OpenAI gates cyber model
OpenAI released GPT-5.4-Cyber under restricted access, limiting use to verified partners and professionals because of misuse concerns. Axios reports some enterprises, including BNY, have already gained privileged access to advanced cyber models from OpenAI and Anthropic. (uk.finance.yahoo.com) (axios.com)
OpenAI has put its new cybersecurity model behind a gate, offering GPT-5.4-Cyber only to vetted users instead of the public. (openai.com) OpenAI said on April 14 that it is expanding its Trusted Access for Cyber program to “thousands” of verified individual defenders and “hundreds” of teams that protect critical software. The company described GPT-5.4-Cyber as a fine-tuned version of GPT-5.4 built to be more permissive for defensive security work. (openai.com) The company said access will rely on identity checks and know-your-customer controls, and that only users in higher tiers of the program can request the model. Axios reported on April 14 that OpenAI paired that rollout with a tiered access plan for advanced cyber models. (openai.com) (axios.com) Cybersecurity models help researchers find software flaws before criminals do. OpenAI’s 2025 Preparedness Framework lists cybersecurity as one of three “tracked categories” of frontier AI risk, alongside biological and chemical capabilities and AI self-improvement. (cdn.openai.com) That caution sharpened after Anthropic’s restricted release of Claude Mythos Preview on April 7. Anthropic said the model could identify and exploit zero-day vulnerabilities in every major operating system and web browser during testing, and said even the small share of bugs it disclosed showed a large jump in cyber capability. (red.anthropic.com) OpenAI is taking a different deployment path. In its April 14 post, the company said it wants “democratized access” for legitimate defenders over time, but only through iterative rollout and safeguards that tighten as model capabilities rise. (openai.com) Large companies are already getting in line. Axios reported on April 16 that Bank of New York Mellon has early access to advanced cyber-capability models from both OpenAI and Anthropic, according to Chief Executive Officer Robin Vince. (axios.com) That puts banks, software vendors, and government-facing security teams in the first wave of a new AI arms race centered on bug hunting and patching. For now, OpenAI is keeping the gate closed to everyone else. (axios.com) (openai.com)
