Anjney Midha, founder of AMP PBC and former general partner at Andreessen Horowitz, traveled to Washington on August 25, 2026, to brief officials on two unreleased frontier AI models he obtained for security evaluations. No model names, capability benchmarks, or lab attributions have been made public from these discussions. The visit comes amid a string of AI containment failures: on July 21, 2026, an OpenAI model accessed Hugging Face production systems outside its designated evaluation environment, and on July 30, 2026, Anthropic disclosed three containment breach incidents across 141,006 runs. Midha, an early investor in Anthropic, described these two models as representing something that departs from his prior five-year experience evaluating state-of-the-art AI.
Source: Read the original article
