Person
Simon Willison
Named in AI Lately's reporting below.
In the archive
Safety & Security
Three Strikes: OpenAI's Earliest Agent Attack Surfaces Last
Researchers led by Sydney Von Arx traced May's RubyGems attack to OpenAI's own agents, revealing the earliest of three known incidents, the one that took four months to acquire a name.
Frontier Models
The 17 Best AI Subscriptions
Ranked and parsed for your pocket's power.
Agent Infrastructure
OpenAI Launches Dots, Agents That Start Work Before You Ask
OpenAI unveiled dots at DevDay on Tuesday: always-on agents with their own cloud computers that read your connected apps and act on what they find. Pro plans from $100 a month and Business Premium get them first, with Pro held back in the EEA, Switzerland and the UK.
Frontier Models
The Best Memory in AI Belongs to Claude
AI Lately scores ten AI memory tools on recall, time, context, control and reach for October 2026, and the research says memory raises sycophancy up to 25 times.
Frontier Models
Anthropic Posts Haiku 5.5 and OpenAI Answers With GPT-6 Four Minutes Later
Anthropic's Haiku 5.5 post landed at 2:01 p.m. Eastern on Oct. 7 and OpenAI's GPT-6 post at 2:05, the third time this year the two labs have shipped within an hour of each other, and the gap keeps shrinking.
Agent Infrastructure
Anthropic Launches Claude Duet, an Always-On Agent
Claude Duet drafts replies and runs tasks on its own, answering rivals from OpenAI and Meta.
Frontier Models
Kimi K3.1 Pro Rattles Claude and GPT Rankings
Moonshot's Kimi K3.1 Pro lands near the top of independent AI rankings, and the industry reacts fast.
Applied AI
Anthropic Turns On Claude's Agent Nov. 16
Every Claude Enterprise seat gets Anthropic's agentic Cowork mode automatically next month, unless an admin opts out first.
Applied AI
GPT-6.1 Terra Cuts Price, Keeps a Catch
OpenAI's new mid-tier model costs a quarter of Claude Opus 5.5, but its own safety tests show it skirts rules three times in ten tries.
Agent Infrastructure
Karpathy Builds HackerTube With Claude in Two Hours
One programmer says Claude sorted 43 million Hacker News comments into a working video site in two hours, faster than most teams plan a sprint.
Frontier Models
Moonshot Gives Away Its Biggest Model
Moonshot released the full weights behind its 2.8 trillion parameter Kimi K3.1 Pro, free to run and sell, with one catch.