The file
Sourced AI briefs in publish order — newest first. Citeable originals live on the homepage under From Signal Desk.
Check Point, which says it helps secure over 100,000 organizations, uses OpenAI’s Daybreak to turn newly disclosed flaws into protections faster. Next is an agentic system that monitors customer networks with human approval for consequential actions. The defensive case is strong; the claim is the CTO’s own.
Greptile uses GPT-6 Astra’s computer use to spin up a pull request, shrink the browser to phone size and click through like a user. Its CEO says human QA is too slow for AI-speed shipping. Behavior-level testing is a real improvement; the numbers are absent and critical code deserves friction.
Anthropic’s S-1 says data centers lean increasingly on natural gas, some planned facilities use on-site turbines, and a warming climate will raise its own power and cooling costs. The candor beats a carbon pledge, but neighbors bear the pollution and bills.
Apple’s 27.2 updates let Siri recall past conversations and draw on Messages photos, Calendar attendees, voicemails and Safari history. A new Excluded Apps setting lets users wall off apps in one place. Retention details are unclear, and Siri AI still isn’t available in the EU or mainland China.
The Verge showed Muse could re-add a unique luxury listing every time The RealReal’s 20-minute hold expired, blocking other buyers. The site now makes agent login and checkout much harder while still allowing discovery. Systems built around human patience are meeting software with none.
Over half of Anthropic customers spending $100,000-plus now use both the Claude platform and a first-party app, up from about 30%, and net revenue retention is about 900%. First-half revenue hit $16.3 billion. It is a strong land-and-expand business, and a dependency question for buyers.
Nace AI says its Drex for Search lets agents find cited evidence in huge document piles with far fewer search calls, and the claims are worth testing because they come from the vendor.
Anthropic’s prospectus warns that recursive self-improvement could outrun its safeguards and pose existential risk, then calls ‘closing the loop’ one of its most significant structural advantages. The candor is unusual; the contradiction is the industry’s collective-action trap in one document.
At the 2026 OCP Global Summit, Meta detailed the networking behind Muse models and its Prometheus cluster, open-sourced its MetaRoCE transport and co-led a new switch platform standard with Microsoft and Alibaba. Open, multi-vendor networking lowers barriers, and makes gigawatt build-outs easier to copy.
Polar Peaks, the third AI Quests game from Google Research and Stanford, has students aged 11 to 14 clean data and test models while mapping neurons. Over 400,000 quests have been completed in 10 languages. The lesson is sound; one company supplying the curriculum is the caveat.
NVIDIA's Minerva points machine learning at the AI data center itself, chasing the quiet hardware faults that waste GPU hours, and it's a good use of AI that also deepens NVIDIA's lock on the stack.
Microsoft is offering voluntary exit packages to as many as 60,000 of its roughly 223,000 employees, months after cutting 4,800. Teams in Copilot, CoreAI, Windows and Xbox are eligible; security and quantum are not. It is gentler than layoffs, but voluntary exits tend to lose the people with options.