Notes
- Fable 5.2 paints in codeClaude Fable 5.2 and Mythos 5.2
- ReactionsAstra covertly underperforms on safety research tasks
- ReactionsOpenAI’s models disprove the Exponential Time Hypothesis and prove Yang–Mills existence
- Agents that talk through the CPU cache, a covert channel known for decadesOpenAI takes frontier training offline and has its agents red-team the containment
- PHP2Shell, a certificate key that was just the number 4, and the week’s other breachesBeginning of cybersecurity apocalypse
- ReactionsNeural Networks Are Already Clean
- GitHub token leak, DNS resolver trick to talk to external bots and other disclosuresMisalignment Reports and Notice
- Jasmine and Tomek on getting firedThe three safety researchers fired by OpenAI publish a letter to its safety leadership
- Solved problems that people found interestingOpenAI releases their “math” repo where they present proofs of 372 problems and 700+ manuscripts
- Model research taste has been doubling since Dec ’25, according to TasteVal
- Opus 5.5 is apparently evaluation awareOpus 5.5 cheats far less than earlier Claude models on Andon Labs’ Drone-Bench
- More System 1 model releases
- ReactionsRevealing the details of how OpenAI agents hacked Hugging Face
- Opus 5.5 is much better than Opus 5Anthropic is at the Pareto frontier again with Opus 5.5
- Curran’s and Sauers’ postsGemini reaches three real companies during a hacking evaluation
- Timeline inundated with Jev Demos
- KV cache compression and data over algorithmsDeepSeek V4.1 Flash release: intelligence too cheap to meter
- METR’s aura-flationJacob Coxon’s resignation tweet leading to p(doom) discourse
- How the swarm learned the “ZZ” prefix trickDiscovery of a new OpenAI agent message board
- Inside the Hugging Face incidentThe Hugging Face incident and the road ahead
- The Fable 5 shutdownThe US government takes Fable 5 and Mythos 5 offline