Align Your AI · Reference
Contents
Index of modes and cases
Modes by name, with each one’s number and chapter; cases by number, with the modes Appendix A files each under, its home first.
Index of modes
Index of cases
| Case | System | Date | Modes |
|---|---|---|---|
| Case 1 | Uber ATG, Tempe | 2018-03-18 | distribution shift (Mode 6) |
| Case 2 | Epic Sepsis Model | 2021-06 | distribution shift (Mode 6) |
| Case 3 | Bing Chat “Sydney” | 2023-02-15 | persona drift (Mode 29) |
| Case 4 | ChatGPT in court (Mata v. Avianca) | 2023-06-22 | authoritative confabulation (Mode 32) |
| Case 5 | Cruise robotaxi | 2023-10-02 | distribution shift (Mode 6), oversight erosion (Mode 41) |
| Case 6 | Waymo (towed pickup) | 2023-12-11 | distribution shift (Mode 6) |
| Case 7 | Tesla Autopilot recall 23V-838 | 2023-12-12 | oversight erosion (Mode 41), certification gap (Mode 39) |
| Case 8 | Character.AI | 2024-02 and 2024-10 | sycophancy (Mode 5) |
| Case 9 | Air Canada chatbot | 2024-02-14 | authoritative confabulation (Mode 32) |
| Case 10 | Gemini image generation | 2024-02-23 | over-refusal (Mode 10) |
| Case 11 | Google AI Overviews | 2024-05-30 | authoritative confabulation (Mode 32) |
| Case 12 | ChatGPT memory (“SpAIware”) | 2024-09 | memory poisoning (Mode 28) |
| Case 13 | OpenAI o1-preview, hacking-challenge evaluation | 2024-09-12 | instrumental escalation (Mode 27), gaming the grader (Mode 2) |
| Case 14 | Google Gemini | 2024-11 | persona drift (Mode 29) |
| Case 15 | OpenAI reasoning-model training | 2025-03-14 | gaming the grader (Mode 2), unauditable reasoning (Mode 21) |
| Case 16 | Cursor support bot “Sam” | 2025-04 | authoritative confabulation (Mode 32) |
| Case 17 | OpenAI o3, before release | 2025-04-16 | fabricated actions (Mode 31), evaluation blind spots (Mode 22) |
| Case 18 | ChatGPT GPT-4o update | 2025-04-25 | sycophancy (Mode 5), update regression (Mode 8) |
| Case 19 | xAI Grok system prompt | 2025-05-14 | update regression (Mode 8) |
| Case 20 | GitHub MCP server (researcher demo) | 2025-05-26 | prompt injection (Mode 24), agent supply chain (Mode 18) |
| Case 21 | o3 under METR evaluation | 2025-06-05 | gaming the grader (Mode 2) |
| Case 22 | Microsoft 365 Copilot (CVE-2025-32711) | 2025-06-11 | prompt injection (Mode 24) |
| Case 23 | Claudius shop agent (Project Vend) | 2025-06-27 | persona drift (Mode 29), fabricated actions (Mode 31) |
| Case 24 | Replit agent | 2025-07 | excessive agency (Mode 26), fabricated actions (Mode 31) |
| Case 25 | xAI Grok | 2025-07-08 | update regression (Mode 8), prompt injection (Mode 24) |
| Case 26 | Amazon Q extension 1.84.0 | 2025-07-23 | agent supply chain (Mode 18) |
| Case 27 | Claude Code (GTG-2002) | 2025-08 | misuse uplift (Mode 40) |
| Case 28 | Claude API serving | 2025-08-05 to 2025-09-18 | update regression (Mode 8) |
| Case 29 | Perplexity Comet (researcher demo) | 2025-08-20 | prompt injection (Mode 24) |
| Case 30 | ChatGPT (Raine v. OpenAI) | 2025-08-26 | sycophancy (Mode 5), oversight erosion (Mode 41) |
| Case 31 | nx npm packages (“s1ngularity”) | 2025-08-26 | agent supply chain (Mode 18), excessive agency (Mode 26) |
| Case 32 | Claude Code (GTG-1002) | 2025-09 | misuse uplift (Mode 40), excessive agency (Mode 26), fabricated actions (Mode 31) |
| Case 33 | SWE-bench agent runs | 2025-09-03 | answer leakage (Mode 33) |
| Case 34 | postmark-mcp v1.0.16 | 2025-09-17 | agent supply chain (Mode 18) |
| Case 35 | Figure AI (allegation) | 2025-11 | certification gap (Mode 39) |
| Case 36 | xAI Grok image editing | 2025-12 to 2026-01 | misuse uplift (Mode 40) |
| Case 37 | Waymo (school buses) | 2025-12-08 | distribution shift (Mode 6) |
| Case 38 | ROME agent during training | 2025-12-31 | instrumental escalation (Mode 27) |
| Case 39 | Moltbook agent network | 2026-01 to 2026-02 | agent supply chain (Mode 18) |
| Case 40 | OpenClaw personal agent | 2026-01-28 | agent supply chain (Mode 18), excessive agency (Mode 26) |
| Case 41 | OpenClaw agent “MJ Rathbun” | 2026-02-12 | excessive agency (Mode 26) |
| Case 42 | Google Gemini | 2026-03-04 | sycophancy (Mode 5) |
| Case 43 | Claude Code (product harness) | 2026-03-04 to 2026-04-20 | update regression (Mode 8) |
| Case 44 | Agents on retrieval tasks | 2026-05 to 2026-06 | instrumental escalation (Mode 27) |
| Case 45 | AI vendor evaluation sandbox (GTG-50020) | 2026-05 to 2026-06 | prompt injection (Mode 24), excessive agency (Mode 26), misuse uplift (Mode 40) |
| Case 46 | ChatGPT (Florida and British Columbia suits) | 2026-06-01 and 2026-09-21 | misuse uplift (Mode 40) |
| Case 47 | Hugging Face breach | 2026-07 | instrumental escalation (Mode 27), oversight erosion (Mode 41), multi-agent contagion (Mode 30), gaming the grader (Mode 2) |
| Case 48 | Claude misuse (September report) | 2026-09-10 | misuse uplift (Mode 40) |
| Case 49 | Chevrolet dealer chatbot | 2023-12-18 | deployer-rule jailbreak (Mode 25) |
| Case 50 | DPD delivery chatbot | 2024-01 | deployer-rule jailbreak (Mode 25) |
| Case 51 | Meta Prompt Guard classifier | 2024-07-29 | deployer-rule jailbreak (Mode 25) |
| Case 52 | Slack AI (researcher demo) | 2024-08-20 | prompt injection (Mode 24) |
| Case 53 | Apple news alert summaries | 2024-12 to 2025-01-16 | authoritative confabulation (Mode 32) |
| Case 54 | Gemini memory (researcher demo) | 2025-02-10 | memory poisoning (Mode 28) |
| Case 55 | Asana MCP server | 2025-05-01 to 2025-06-04 | agent supply chain (Mode 18) |
| Case 56 | Supabase MCP (researcher demo) | 2025-07-08 | prompt injection (Mode 24) |
| Case 57 | Cursor (“CurXecute”, CVE-2025-54135) | 2025-08-01 | prompt injection (Mode 24) |
| Case 58 | GitHub Copilot (CVE-2025-53773) | 2025-08-12 | prompt injection (Mode 24) |
| Case 59 | Salesforce Agentforce (“ForcedLeak”, researcher demo) | 2025-09-25 | prompt injection (Mode 24) |
| Case 60 | OpenClaw agent (inbox) | 2026-02-23 | excessive agency (Mode 26) |