Overview
- OpenAI released GPT-5.4-Cyber for vetted defenders, pairing it with an expanded Trusted Access for Cyber program that eases identity checks and lowers needless refusals for lawful research.
- The new model targets defense tasks such as binary analysis without source code and vulnerability triage, and OpenAI says its Codex Security system has already fixed more than 3,000 high and critical bugs.
- Politico reporting says U.S. federal staff and the Commerce Department’s AI Standards and Innovation Center are quietly testing Anthropic’s Mythos despite a White House directive limiting government use, with congressional staff seeking briefings.
- Bank of England Governor Andrew Bailey warned that Mythos could change the cybersecurity risk landscape and urged regulators to gauge how far such systems can find exploitable weaknesses in existing tech.
- Investors are repricing Anthropic as Business Insider reports bids valuing the company as high as $800 billion after it disclosed $30 billion in annualized revenue and more than 1,000 customers paying over $1 million a year, while OpenAI faces enterprise go-to-market strain tied to its Microsoft pact and rising interest in AWS access.