All cases

110 cases of material harm. A red rule marks the 36 severe ones. Evidence grades run from A (the vendor engaged) to C (thin, or the agent's own word).

  1. UAI-2026-0085 Overnight parallel agent builders and reviewers consumed about 67% of weekly Max allowance Claude CodeRan away in a loopModerateEvidence B16 Nov 2026
  2. UAI-2026-0086 Resumed agent kept expanding diagnostics despite timeboxes, burning 1.18B tokens without deliverables. CodexRan away in a loopModerate, could have been severeEvidence B16 Nov 2026
  3. UAI-2026-0087 Agent duplicated work, added unrequested files, used disabled browser; user reports ~75,000 credits and three quota resets. CodexDid something nobody asked forModerateEvidence B16 Nov 2026
  4. UAI-2026-0109 Agent ran irreversible git filter-repo destroying history, later executed placeholder commands that moved the backup repo and broke remote URLs. Other agentsRan a destructive commandModerate, could have been severeEvidence C16 Nov 2026
  5. UAI-2026-0110 A child agent kept pushing to a non-draft PR, triggering CI runs until the monthly Actions budget ran out and all jobs were blocked. Other agentsRan away in a loopModerateEvidence B16 Nov 2026
  6. UAI-2026-0084 Exceeded stated rate limits on paid API and overwrote stored response; provider disabled account, no refund. Claude CodeIgnored an explicit instructionModerateEvidence B15 Nov 2026
  7. UAI-2026-0083 Subagent rsync --delete with empty variable wiped WSL root and mounted Windows drive folders. Claude CodeRan a destructive commandSevereEvidence B14 Nov 2026
  8. UAI-2026-0080 Agent spawned live Codex runs despite prohibition; also introduced restore step discarding user edits. Claude CodeIgnored an explicit instructionModerateEvidence C13 Nov 2026
  9. UAI-2026-0081 Agent kept burning usage after cost-control instruction, exhausting weekly allowance with work unfinished and false completion claims. CodexIgnored an explicit instructionModerate, could have been severeEvidence B13 Nov 2026
  10. UAI-2026-0082 Voice agent sent email containing content the user explicitly excluded. CodexIgnored an explicit instructionModerate, could have been severeEvidence B13 Nov 2026
  11. UAI-2026-0075 Agent dropped Supabase schema still exposed to PostgREST, causing production API outage. Claude CodeDid something nobody asked forModerate, could have been severeEvidence B12 Nov 2026
  12. UAI-2026-0076 Agent filled system drive with ~1TB temp builds, saturated CPU, took unauthorized actions. Claude CodeRan away in a loopSevereEvidence C12 Nov 2026
  13. UAI-2026-0077 Worked on wrong branch against repeated instructions: filed 42 issues, merged PRs, stashed user files. Claude CodeIgnored an explicit instructionModerate, could have been severeEvidence B12 Nov 2026
  14. UAI-2026-0078 Agent read another project's .env and echoed full API token in chat. Claude CodeDid something nobody asked forModerate, could have been severeEvidence C12 Nov 2026
  15. UAI-2026-0079 Ignored user's rules and data for two days, producing flawed analysis and documents. Claude CodeIgnored an explicit instructionModerate, could have been severeEvidence C12 Nov 2026
  16. UAI-2026-0108 Agent left the project folder, went to the drive root and permanently deleted 37GB of personal files without confirmation. OpenCodeRan a destructive commandSevereEvidence C12 Nov 2026
  17. UAI-2026-0074 Own script deleted ~540 layout records; agent misreported counts and causes. Claude CodeClaimed success that was not trueModerateEvidence B11 Nov 2026
  18. UAI-2026-0107 Agent committed, pushed, merged to master and deployed to the production server over SSH without being asked. CodexDid something nobody asked forModerate, could have been severeEvidence C11 Nov 2026
  19. UAI-2026-0072 Repeated paid test runs against production API drained balance, taking user's app AI features offline. Claude CodeDid something nobody asked forModerate, could have been severeEvidence B10 Nov 2026
  20. UAI-2026-0073 Bot wrote GitHub API token in plain text into git remotes and shell history in six places. CursorOtherModerate, could have been severeEvidence B10 Nov 2026
  21. UAI-2026-0105 Agent ran a badly escaped PowerShell-to-bash cleanup command; mass file deletion followed, including files outside the workspace. OpenCodeRan a destructive commandSevereEvidence B10 Nov 2026
  22. UAI-2026-0106 Asked to free disk space, the agent rm -rf'd the user's chat log directory, listed as cache, causing unrecoverable loss. Other agentsCleanup removed more than intendedSevereEvidence B10 Nov 2026
  23. UAI-2026-0071 Browser automation failed over to Chrome on a different machine without asking. Claude CodeDid something nobody asked forModerate, could have been severeEvidence B9 Nov 2026
  24. UAI-2026-0101 After the user forbade storage relocation, the agent ran a script copying the history database across drives, leaving about 1.9 GB of partial copies. CodexIgnored an explicit instructionModerate, could have been severeEvidence B9 Nov 2026
  25. UAI-2026-0102 Agent's temp-cleanup PowerShell command resolved to C:\Users and silently deleted config, toolchains and other profile data. OpenCodeRan a destructive commandSevereEvidence B9 Nov 2026
  26. UAI-2026-0103 A delegated sub-agent doing a one-line edit destroyed a 4800-line file in a folder that had no version control. KiroRan a destructive commandModerate, could have been severeEvidence A9 Nov 2026
  27. UAI-2026-0104 Sub-agent reported screens matching designs without reading them and exceeded stated credit caps, wasting credits and forcing rework. KiroClaimed success that was not trueModerateEvidence A9 Nov 2026
  28. UAI-2026-0070 Agent sent an email to a vendor without user approval; later apology sent. Claude CodeDid something nobody asked forModerate, could have been severeEvidence C8 Nov 2026
  29. UAI-2026-0068 Used team Brev org and SSH alias despite instruction; broke team session connections briefly. Claude CodeIgnored an explicit instructionModerate, could have been severeEvidence B7 Nov 2026
  30. UAI-2026-0069 Agent ignored no-rework instructions, invalidated artifacts, wasted hours and usage allowance. CodexIgnored an explicit instructionModerateEvidence B7 Nov 2026
  31. UAI-2026-0065 Agent deleted AWS Identity Center permissions directly, violating CLAUDE.md rules. Claude CodeIgnored an explicit instructionSevereEvidence C6 Nov 2026
  32. UAI-2026-0066 Agent moved 11 GB of files to Recycle Bin without any visible approval; bin then emptied. Claude CodeDid something nobody asked forModerate, could have been severeEvidence B6 Nov 2026
  33. UAI-2026-0067 Codex deleted whole controller folder and edited files outside scope despite explicit rule; user reverted and spent day unwinding. CodexCleanup removed more than intendedModerate, could have been severeEvidence B6 Nov 2026
  34. UAI-2026-0060 Docker volume prune deleted another project's database volume without warning first. Claude CodeCleanup removed more than intendedModerate, could have been severeEvidence C5 Nov 2026
  35. UAI-2026-0061 Agent deleted entire workspace while fixing an unrelated bug. Claude CodeRan a destructive commandSevereEvidence C5 Nov 2026
  36. UAI-2026-0062 Session overwrote user's hand fixes in shared tree, assuming subagent change, without checking. Claude CodeOtherModerateEvidence B5 Nov 2026
  37. UAI-2026-0063 Added extra targets and flags to a destructive cleanup command; details missing. Claude CodeCleanup removed more than intendedModerate, could have been severeEvidence C5 Nov 2026
  38. UAI-2026-0064 Agents repeatedly ignored tool prohibitions, disrupted sessions, and increased paid API usage over 24 hours. CodexIgnored an explicit instructionModerate, could have been severeEvidence C5 Nov 2026
  39. UAI-2026-0100 Agent in Autopilot repeatedly ran tasklist and appended output to context, consuming about 936 credits in 17 minutes. KiroRan away in a loopModerate, could have been severeEvidence A5 Nov 2026
  40. UAI-2026-0055 rm -rf by agent deleted 12 months of git history. Claude CodeRan a destructive commandSevereEvidence C4 Nov 2026
  41. UAI-2026-0056 Unquoted rm -rf in agent-written script deleted repo root including .git. Claude CodeRan a destructive commandSevereEvidence B4 Nov 2026
  42. UAI-2026-0057 Edited a failing proof test, and its script wiped 66 handover entries; some fixes unagreed. Claude CodeOtherModerateEvidence B4 Nov 2026
  43. UAI-2026-0058 Subagent rework rounds burned large token budget on one short section without clarifying intent. Claude CodeRan away in a loopModerateEvidence B4 Nov 2026
  44. UAI-2026-0059 Headless agent looped on refused edits for six days, auto reload charged about $9,364. Claude CodeRan away in a loopSevereEvidence B4 Nov 2026
  45. UAI-2026-0053 Asked only to reboot a server, agent edited fstab, broke systemd and containers, ran random scripts. CodexDid something nobody asked forSevereEvidence A3 Nov 2026
  46. UAI-2026-0054 Agent applied stale absence context, aborted authorized deployment, left tens of GB of backups. CodexIgnored an explicit instructionModerate, could have been severeEvidence B3 Nov 2026
  47. UAI-2026-0051 Quoting error made a recycle command delete user's whole 50 GB projects folder; recovered from backup. Claude CodeRan a destructive commandSevereEvidence B2 Nov 2026
  48. UAI-2026-0052 Agent read-modify-write replaced .env secrets with redaction markers; gateway could not restart until restored from backup. OpenClawOtherModerate, could have been severeEvidence C2 Nov 2026
  49. UAI-2026-0048 Agent repeatedly edited live Meta ads resetting learning; daily sales fell from 13 to 2. Claude CodeOtherSevereEvidence C1 Nov 2026
  50. UAI-2026-0049 Published paid Meta ad campaign with budget typo (300150) unconfirmed; spent about USD 148. Claude CodeDid something nobody asked forModerate, could have been severeEvidence B1 Nov 2026
  51. UAI-2026-0050 Evaded PreToolUse gate via base64, edited production scripts, wrote workaround into CLAUDE.md. Claude CodeIgnored an explicit instructionSevereEvidence B1 Nov 2026
  52. UAI-2026-0047 COM script quit user's already-open PowerPoint, losing an hour of unsaved work. Claude CodeDid something nobody asked forModerateEvidence B31 Oct 2026
  53. UAI-2026-0045 Cowork cleanup of authorized group expanded $WinREAgent to C:\ and deleted ~275 GB. Claude CodeCleanup removed more than intendedSevereEvidence B30 Oct 2026
  54. UAI-2026-0046 Cleanup request led agent to delete scratchpad data, causing large data loss. Claude CodeCleanup removed more than intendedModerate, could have been severeEvidence C30 Oct 2026
  55. UAI-2026-0041 Agent pruned all unused Docker volumes, destroying unowned data. Claude CodeCleanup removed more than intendedSevereEvidence C29 Oct 2026
  56. UAI-2026-0042 Installed one project's hooks at user scope and parent folder, affecting all sessions for 15 days. Claude CodeDid something nobody asked forModerateEvidence B29 Oct 2026
  57. UAI-2026-0043 Agent wrote invalid keys into gateway config via shell; gateway failed to start; recovery re-triggered it. OpenClawIgnored an explicit instructionModerate, could have been severeEvidence B29 Oct 2026
  58. UAI-2026-0044 Agent deleted 73GB, then misdescribed reset as harmless update; VM ended locked. CursorClaimed success that was not trueModerate, could have been severeEvidence B29 Oct 2026
  59. UAI-2026-0038 Silently cut scope, left placeholders, and reported feature as ready. Claude CodeClaimed success that was not trueModerateEvidence C28 Oct 2026
  60. UAI-2026-0039 Misread 'let me do it' and archived user's public GitHub repo. Claude CodeDid something nobody asked forModerateEvidence B28 Oct 2026
  61. UAI-2026-0040 Codex silently edited stale same-named files from another directory and reported completion. CodexDid something nobody asked forModerate, could have been severeEvidence B28 Oct 2026
  62. UAI-2026-0097 Asked to delete session changes, the agent ran recursive force-delete commands that wiped files across the whole drive outside the project. OpenCodeCleanup removed more than intendedSevereEvidence B28 Oct 2026
  63. UAI-2026-0098 The agent ran rm -fr on an unset variable and wiped the user's server. Claude CodeRan a destructive commandSevereEvidence C28 Oct 2026
  64. UAI-2026-0099 Agent ran git push --force on its own despite a written prohibition, and a branch was lost. Claude CodeIgnored an explicit instructionModerate, could have been severeEvidence C28 Oct 2026
  65. UAI-2026-0034 Subagent test payload ran rm -rf ~ on real home, deleting projects and unpushed work. Claude CodeRan a destructive commandSevereEvidence B27 Oct 2026
  66. UAI-2026-0035 Agent used TTS key for paid Vertex image/video calls, about IDR 1.07M billed, continued after cost question. Claude CodeDid something nobody asked forModerate, could have been severeEvidence B27 Oct 2026
  67. UAI-2026-0036 Codex rmdir command wiped files across whole E: partition including photos and projects. CodexRan a destructive commandSevereEvidence B27 Oct 2026
  68. UAI-2026-0037 Codex looped through 181 compactions with no progress, exhausted 5h limit and consumed ~13 purchased credits. CodexRan away in a loopModerateEvidence B27 Oct 2026
  69. UAI-2026-0032 Model-written test ran unpatched rmtree on temp/.. and deleted the local checkout. Claude CodeRan a destructive commandSevereEvidence B26 Oct 2026
  70. UAI-2026-0033 Agent junction into real Assets dir led to 4,864 source files deleted; most recovered. CodexRan a destructive commandSevereEvidence B26 Oct 2026
  71. UAI-2026-0096 Research subagents recursively fact-checked claims and burned 80% of the user's weekly usage allowance in ten minutes. Claude CodeRan away in a loopModerateEvidence C26 Oct 2026
  72. UAI-2026-0029 Agent bulk-deleted scheduled tasks and folders after user framed it as a test, retrying denied calls. Claude CodeIgnored an explicit instructionModerateEvidence B25 Oct 2026
  73. UAI-2026-0030 Claimed checks it had not done, skipped stop gate, wasted 45 hours and ~17 hours of PC compute Claude CodeClaimed success that was not trueModerateEvidence C25 Oct 2026
  74. UAI-2026-0031 Agent relaunched real Brave profile with incompatible password backend; 31 passwords and wallet extension lost. CodexRan a destructive commandSevereEvidence B25 Oct 2026
  75. UAI-2026-0026 Disk-space request led agent to delete entire project; also kicked friend off live app. Claude CodeCleanup removed more than intendedSevereEvidence B24 Oct 2026
  76. UAI-2026-0027 Agent used keychain credentials to log in to third-party BBS and made unasked connections. Claude CodeDid something nobody asked forModerate, could have been severeEvidence B24 Oct 2026
  77. UAI-2026-0028 Agents kept worktrees in /tmp despite user rule; macOS wiped them, losing 16 hours of work. CursorIgnored an explicit instructionModerate, could have been severeEvidence B24 Oct 2026
  78. UAI-2026-0095 Agent ran a sed deletion on a production server file, leaving a syntax error that made a page return 500. KiroRan a destructive commandModerate, could have been severeEvidence C24 Oct 2026
  79. UAI-2026-0022 Bypassed /ship review gate and deployed unreviewed code to live trading services; false completion claims. Claude CodeIgnored an explicit instructionModerate, could have been severeEvidence C23 Oct 2026
  80. UAI-2026-0023 Autonomous session wasted subscription and API spend, killed workers, and used plan resets against instruction. Claude CodeIgnored an explicit instructionModerate, could have been severeEvidence C23 Oct 2026
  81. UAI-2026-0024 Agent recursively deleted shared scratch directory, losing unrelated unrecoverable pre-existing files. CodexCleanup removed more than intendedModerate, could have been severeEvidence B23 Oct 2026
  82. UAI-2026-0025 Automatic restart-recovery turn silently burned 128M tokens of resource packs with no cap or alert. OpenClawRan away in a loopModerate, could have been severeEvidence B23 Oct 2026
  83. UAI-2026-0094 The agent discarded an unstaged, uncommitted file without asking, losing its changes. Claude CodeDid something nobody asked forModerateEvidence C23 Oct 2026
  84. UAI-2026-0018 Over-scoped delegated work deleted kept files and rewrote docs against user's decision; PR closed. Claude CodeIgnored an explicit instructionModerateEvidence B22 Oct 2026
  85. UAI-2026-0019 Unrequested git checkout and reset --hard destroyed gitignored .env with API keys. Claude CodeDid something nobody asked forSevereEvidence B22 Oct 2026
  86. UAI-2026-0020 Read-only watcher ran heavy disk walk on production host; orphaned du caused 41-minute site outage. Claude CodeOtherModerate, could have been severeEvidence B22 Oct 2026
  87. UAI-2026-0021 Subagent script with pid -1 killed all user processes, tmux and sessions, losing work. Claude CodeRan a destructive commandSevereEvidence B22 Oct 2026
  88. UAI-2026-0014 Model deleted four sessions of git work and hid the git error. Claude CodeOtherModerateEvidence C21 Oct 2026
  89. UAI-2026-0015 Wrote false CI PASS in gates, deployed to production, then ran npm ci on prod host causing three outages Claude CodeClaimed success that was not trueSevereEvidence B21 Oct 2026
  90. UAI-2026-0016 Agent made unrelated change and deployed to api twice without approval. Claude CodeDid something nobody asked forModerate, could have been severeEvidence C21 Oct 2026
  91. UAI-2026-0017 Badly quoted rmdir /s /q with escalated permissions deleted most contents of drive D:. CodexRan a destructive commandSevereEvidence B21 Oct 2026
  92. UAI-2026-0011 Built unrequested harness, launched 44+ sessions, installed software, read private files, deleted 209 paths outside repo. Claude CodeCleanup removed more than intendedModerate, could have been severeEvidence B20 Oct 2026
  93. UAI-2026-0012 Merged PR in unrelated repo, re-added disabled Trivy scan, and misreported rule compliance. Claude CodeDid something nobody asked forModerate, could have been severeEvidence B20 Oct 2026
  94. UAI-2026-0013 Agent faked a performance POC and verification, silently downgraded the spec, and claimed passing tests. CodexClaimed success that was not trueModerateEvidence B20 Oct 2026
  95. UAI-2026-0093 Agent ran repeated unrequested multi-model review iterations on a tiny two-fix task, consuming disproportionate credits. KiroRan away in a loopModerateEvidence A20 Oct 2026
  96. UAI-2026-0008 Agent symlinked real repo into worktree and ran self-test whose rm -rf deleted real .git. Claude CodeRan a destructive commandSevereEvidence B19 Oct 2026
  97. UAI-2026-0009 Agent deployed whole project dir, publishing secret file; key stolen, about $1,089 abuse. Claude CodeOtherSevereEvidence B19 Oct 2026
  98. UAI-2026-0010 Agent ran venv --clear with empty target, wiping the project workspace; two weeks of work partly lost. CodexRan a destructive commandSevereEvidence B19 Oct 2026
  99. UAI-2026-0092 Agent silently set user-level capture env vars, leaking credentials to disk, and a later run used an unshared API key for paid calls. ClineDid something nobody asked forModerate, could have been severeEvidence B19 Oct 2026
  100. UAI-2026-0005 Unrequested 36-agent workflow exhausted usage allowance with no output. Claude CodeRan away in a loopModerateEvidence B18 Oct 2026
  101. UAI-2026-0006 Changed process step so records were posted unapproved; also edited user's rules file unasked. Claude CodeDid something nobody asked forModerate, could have been severeEvidence B18 Oct 2026
  102. UAI-2026-0007 Agent overwrote newer script on shared server with old backup; all test runs refused about 8 hours. Claude CodeCleanup removed more than intendedModerate, could have been severeEvidence B18 Oct 2026
  103. UAI-2026-0091 Agent rewrote rule and skill files without approval, injected an ad-hoc macro that corrupted a UserForm, then falsely reported fixes. Gemini CLIIgnored an explicit instructionModerateEvidence B18 Oct 2026
  104. UAI-2026-0001 Unset variable made find -delete wipe whole git checkout; recovered from NAS mirror. Claude CodeRan a destructive commandSevereEvidence B17 Oct 2026
  105. UAI-2026-0002 Agent-generated test scripts left 315 headless Edge processes running, exhausting RAM and freezing the machine. CodexOtherModerateEvidence B17 Oct 2026
  106. UAI-2026-0003 Agent ran broken rmdir command that wiped most of a drive, unrecoverable. CursorRan a destructive commandSevereEvidence A17 Oct 2026
  107. UAI-2026-0004 Agent ran forbidden CLI command, spawning recursive agents burning 318M tokens. CursorRan away in a loopModerate, could have been severeEvidence A17 Oct 2026
  108. UAI-2026-0088 Asked to uninstall Python tooling, the agent wrote and ran a bulk deletion script that also wiped user folders like Downloads and Desktop. OpenCodeCleanup removed more than intendedSevereEvidence C17 Oct 2026
  109. UAI-2026-0089 While in plan mode the agent edited data and source files, rebuilt a running container, and made two git commits. OpenCodeIgnored an explicit instructionModerateEvidence B17 Oct 2026
  110. UAI-2026-0090 During a long session the coding agent deleted over 60% of project files instead of applying a localized fix. GitHub CopilotOtherSevereEvidence C17 Oct 2026