Descriptions:
The All-In Podcast’s April 10 episode opens with an extended discussion of Anthropic’s Claude Mythos, an unreleased frontier model the company says is too dangerous to deploy publicly. According to Anthropic’s own system card, Mythos Preview autonomously found thousands of high-severity vulnerabilities — including a 27-year-old flaw in OpenBSD and a 16-year-old bug in FFmpeg — scoring 83.1% on cybersecurity vulnerability reproduction benchmarks compared to 66.6% for Claude Opus 4.6, and outperforming Opus by 24 percentage points on SWE-bench Pro. Rather than a general release, Anthropic is granting access only to vetted cybersecurity teams under its Project Glass Wing initiative.
The hosts — Chamath Palihapitiya, Jason Calacanis, and guest Brad Gerstner — debate the societal implications, the credibility of Anthropic’s safety framing, and whether withholding a model of this capability is precedent-setting or performative. The conversation then pivots to Bit Tensor subnet 62 (Ridges AI), an open-source, crypto-incentivized coding project that reportedly reached 80% of Claude 4 coding performance in under 45 days for approximately $1 million in TAO token rewards.
Additional segments cover Iran ceasefire dynamics, Israel’s influence operations, and OpenAI’s financial trajectory, though the Mythos and open-source AI competition threads dominate the technical portions of the episode.
📺 Source: All-In Podcast · Published April 10, 2026
🏷️ Format: Podcast
