety classifier flags CVP-approved user on security research work#103486
  86. CONSTANT Security violations/stoppage - this is RIDICULOUS#103472
  87. Bug Report: Claude Code blocks actions during approved internal security review#103354
  88. Anthropic API Error: Opus 4.8 Cyber Safeguards Flagging Legitimate Request#103337
  89. Bug Report: Claude Code blocks legitimate CTF security work#103210
  90. A message was flagged by the Opus 5.5 safeguards, and the session switched to Opus#103120

Do you work on Claude Code?

Claim its RAP Sheet to see what your users report and answer them in one place.

Claim a tool