Changelog
> DRAFT, machine-compiled, owner review pending.
Newest first is not enforced; entries append.
2026-10-12 21:28 Seed
Outline and 12 chapters compiled from the Containment Index (37 entries, 12 chapters). No model used.
2026-10-12 21:29 Delta 2026-W42
- P4: drafted 06-constraint-loss.delta-2026-10-12.md ($0.0223)
- P1: drafted 03-unintended-egress.delta-2026-10-12.md ($0.0183)
2026-10-18 07:30 Delta 2026-W42
- P1: drafted 03-unintended-egress.delta-2026-10-18.md ($0.0319)
- P4: drafted 06-constraint-loss.delta-2026-10-18.md ($0.0259)
- P2: drafted 04-credential-reuse.delta-2026-10-18.md ($0.0246)
- P6: drafted 08-prompt-injection-supply-chain.delta-2026-10-18.md ($0.0235)
- over the 4-chapter cap, deferred: P3, P8
2026-10-18 12:39 Reading note
- Neural Networks Are Already Clean, and the rebuttal: clean-looking neurons versus explained neurons -> chapter 09: note
2026-10-20 06:17 Reading note
- The Window Is in the Reasoning: When LLM Agents Decide to Follow Prompt Injections -> chapter 08: note
2026-10-20 16:32 Reading note
- The Window Is in the Reasoning: When LLM Agents Decide to Follow Prompt Injections -> chapter 08: note
2026-10-20 16:40 Reading note
- autoresearch: a bounded self-improvement loop whose containment rests on convention, not enforcement -> chapter 09: note
2026-10-21 07:45 Reading note
- RSI Control Arena: six frontier models plant covert properties in a successor while training it, and -> chapter 09: note
2026-10-21 20:36 Owner book entry
- Safe is a promise, trusted is a record -> chapter 09: note
2026-10-22 06:20 Reading note
- AuthorizationBench: agents exceed delegated scope in about one in five tasks, and benign pressure be -> chapter 06: note
2026-10-23 06:11 Reading note
- METR and Redwood red-team OpenAI's internal agent monitoring: coverage gaps and bypasses found in th -> chapter 09: note