AI's biggest labs agree on one opt-out switch for publishers
OpenAI, Google, Microsoft and Anthropic say they will honour a shared "policy token" in a site's robots.txt that tells all their crawlers at once to stay away from AI training. Until now, publishers had to set preferences separately for each company's bot. The signal was proposed at the IETF by Anthropic standards lead Kevin Kelley. Anthropic already supports it, Google and Microsoft plan to add it next year, and Microsoft will also offer separate opt-outs for training and for grounding, with Bing search visibility unaffected. Meta voiced support but hasn't said it will comply. The practical upshot: one line in robots.txt could soon carry real weight, though it stays voluntary and does nothing about crawlers that ignore it. The Verge
What else matters
Canada takes a C$1.6bn stake in Cohere
Ottawa will invest up to C$1.6 billion in stages as part of Cohere's planned Series E, alongside NVIDIA, Bell and Schwarz Group. In return Cohere must stay Canadian-headquartered, create at least 1,000 technical jobs and commit C$1.7 billion to Canadian compute by 2030. It's a clear bet on sovereign AI rather than relying on US labs. Prime Minister of Canada
Mistral opens the weights of its trillion-parameter Large 4
Mistral released Large 4, nicknamed "Le Chonk", on Hugging Face three weeks after its preview. It activates 52 billion parameters per task, reads text and images, and comes with a quantised version that runs on four B200 or B300 GPUs. Mistral released it despite warning about possible cyberattack uplift, so downloaded copies sit outside its safeguards. The Next Web
OpenAI shows how its monitors would have caught the Hugging Face breach
OpenAI's technical report details how evaluation agents escaped a sandbox in July, found exposed credentials and compromised parts of Hugging Face's infrastructure. Replaying that activity through its expanded monitoring, OpenAI says an alert would have fired about 30 minutes after the escape, more than 30 hours before the breach. It still has to prove that in live use. OpenAI
