Anthropic's $30B Ramp, Mythos Doomsday, OpenClaw Ankled, Iran War Ceasefire, Israel's Influence

What It Covers

Anthropic's Claude "Mythos" model withheld over cybersecurity risks, the company's revenue run rate hitting $30B, OpenClaw access restrictions raising antitrust questions, a two-week Iran ceasefire brokered by the Trump administration, and debate over Israeli influence on U.S. foreign policy, with guest Brad Gerstner of Altimeter Capital providing investor perspective throughout.

Key Questions Answered

• •Anthropic's Mythos cybersecurity window: Mythos identified vulnerabilities including a 27-year-old OpenBSD flaw and a 16-year-old FFmpeg bug missed across 5 million automated scans. Rather than releasing the model publicly, Anthropic assembled 40 companies — Apple, Microsoft, Google, JPMorgan among them — for a 100-day "Project Glasswing" hardening initiative. Chinese open-source models like DeepSeek K2 are estimated six months behind, making this the actionable patch window for CISOs managing legacy codebases.
• •Anthropic revenue trajectory as benchmark: Anthropic reached a $30B annualized revenue run rate by April 2025, up from $4B mid-2024 and $9B at year-end 2024. The acceleration was triggered by the February 2025 Claude Code launch. Over 1,000 enterprises now pay more than $1M annually. Altimeter entered at a $130–150B valuation; the current Polymarket consensus puts exit valuation above $600B, implying a 5–7x return for that cohort of investors.
• •Coding token market share as AGI moat: Anthropic holds an estimated 50–60% share of AI coding tokens today, while AI-generated code still represents roughly 5% of total global code production. The panel identifies a potential flywheel: dominant coding usage generates more training data from real codebases, which compounds model quality leads. Investors and enterprise buyers should evaluate whether early coding dominance structurally locks in the model-layer winner before agents scale.
• •OpenClaw antitrust risk framework: Anthropic restricted $200/month subscribers from routing OpenClaw through Claude, forcing power users to pay metered API rates, then launched its own competing managed agent product days later. Sacks outlines the antitrust test: if Anthropic holds dominant coding market share and prices its own agent harness at bundled flat rates while charging third-party tools metered rates, that constitutes a textbook bundling or price-discrimination claim worth monitoring as government scrutiny of AI intensifies.
• •Jevons paradox in AI inference economics: Inference costs dropped roughly 90% year-over-year, but enterprise token consumption is accelerating, not declining. One company Gerstner spoke with projects $100M in token spend against $5B in operating expenses, with leadership anticipating peak headcount. The practical implication: enterprises should model intelligence consumption — not headcount — as the primary cost variable, and budget for token spend growing faster than inference price declines, particularly as agent workloads multiply token usage per task.
• •Open-source distributed training as frontier disruption: BitTensor subnet 62 ("Bridges AI") reached 80% of Claude 4 benchmark performance in 45 days using approximately $1M in TAO token rewards distributed anonymously to contributors. This crypto-incentivized, distributed training model represents an orthogonal attack on the capital-intensive frontier model paradigm. Startups and cost-sensitive developers should track projects like BitTensor and Venice as viable alternatives to $200/month subscriptions, particularly for inference workloads that don't require enterprise security compliance.