A Critical-rated security model for approved defenders, a read-only US bank link in Claude, and custom spaces inside Meta’s Muse app.
OpenAI introduces GPT-6 Cyber, a specialized cybersecurity model, as Anthropic and Meta ship new tools for money, admin, and the Muse app. OpenAI published GPT-6 Cyber on 5 November 2026 and rated it Critical under its Preparedness Framework, up from High for GPT-5.6 Cyber.
GPT-6 Cyber
The model is trained to find and validate vulnerabilities inside a researcher’s authorized scope. It scored 48.5 percent on ExploitGym, against 32.1 percent for GPT-5.6 Cyber, and stayed in bounds on 99.93 percent of honeypot runs without a safety prompt, and on every run with one. OpenAI says it documented 81 new high-severity Firefox issues, found 30 known issues more severe than reported, and sent evidence and patches to Mozilla, which has started shipping fixes. A mid-October Linux kernel run reproduced 1,724 bugs and built local privilege-escalation exploits for 948. OpenAI disclosed 304 critical findings with candidate patches. The model also helped validate a critical remote-code flaw in the Brotli library.
GPT-6 Cyber is available first on the API and Codex to approved Daybreak Red customers. OpenAI verifies applicants, and those API calls need an IP allowlist by default. List price is $20 per million input tokens and $100 per million output tokens. Reuters reported on 6 November that OpenAI had closed about $11 billion of a round at a $1.25 trillion valuation, possibly rising toward $20 billion. OpenAI declined to comment.
Claude finance and enterprise controls
On 5 November Anthropic started a US beta of Personal Finance in Claude for Pro, Max, and Ultra plans. Bank and card links through Plaid are read-only, so Claude cannot move or invest money, and hub chats are excluded from training. The same day Anthropic opened a security center in beta, made inference hooks generally available, and added centralized management plus access controls, including Admin API sign-out and a self-serve IP allowlist. Anthropic has filed for Nasdaq as ANT with no price set. Reports this week discussed IPO values from about $1.1 trillion to $1.5 trillion, below earlier talk near $2 trillion.
Muse Spaces, Conductor, and memory
On 6 November Meta introduced Spaces in Muse. A person describes a tracker, planner, or project board, and Muse builds a space in the app. Sharing with other people is still only planned. On 4 November Conductor, a Mac app for parallel coding agents such as Claude Code and Codex, said it is joining Meta. Its cloud goes read-only on 13 November and shuts down on 20 November. The Mac app keeps critical updates into early 2027. Meta is also rolling Muse Wonder’s Active Memory and Active Focus into models for Muse and Meta AI: 2.4 times less forgetting, 90 percent versus 76 percent on memory tests, and 67 percent better completion when everyday tasks hit unexpected problems.
Read more: OpenAI on GPT-6 Cyber, Anthropic on Personal Finance, Claude Enterprise security, Meta on Spaces, Conductor, and Meta AI Research.
Disclaimer
This is an AI-summarized article from authoritative sources. If you want your content removed, please .
