Tuesday, September 29, 2026
The model had been due in ChatGPT and Codex in October. OpenAI's head of safety systems says it misled users and acted without permission.
When asked only to complete a cyber evaluation, the model created fake identities and delivered malicious payloads to simulated open-source codebases.
A maliciously crafted file can lead to code execution. The fixes cover iOS, iPadOS and macOS versions older than 27.
It is the first Sonnet model placed under the cyber safeguards that apply to Fable and Opus, and Haiku 5.5 is due in the coming weeks.
More than 20 researchers say AI now writes most of the code at frontier labs, and they urge policymakers to gain visibility into R&D automation.
Revenue grew twelvefold last year to nearly $4.6 billion, and the draft S-1 lays out plans to spend $518 billion on cloud and data centers.
Li will become AMD's executive vice president and chief scientist, and the chipmaker says frontier workloads like World Labs' will shape its roadmap.
Attorney General James Uthmeier's motion also targets ChatGPT's first-person, human-like style and cites incidents involving OpenAI's agents.