Anthropic released Mythos, a frontier-class model previously restricted to cleared defense contractors, to the public. The move eliminates the distinction between government-cleared AI systems and commercially available models. DeepMind head Demis Hassabis shortened his AGI timeline from five to ten years to 2029, citing AlphaProof Nexus solving nine unsolved Erdős mathematical problems as evidence of rapid capability acceleration.
Security vulnerabilities dominated the week. Starlette, a web framework powering roughly one million AI agents, disclosed critical zero-days. CrowdStrike led a coordinated operation against Glassworm, a developer botnet operating across four command-and-control channels.
Geopolitical AI moves intensified. BNP Paribas signed a sovereign AI security agreement with French model provider Mistral, signaling EU financial institutions' shift toward European AI infrastructure. Simultaneously, Beijing restricted overseas travel for senior AI engineers at Alibaba and DeepSeek, tightening state control over talent mobility in the AI sector.
Economic disruption entered harder numbers. Uber exhausted its full annual AI token budget by April, with the company deploying models for operational decisions faster than initially projected. This suggests real-world AI adoption outpacing internal forecasts.
The week compressed several converging trends. Frontier AI capabilities now sit in developer hands rather than government labs. Mathematical breakthroughs accelerate AGI feasibility claims. Security holes grow with deployment scale. Nation-states treat AI talent as strategic assets. And cost-per-capability keeps dropping, forcing companies to accelerate spending or fall behind.
The Mythos release carries the sharpest signal. Public availability of contractor-grade systems collapses years of gatekeeping in weeks. This reshapes the competitive landscape. Every organization with API access gains equivalent compute to defense labs