Exclusive: OpenAI briefs feds and Five Eyes on new cyber product - Axios

OpenAI spent the past week bending the ears of 50 federal cyber defense types about its shiny new GPT-5.4-Cyber model, because nothing says 'trust us, we've got this' quite like selling hacking tools to Five Eyes allies and state governments simultaneously. The demo in D.C. was all about bolstering defenses, naturally, but the smirk-inducing detail is that this model — by design — knows how to break systems to patch them. The old 'let us find the hole before the bad guys do' pitch. But every offensive capability revealed to a government official in a briefing room is also a capability that, should the wrongness leak or a state actor acquire similar tech, becomes ammunition in a digital arms race. The irony: we're probably about to see more ransomware attacks than ever, because the same GPT-5.4-Cyber that writes airtight code can also craft perfect phishing emails and discover zero-days. Five Eyes may get a leg up on cyber defense, but they've also got a one-time head start on autonomous attack scripts.