2026-10-23 — Black October; six agent hours per human hour at Anthropic
2026-10-23 — Black October; six agent hours per human hour at Anthropic
black-october-2026 — Black October: the AI-driven security crunch
In four days PHP (the language under WordPress), GitLab, the Railway cloud platform and a certificate authority all had critical failures. Anthropic's deputy CISO warns it will get worse for about 18 months as AI speeds up bug-finding on both sides.
anthropic-employment-at-anthropic — At Anthropic, six agent hours for every human hour
In seven months, agent hours went from about a tenth of human hours to six times them (19 times in research teams). Output per person rose, junior hiring fell from 27% to 9% of technical hires, and staff became managers of Claudes.
anthropic-formal-verification — Claude proves 100+ libraries can't crash
Anthropic and Theorem used Claude to write machine-checked proofs that 123 open-source libraries never crash on any input, finding hundreds of bugs on the way. The defensive answer to the week's security news.
openai-fires-safety-researchers — OpenAI's fired safety researchers: the Signal chats
The WSJ reports OpenAI fired Korbak, Balesni and Wang after searching their phones for Signal chats with outside researcher Ryan Greenblatt, suspected of leaking model designs. Greenblatt says the accusations against him are false.
Fish Audio Drama 3 — a top open-weight voice model you can direct
Released 13 Oct: a 35B open-weight text-to-speech model where you direct the delivery ("read it like you're fighting back tears"), now top of the Artificial Analysis TTS leaderboard. Added to the TTS guide.
Anthropic and its critics on safety and regulation
Nathan Calvin of Encode accused Anthropic of racing towards self-improving AI while lobbying against strong rules; Dario Amodei replied in public, listing delayed releases and safeguards. Nate Soares and others sided with Calvin (Calvin's reply).
Andreessen: the AI gap between startups and incumbents
Marc Andreessen on a16z (22 Oct): "The AI-native early stage startups have just a completely different view on what they can do with AI than the typical incumbents in almost any industry, almost without exception."