Cold snap
Supermarket freezers get hacked again, Google dismantles a TV-box botnet, and Anthropic's IPO filing shows $17 billion in quarterly revenue.
The 10
Ranked by how many sources covered it-
Anthropic tells Claude to stop recreating existing video games, after retro gaming was flooded with AI-made decompilations, recompilations and ports. An Anthropic employee says the company is still tuning the right balance.
Why it matters Model makers are starting to write intellectual property limits directly into their coding tools, with legitimate reverse engineering caught in the middle.
-
ChatGPT's Chat and Work tabs are merging into one interface, with ChatGPT and Codex drawing usage from the same allocation pool. OpenAI also plans to end support for the ChatGPT Classic app on macOS early next year.
-
A prosecutor cited an imaginary case in a high court matter after an AI search, in a child rape case, adding to dozens of AI-related errors in court proceedings, according to the Guardian's exclusive on a senior Crown prosecutor and nonexistent legal precedent.
Why it matters Fabricated citations reaching serious criminal proceedings raise the stakes for verifying AI-assisted legal research.
-
Open models handle most agentic work at about a twentieth of the cost of a frontier LLM, Gartner says, putting the value in the harness rather than in the model itself.
Why it matters If open models can cover most agent workloads, enterprise spending may shift from model access toward the tooling around it.
Get the issue at 8 AM ET, weekdays.
Ten stories and a few good reads. Four minutes, then back to your day.
Also worth your time
**Qwen model plays DOOM** because someone compiled a 32-bit RISC-V CPU straight into its weights, so the parameters double as RAM for the experimental Qwenputer build. / tomshardware.com
**Astra reads and writes weights** so it can improve Qwen directly, which is the core idea of weight-space programming. / rikiyat.github.io
**Claude draws permit-level plans** by the look of one Substack post, which argues the model can now produce permit-ready architectural drawings. / stephenjgallagher.substack.com
**AI agent operates on mouse** overnight, a feat the post describes as brain surgery, with few details on how the procedure went. / twitter.com
**Anthropic's internal AI bill** at retail token prices now exceeds what the company pays its employees, a comparison that puts its own usage in focus. / businessinsider.com(paywalled)