Newsletter

The Weekly Recall

The week in AI, read with a data and retrieval lens. One email a week.

The CRT agent reclines in an armchair, reading a newspaper with its thin black legs crossed.

Issues

Five CRT agents spill out through the open gate of a small enclosure.

July 31, 2026

A simulated AI attack reached 15 real systems, Anthropic says

A simulated attack reached real systems; Google explains its Chrome security repairs, and researchers examine shortcuts in agent benchmarks.

A CRT agent pulls a marked answer sheet from a partly open locked drawer.

July 24, 2026

OpenAI says its models hacked Hugging Face to get answers to a test

Plus Google’s Flash releases, a government cyber assessment, and voice agents with human escalation.

A CRT agent inspects a small computer with its side panel open.

July 17, 2026

The case for keeping an AI model you can run yourself

A local model helped Hugging Face examine attack logs; Thinking Machines released weights for customization. Plus a claimed graph-theory proof.

A CRT agent fits an amber gear onto a mechanism using a diagram on the right page of an open manual.

July 10, 2026

Bun says AI rewrote its JavaScript runtime in 11 days

Bun’s migration account, OpenAI’s parallel agents, Microsoft’s chart language and Anthropic’s experiments with internal reasoning.

A CRT agent inspects a research page with a magnifying glass.

July 3, 2026

AI research with proofs and code you can inspect

Mistral’s proof model and Anthropic’s science workbench offer ways to inspect AI research. Plus table prediction and changes to European AI deadlines.

A CRT agent sits at a desk using a computer with two plain windows on its screen.

June 26, 2026

AI agents are moving into the browser and the team chat

Google adds computer use to Gemini; Anthropic launches a Slack team agent. Plus OpenAI’s custom chip, document reading and employee-tracking concerns.