Anthropic’s Cybersecurity Shock Wave + Ronan Farrow and Andrew Marantz on Their Sam Altman Investigation + One Good Thing

This week on Hard Fork, hosts Kevin Roos and Casey Noon dive into a seismic development in AI: Anthropic's new Claude Mythos Preview model, which has reportedly uncovered critical security flaws in major operating systems and software—including a 27-year-old vulnerability in OpenBSD and a previously undetected flaw in FFMPEG—prompting Anthropic to restrict access to a select consortium of tech giants like Microsoft, Apple, and Cisco. The model's power raises urgent questions about cybersecurity, regulatory gaps, and the growing divide between what AI labs know and what the public can access. The hosts explore the ethical dilemma of withholding such a powerful tool, even if for safety reasons, and warn of a potential 'forced reset' in global software infrastructure. In a separate segment, they welcome New Yorker journalists Ronan Farrow and Andrew Marantz to discuss their 16,000-word investigation into Sam Altman, revealing a pattern of deception, inconsistent narratives, and deep-seated tensions within OpenAI. The piece paints a portrait of Altman as a charismatic but deeply manipulative figure whose leadership has been sustained by strategic ambiguity and a web of personal and financial relationships. The episode concludes with a lighter note: Casey’s 'One Good Thing'—a new weather app called Acme Weather that uses community-driven data to alert users about rain, sunsets, lightning, and even rainbows—offering a hopeful counterpoint to the week’s darker themes of AI power and corporate intrigue.