OpenAI Launches GPT-5.4-Cyber as Banks Scrutinize Risks From Anthropic’s Mythos

Overview

• OpenAI introduced GPT-5.4-Cyber for early users in its Trusted Access for Cyber program, pitching the model to help organizations spot software flaws.
• Anthropic is holding back a broad release of Claude Mythos and is running private tests through Project Glasswing with major tech firms and banks, including JPMorgan Chase.
• Government officials in the United States, the United Kingdom, and Canada have met with large banks to discuss cyber risks tied to Mythos, according to the U.S. Treasury and news reports.
• Anthropic says the preview model found thousands of high‑ and critical‑severity bugs, including a long‑standing flaw in the FFmpeg media library, while independent groups note the trials lacked real‑world defenders.
• The Pentagon still bars Anthropic after branding it a supply‑chain risk, and a company cofounder says he is in talks with the Trump administration about the model.