Anthropic released Mythos, a frontier-grade AI model previously restricted to cleared defense contractors, directly to the public. The move eliminates the technical boundary between government-cleared AI systems and commercial developer tools, fundamentally shifting access to state-of-the-art capabilities.
DeepMeind's Demis Hassabis accelerated his AGI timeline from five to ten years to 2029, citing AlphaProof Nexus solving nine open Erdős problems as evidence of rapid mathematical reasoning progress. These problems had resisted solution for decades. The breakthrough cost minimal computational resources, suggesting efficiency gains in reasoning systems.
Security incidents accelerated simultaneously. Starlette, a web framework powering roughly one million AI agents, had critical zero-day vulnerabilities exposed. CrowdStrike coordinated a takedown of Glassworm, a developer botnet operating across four command-and-control channels, indicating organized AI infrastructure threats are materializing.
Geopolitical stakes sharpened. BNP Paribas formalized a sovereign-AI security partnership with Mistral, France's homegrown model builder, in response to U.S. and Chinese AI dominance. Beijing froze overseas travel for top AI engineers at Alibaba and DeepSeek, restricting brain drain while centralizing AI development.
The labor impact became undeniable. Uber exhausted its full annual AI token budget by April, suggesting AI cost efficiency improved enough that planned computing spend now depletes in months rather than years. This compression signals either rapid productivity gains or aggressive internal AI adoption outpacing budget expectations.
The week exposed three trends. Commercial access to frontier models is collapsing compartmentalization between defense and civilian use. Technical breakthroughs in reasoning are compressing AGI timelines based on real mathematical proofs, not extrapolation. Security gaps and ge