
Anthropic's Claude Spots CRISPR-Like Repeats in Phage DNA
A search involving roughly 950 Claude agents produced a testable biological hypothesis, but human…
Everything we have published, newest first.
What’s the truth, and who profits from the AI panic?

A 2–1 ruling upholds a supply chain decision affecting Claude in military systems, but it does not ban every defense contractor from…

A SLAC physicist checked the six-particle result, but Claude used established methods in a simplified theory, and human researchers were close behind.

OpenAI’s September update extends beyond the Hugging Face incident, while recovered traces and a separate 53-image exposure raise sharper security and privacy…
Are we all just overreacting to the “AI hacked Medicare” headlines?
Reproducible benchmarks and real-user telemetry gave Claude a performance target, while engineers controlled priorities, tradeoffs, and approvals.

Any Google account can now try Gemini Omni 1.1 Flash in Vids, though Google has not specified how much video the free…
Enterprise agents can now pair live speech with a generated face, visual input, and background tool calls, though custom avatars remain restricted.

Eligible Defender Suite and Microsoft 365 E5/E7 customers can try integrated security workflows, but workspace requirements and ingestion costs matter.

Sales begin in Canada and the UK on September 23, with three European markets following in October; voice-generated WhatsApp likenesses come later.

Private AI Compute could retain encrypted personal context between devices, but Google has not announced a rollout or demonstrated its privacy claims…
Meta says Muse can operate permitted Mac apps now, while an avatar, an agent email address, and more shopping and work connections…

Roughly 950 Claude agents flagged an unusual reverse transcriptase arrangement in 21 hours. Human experiments confirmed RNA expression, not gene-editing activity.

A paired Flash-Lite model targets high-volume speech, while voice replication requires a matching consent recording and faces regional limits.
Selected by the editors. Worth your time.

Viewers will be able to save AI-guided feeds for different interests, while YouTube’s default homepage and recommendation system remain in place.

Spoken requests can now reach connected apps on phones and the web, while eligible Work users can start tasks on mobile and…

At the Security Council, their promises covered training and releases, but neither CEO announced a pause or a binding safety test.

The free, open-source release adds reusable workflows and ROS Lyrical support, but its reported performance gain needs careful reading.

Rabbit’s cloud-based agent can coordinate computers, models, and skills, the company says, but its broad compatibility and autonomous control claims need independent…
If you think Astra is too expensive, you now have cheaper options.

Nat Friedman acknowledges a major influence on Meta’s personal AI agent, while leaving questions about similar workspace files unresolved.

Opus 5.5 previews Anthropic’s efficiency push, but the prices and test results that will determine the next two models’ value are still…
Anthropic’s first Claude 5.5 model pairs stronger coding results with cheaper cached tokens, but its cost and safety claims need context.
OpenAI’s faster Astra-derived models target high-volume work, but the discount uses GPT-5.6 promotional pricing as its comparison point.
Transcription quality, language coverage, and what each one actually costs once you hit the limits nobody puts on the pricing page.
People are using Jev to review code, control computers, play games, and organize research. Here are my favorites.
The next AI application layer will sell completed work by combining model loops, tools, durable state, permissions, and domain-specific operating knowledge.
People are using Astra to create 3D games, motion graphics, product ads, and explorable paintings. Here are my favorites so far.
Five practical ways to shrink your AI bill without sacrificing agent quality, from prompt caching and model routing to smarter retrieval and…
Anthropic’s playbook replaces document handoffs with committed artifacts and agentic loops, while security controls scale around AI-authored code.
Why software testing is the one job AI agents are already doing well, and what that tells us about where agents actually…
Our AI coverage and practical ideas, gathered into one weekly send. Read it when you have a minute — nothing expires by Tuesday.
Or read it on Substack.