Jack Roberts and Nick Saraev discuss reports that Anthropic has completed Mythos 2 without plans for a public release. They argue that unreleased internal systemsAn unreleased model is an AI model developed or trained internally but not made available for public use or independent access. make the apparent gap between open models and the frontierA frontier AI model is among the most capable general-purpose models available at a given time. look smaller than the real capability gap.
Roberts and Saraev connect restricted access to cybersecurity, alignment and national-security concernsAlignment is the effort to make AI systems reliably pursue intended goals and follow human values, constraints, and instructions.. They acknowledge that the details remain opaque, while suggesting that government interest in powerful models could shape what laboratories make available to ordinary users.
The hosts also examine Grok 4.6's stronger agentic benchmark result and Sam Altman's prediction that ChatGPT could gain much broader personal contextPersistent personal context is user-specific information that an AI system retains across interactions so future responses can reflect prior preferences, history, and ongoing work.. They distinguish raw model intelligence from practical tool use, then weigh the productivity benefits of persistent context against the need to keep personal data from becoming an uncontrolled surveillance surfaceData privacy governs how personal, confidential, or sensitive information is collected, used, shared, retained, and protected in AI systems..
Watch on YouTube
