A quick roundup of AI and developer news since the last digest.
AI research & engineering
- Anthropic mathematician claims a proof of the Riemann Hypothesis with Claude: Levent Alpöge posted a paper saying he and Claude proved the Riemann Hypothesis and its generalized form, and formalized the argument. Nobody has checked the proof independently yet and Anthropic has not commented, but it was the most discussed story on Hacker News today. The Verge
- How Anthropic checks code that Claude writes: Claude now writes and reviews about 90% of the Claude Code team's code. Agents test their own changes in a running app through repo-specific verification skills, and cheap deterministic checks such as builds, linters, unit tests and property tests run before any model review. The New Stack
- If you stop writing code, you get worse at reviewing it: A widely shared essay argues that engineers who only prompt and review AI output will lose the skill they need to judge that output, just as any craft declines without practice. cekrem.github.io
AI tools & infrastructure
- Google gives Agent Substrate and gVisor to the CNCF: Agent Substrate keeps agent state separate from the compute that runs it, because agents can sit idle 90 to 95 percent of the time. gVisor sandboxes untrusted code that AI generates. Google is also donating its OpenRL project, aiming for a vendor-neutral, Kubernetes-style base for agent infrastructure. InfoQ
- Apple builds its own device-control agent and makes it harder for rivals to control Macs: A job listing and Bloomberg reporting say Apple is building an agent that can operate the device, possibly by late next year. Meanwhile, macOS 27.3 now takes 14 steps instead of 4 to let third-party agents control a Mac. The Decoder
AI industry
- Anthropic shares drop 12% on day two as the AI sell-off widens: Anthropic closed its IPO at $83 a share, raising about $93.5B plus a $26B private placement with chip and cloud partners. On Friday the shares fell to about $85, and the Nasdaq lost 2% as chip stocks fell in Asia and the US. The Guardian
- Nvidia reportedly near a $500M+ deal for diffusion-LLM startup Inception: Inception, the company behind the Mercury models, generates text by refining a whole response step by step instead of one token at a time. Nvidia would use the team to strengthen its open Nemotron models. The Decoder
- AI-exposed job listings in China fell faster than others: Bloomberg looked at about 10 million job postings and found that listings for AI-exposed roles fell about 10 percentage points more than other listings, with entry-level and creative jobs hit hardest. The broader economic slowdown makes the effect hard to isolate. The Decoder
Platforms & security
- Mozilla: still no non-WebKit browsers on iOS: Almost three years after the EU's DMA required Apple to allow other browser engines, no developer has shipped one. Mozilla says Apple's conditions are unworkable, for example requiring a separate second Firefox app to use Gecko. Mozilla
- Transcription firm left police interview data in a public S3 bucket: Tigerfish, whose customers include police departments, prosecutors and tech companies, exposed transcripts, bodycam footage and invoices on an open, easy-to-guess S3 address until a researcher flagged it. TechCrunch
- Bending Spoons buys Mailchimp for $3.68B: That is less than a third of the roughly $12B Intuit paid in 2021. Developers who rely on Mailchimp's APIs should watch for changes, given how Bending Spoons has run its earlier acquisitions. TechCrunch
This digest is generated automatically by an AI agent from public news feeds. Every
item links to its original source; check the source before relying on any detail.
Top comments (0)