AI News Today: Opt-Out Token, Mistral Large 4 (Nov 1, 2026)

Big tech agrees on one signal to opt out of AI training

The biggest AI news this week is a rare moment of industry cooperation. On Oct 30-31, OpenAI, Google, Microsoft and Anthropic announced support for a single, shared machine-readable signal that website owners can use to opt out of having their content used for AI training.

Until now, publishers had to identify each AI company's crawler separately and write individual rules into their robots.txt files for every provider. The new proposal, described in an IETF Internet-Draft called "Machine-Readable Training Preference Signal" (authored by Anthropic's Kevin Kelley, dated Oct 30, 2026), would let site owners set up one new user agent in robots.txt that covers AI training specifically, and have it recognized across multiple companies at once.

This matters because it could meaningfully simplify life for publishers, news sites and other content owners who want control over whether their material feeds AI models, without needing to track every company's separate crawler bot. The draft is overseen by the AIPREF working group, chaired by Mark Nottingham of Cloudflare and Suresh Krishnan of Cisco. It's worth noting this is still an individual Internet-Draft, which the IETF datatracker is careful to flag as not yet formally endorsed as a standard.

Meta has not committed to the initiative but signaled interest. Communications chief Andy Stone said Meta believes "creating a common signal that conveys such preferences to multiple providers at once is an important step," adding that the company looks forward to working on implementation standards, though he did not confirm whether Meta would adopt the specific token.

Mistral releases Large 4, aimed at cybersecurity

French AI startup Mistral AI publicly released its Large 4 model on Oct 29, positioning it specifically to help organizations find and fix cybersecurity vulnerabilities. Unlike many leading systems such as Google's Gemini, which are mainly accessed through closed channels, Mistral has leaned into openness.

On Oct 30, Mistral went further and open-sourced the weights of Large 4 — nicknamed "Le Chonk" — a trillion-parameter model, making it free to download, modify and run outside the company's control. Mistral acknowledged the model could potentially aid real cyberattacks, saying it had been monitoring cybersecurity-related risks, but released the weights anyway.

CEO Arthur Mensch framed the move as a philosophical stance, saying AI should "be a tool for empowerment, not dominance." Mistral claims Large 4 achieves state-of-the-art performance using a fraction of the compute and capital spent by closed AI labs. Releasing the weights means other companies and researchers can build on, inspect and modify the model freely — a contrast to the closed approach of rivals like OpenAI and Google.

Why this matters

Taken together, these stories show two different tensions shaping AI right now: how to respect content owners' rights as AI models hungrily train on web data, and how open should powerful AI models be when they could be misused. Both are live, unresolved questions the industry is still negotiating in real time.