OpenAI gates cyber models

OpenAI is putting some of its newest specialist models behind identity checks and invitation lists instead of releasing them broadly. (openai.com) On April 14, OpenAI said it was expanding its Trusted Access for Cyber program to “thousands” of verified defenders and “hundreds” of security teams, and it introduced GPT‑5.4‑Cyber, a GPT‑5.4 variant tuned for defensive cybersecurity work. (openai.com) OpenAI said access to the cyber model will rely on strong know-your-customer checks and identity verification, and it framed the rollout as preparation for “increasingly more capable models” expected in the next few months. (openai.com) Cybersecurity models can help defenders find and fix software flaws faster, but the same skills can also help attackers. Anthropic made the same case on April 7 when it limited Claude Mythos Preview to selected companies under its Project Glasswing program. (anthropic.com) (cnbc.com) Anthropic said Mythos could identify and exploit zero-day vulnerabilities in major operating systems and web browsers during testing, and it restricted access to partners including Microsoft, Amazon, Apple, CrowdStrike and Palo Alto Networks. (anthropic.com) (cnbc.com) OpenAI is using the same gated pattern outside cyber. On April 16, it introduced GPT‑Rosalind, a life sciences model for biology, drug discovery and translational medicine, and made it available as a research preview in ChatGPT, Codex and the application programming interface for qualified customers through its trusted access program. (openai.com) (usnews.com) OpenAI said GPT‑Rosalind is built for tasks such as evidence synthesis, hypothesis generation and experimental planning, and it launched a Codex plugin that connects to more than 50 scientific tools and data sources. The company said customers using the model include Amgen, Moderna, the Allen Institute and Thermo Fisher Scientific. (openai.com) (bloomberg.com) Reuters reported on April 16 that OpenAI’s life sciences push followed its April 14 cyber release, tying both launches to a broader move into higher-stakes specialist tools. OpenAI said researchers can use Rosalind to query databases, read recent papers, use scientific tools and suggest new experiments. (usnews.com) CrowdStrike has also been named publicly in connection with these restricted-access programs. Anthropic listed CrowdStrike as a Project Glasswing participant, and third-party reports on April 16 said CrowdStrike had been selected for OpenAI’s Trusted Access for Cyber program as well. (cnbc.com) (article.wn.com) The result is a new release model for frontier artificial intelligence: broader consumer products on one track, and the most cyber-capable or lab-oriented systems on another, with access tied to verification, enterprise relationships and controlled previews. (openai.com 1) (openai.com 2)