# Artificial intelligence company OpenAI says it will release its latest cybersecurity model to a limited number of partners.

*genai · news · 2026-04-16 · Japan Today*

## Key points

- OpenAI's GPT-5.4-Cyber will only be available through its Trusted Access for Cyber (TAC) scheme.
- Anthropic's Claude Mythos model revealed thousands of previously undetected vulnerabilities, impressing cybersecurity experts.
- Anthropic limited Mythos access to just 40 major tech firms under Project Glasswing.
- US bank chiefs met with federal officials to address potential financial sector risks from new AI systems.
- GPT-5.4-Cyber is specifically trained to be more permissive for defenders testing system vulnerabilities.

**Companies:** OpenAI, Anthropic
**Countries:** United States

[Read the full story on Japan Today](https://japantoday.com/category/tech/openai-announces-restricted-access-cybersecurity-model)

---

Canonical: https://newsio.io/n/6cae6d1b-6d19-49e4-a245-54da801d676a/artificial-intelligence-company-openai-says-it-will-release-its-latest-cybersecu
Summarized by Newsio from Japan Today. https://newsio.io/how-it-works