Glaneur
Glaneur AI

NOV 6 - OpenAI unveils GPT-6 Cyber

Friday, November 6th · 2 min 34

In this episode

Full transcript

Good morning. It's Friday, November 6th, and here's the AI news, from France.

OpenAI has unveiled GPT-6 Cyber, a model trained specifically to find security flaws, build working exploits and suggest fixes. In targeted tests, it found 81 previously unknown critical vulnerabilities in the Firefox browser, which Mozilla has since patched. In the Linux kernel, it reproduced 1,724 known flaws and built working exploits for 948 of them. That is more than five times its predecessor. OpenAI rates the model Critical for cybersecurity under its own risk framework. Separately, a startup engineer told TechCrunch he used an earlier OpenAI model to find a severe flaw in Ubuntu, in about two hours.

So how good are AI agents at real work? Not very, according to GDPx, a new benchmark from the analytics firm SemiAnalysis. It drops agents into simulated jobs for up to eight hours, across 206 tasks in fields like chip design, biology and law. Claude Opus 5.5 leads with just 12 percent. Google's much cheaper Gemini 4 Carbon follows at 10.9 percent, then GPT-6 Astra at 9.6. Critics question how meaningful the test is, since the tasks are kept secret.

Airbnb says its own AI travel agent will launch in the first half of 2027. Chief executive Brian Chesky argues that a chatbot is the wrong way to plan a trip. The agent will instead compose its answers from maps, photos, comparisons and itineraries.

In brief. Anthropic is adding a Personal Finance tool to Claude, in beta. Users can connect bank and card accounts through Plaid, with read-only access, and ask where their money went.

Meta's president, Dina Powell McCormick, warned that Democratic wins in the midterms will make data centers harder to build. She said the industry has not done a good enough job explaining their benefits.

And China's Tencent has released Hy4, a model that can read images and operate a computer. Independent tests rank it second among Chinese models, but well behind the leaders on hard math and coding.

That's it for today. See you tomorrow.

Subscribe

RSS feed Apple Podcasts, Spotify