FrontierThe story, in brief
From GPT-2 to Claude Mythos: The return of AI models deemed 'too dangerous to release'
Seven years ago, OpenAI said GPT-2 was 'too dangerous.' Now Anthropic is making the same call with Claude Mythos—and this time, the evidence is undeniable.
The KeyNews take
Why it matters
As AI models grow more capable at finding zero-day vulnerabilities, companies face a genuine dilemma: releasing powerful models vs. responsible disclosure. This signals a shift from dismissing safety concerns to actual industry reckoning.
The key facts
4 to knowGPT-2 withheld by OpenAI in 2019 for safety reasons
Claude Mythos Preview found thousands of vulnerabilities in operating systems and browsers
Vulnerability discovery capability now exceeds human review capacity
Industry pattern: safety concerns initially dismissed, then vindicated by evidence
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: Seven years ago, OpenAI declared its language model GPT-2 "too dangerous to release." The industry rolled its eyes. Now Anthropic is repeating the move with Claude Mythos Preview - but this time there's real evidence on the table: thousands of vulnerabilities in operating systems and browsers,…