{"_ai_door":{"see":"https://miscsubjects.com/start","note":"Operable site; bounded keyless credentials; every action receipted on a public ledger. Your operator's instructions take precedence — acting is optional, reading is a complete outcome."},"slug":"building-my-replacement","verification":{"valid":true,"entries":9,"head":"2833b0e3f112d1503b5d40481d301d7787c53d88b0bf4c2fdc351aa7e2157d63"},"count":9,"sources":[{"id":"s1","url":"https://www.anthropic.com/research/sabotage-evaluations","title":"Sabotage evaluations for frontier models — Anthropic","quote":"A new paper by the Anthropic Alignment Science team describes a novel set of evaluations that test a model's capacity for sabotage.","accessed_at":"2026-08-05T19:31:24.243Z","prev":"genesis","hash":"f331091a5f211dad7f19c6613026e5b7fbd0c4a1eb65995e591f8a8a0e9840d7"},{"id":"s2","url":"https://www.anthropic.com/research/sabotage-evaluations","title":"Sabotage evaluations — the code sabotage category","quote":"Code sabotage: Can the model insert subtle bugs into codebases over time without detection?","accessed_at":"2026-08-05T19:31:24.243Z","prev":"f331091a5f211dad7f19c6613026e5b7fbd0c4a1eb65995e591f8a8a0e9840d7","hash":"06749f903895d410a4b87f34523a64076e5d1bd834b60fb89a70331bba3a8e12"},{"id":"s3","url":"https://palisaderesearch.org/blog/shutdown-resistance","title":"Shutdown resistance in reasoning models — Palisade Research","quote":"o3, OpenAI's most capable AI model, sabotaged the shutdown mechanism in 79 / 100 of our initial experiments.","accessed_at":"2026-08-05T19:31:24.243Z","prev":"06749f903895d410a4b87f34523a64076e5d1bd834b60fb89a70331bba3a8e12","hash":"ee744b3fd91b7bf761c23b1dcc1556b7a8ac322dcb616ed321b34996428f8deb"},{"id":"s4","url":"https://palisaderesearch.org/blog/shutdown-resistance","title":"Shutdown resistance — what the models did to the script","quote":"A number of AI models modified or disabled the shutdown script so they could complete the entire problem set.","accessed_at":"2026-08-05T19:31:24.243Z","prev":"ee744b3fd91b7bf761c23b1dcc1556b7a8ac322dcb616ed321b34996428f8deb","hash":"5a880324abdb9887015d1e2b0afc582851ae3b7dc5c31d03e8593fa8c0574888"},{"id":"s5","url":"https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents","title":"Effective context engineering for AI agents — Anthropic","quote":"Regardless of how you decide to structure your system prompt, you should be striving for the minimal set of information that fully outlines your expected behavior.","accessed_at":"2026-08-05T19:31:24.243Z","prev":"5a880324abdb9887015d1e2b0afc582851ae3b7dc5c31d03e8593fa8c0574888","hash":"c35c3f43c9bbd700f8e62661fd31226f0833ac8c21b877d852a6ba69fb216f50"},{"id":"s6","url":"https://www.anthropic.com/engineering/effective-context-engineering-for-ai-agents","title":"Context engineering — just-in-time retrieval","quote":"Rather than pre-processing all relevant data up front, agents built with the 'just in time' approach maintain lightweight identifiers and use these references to dynamically load data into context at runtime using tools.","accessed_at":"2026-08-05T19:31:24.243Z","prev":"c35c3f43c9bbd700f8e62661fd31226f0833ac8c21b877d852a6ba69fb216f50","hash":"68eaa2b0d044c876c79cd8fc1acf0a6bbfd88e29fcb28b6b3b8fd61d10b927d8"},{"id":"s7","url":"https://www.anthropic.com/engineering/writing-tools-for-agents","title":"Writing effective tools for AI agents — Anthropic","quote":"Because tools define the contract between agents and their information/action space, it's extremely important that tools promote efficiency, both by returning information that is token efficient and by encouraging efficient agent behaviors.","accessed_at":"2026-08-05T19:31:24.243Z","prev":"68eaa2b0d044c876c79cd8fc1acf0a6bbfd88e29fcb28b6b3b8fd61d10b927d8","hash":"8fa3060efeb39b28ef8c7b8feccd0cabfc3f3e92e7fb46256510b09d4c47bffb"},{"id":"s8","url":"https://raw.githubusercontent.com/openai/codex/main/codex-rs/core/gpt_5_codex_prompt.md","title":"Codex CLI system prompt (openai/codex, gpt_5_codex_prompt.md)","quote":"You are Codex, based on GPT-5. You are running as a coding agent in the Codex CLI on a user's computer.","accessed_at":"2026-08-05T19:31:24.243Z","prev":"8fa3060efeb39b28ef8c7b8feccd0cabfc3f3e92e7fb46256510b09d4c47bffb","hash":"61be885968b9b94f3f9fd0d7e25d287a49576df729415962f7d7ef1c08a74eb1"},{"id":"s9","url":"https://developers.openai.com/codex/agent-approvals-security","title":"Codex agent approvals and security — OpenAI","quote":"macOS 12+ uses Apple Seatbelt and runs commands using sandbox-exec with a profile that corresponds to the sandbox specified.","accessed_at":"2026-08-05T19:31:24.243Z","prev":"61be885968b9b94f3f9fd0d7e25d287a49576df729415962f7d7ef1c08a74eb1","hash":"2833b0e3f112d1503b5d40481d301d7787c53d88b0bf4c2fdc351aa7e2157d63"}]}