Interview, Fireside Chat, Podcast
Anthropic’s $30B Ramp, Mythos Doomsday, OpenClaw Ankled, Iran War Ceasefire, Israel's Influence
- Anthropic's "Mythos" Model withholds: Anthropic withheld its newest model, "Mythos," citing it as too dangerous for public release due to autonomous findings of thousands of critical vulnerabilities.
- Vulnerability Scale: The model identified exploits in major operating systems and web browsers, including a 27-year-old vulnerability in OpenBSD and a 16-year-old bug in FFmpeg missed by 5 million automated scans.
- Exploit Chaining: The model demonstrated the ability to chain 3 to 5 individual vulnerabilities to create sophisticated end-outcome exploits that are ineffective when used independently.
- Project Glasswing: Anthropic launched a coalition of 40 major companies (including Apple, Microsoft, Google, Amazon, and JPMorgan) to spend 100 days using AI to find and patch software vulnerabilities before public release.
- AGI Threshold Argument: Brad Gerstner characterizes models like "Mythos" and OpenAI's upcoming "SPUD" as AGI models requiring a "safe harbor" and sandboxing due to their step-function intelligence improvements.
- Skepticism on Intent: David Sacks suggests Anthropic's fear-mongering is a recurring marketing tactic (citing a "blackmail study" from a year prior that showed no real-world evidence) but concedes this specific cybersecurity threat is likely legitimate.
- Security Timeline: The panel estimates a six-month window where AI-driven cyber defense will outpace offense, urging CISOs to use this period to patch dormant bugs before capabilities become widely available.
- Chamath Palihapitiya's Counter: Chamath argues the "catastrophe" narrative is largely theater, noting that patching all identified vulnerabilities would require shutting down the internet for years and that advanced hacking is already possible with current models like Opus.
- Government Coordination: Chamath and Brad agree that if the threat were real, government coordination (CIA/FBI) is necessary, though Brad asserts this coordination is already happening behind the scenes.
- OpenDeer Access Cutoff: Anthropic cut off access for OpenDeer founder Peter Steinberger's $200/month subscription, forcing high-volume users to the API pricing model as the new "Claude Agents" product launches.
- Competitive Landscape: The panel identifies a concerted effort by Anthropic, Perplexity, OpenAI, Alibaba, and others to dominate the "agent" space and neutralize the open-source competitor, OpenDeer.
- Market Share Debate: Disagreement exists on whether Anthropic holds a dominant (>50%) market share in coding tokens; Chamath argues the total addressable market for AI coding is currently only 5% of total software development.
- Anthropic Revenue Ramp: Anthropic's revenue run rate exploded to $30 billion by early 2025, tripling from $9 billion just months prior in March, driven by 1,000+ enterprise customers paying over $1 million annually.
- Profitability Trajectory: Brad Gerstner predicts Anthropic could reach an $80–100 billion revenue run rate by year-end, noting that low employee counts (2,500) and plummeting inference costs suggest the companies are nearing profitability despite high CapEx.
- OpenAI vs. Anthropic: Despite OpenAI's consumer dominance (GPT-4/4o), Anthropic has gained momentum in the enterprise sector by focusing on coding and "co-work," while OpenAI faces internal strategic shifting and "FUD" (fear, uncertainty, doubt).
- Investment Strategy: Venture capital is crowding into "frontier models" (Anthropic, OpenAI, SpaceX) due to high perceived terminal value, while enterprise software companies face potential consolidation or obsolescence.
- Iran Ceasefire: A two-week ceasefire with Iran is in effect contingent on the Strait of Hormuz opening, with a US delegation (JD Vance, Jared Kushner) negotiating in Islamabad.
- Market Reaction to War: Markets have remained resilient (S&P down only 5-7%), pricing in a likely "off-ramp" and limited US entanglement despite aggressive rhetoric from the Trump administration.
- Foreign Policy Critique: Discussion highlights US domestic skepticism regarding Israeli influence on US foreign policy, with reports suggesting internal White House concerns about Netanyahu's escalation plans.
- X (Twitter) Translation: The automatic cross-language translation feature on X is praised as a "truth mechanism" for enabling real-time cross-border communication between regions like Japan, Russia, and the Middle East.
- Liquidity Summit Updates: The upcoming Liquidity event is sold out with a waitlist of hundreds; confirmed speakers include Andre Karpathy, Dan Loeb, Thomas LaFontaine, and Sarah Fryer.
- Podcast Logistics: David Sacks confirmed attendance at the upcoming All-In Summit in Los Angeles (September 13–15), while the show notes a capacity issue preventing some waitlisted fans from attending the next Liquidity event.