Daily AI ·
US Partially Lifts Anthropic Mythos 5 Ban; Government Gating Becomes New Norm
The US government granted Anthropic permission to reinstate its Mythos 5 model for critical-infrastructure defenders, replacing a blanket export ban with controlled access for over 100 US organizations. Separately, Google DeepMind lost six top researchers to rivals, and ex-Anthropic founders raised $200M for a self-improving AI startup. OpenAI and Microsoft also joined the UK's AI alignment coalition.
7 sources1 min read
US approves limited return of Anthropic Mythos 5 for critical infrastructure The US government authorized Anthropic to restore access to its most advanced cybersecurity model, Mythos 5, for roughly 100 US critical-infrastructure organizations, including Fortune 500 companies. This partial lift follows a June 12 block and signals a shift from blanket export bans to a government-determined access policy; dialog continues on restoring Fable 5. 12
Google DeepMind loses six top researchers to Meta, OpenAI, and Anthropic Since February, six leading DeepMind researchers have departed, including reasoning pioneer Denny Zhou (Meta Superintelligence Lab), Nobel laureate John Jumper (Anthropic), and security expert Dawn Song (Meta). The exodus is linked to DeepMind's strategic pivot away from core reasoning research, handing rivals critical talent in AI alignment, coding, and safety. 34
Ex-Anthropic researchers raise $200M seed for self-improving AI startup Mirendil Mirendil, founded by Behnam Neyshabur and Harsh Mehta who left Anthropic in December 2025, raised $200M at a $1B valuation from a16z, Kleiner Perkins, and Nvidia. The startup aims to commercialize the AI-to-improve-AI loop that frontier labs keep in-house, packaging it as a platform for scientific research. 5
OpenAI and Microsoft join UK's international coalition for AI alignment The UK's AI Security Institute (AISI) announced that OpenAI and Microsoft have pledged £5.6M to the Alignment Project, joining a global effort led by Yoshua Bengio and Zico Kolter to ensure advanced AI remains controllable. The project now has over £32M in funding and 60+ research initiatives. 6
HP and OpenAI form strategic partnership to embed AI in enterprise workflows HP is integrating OpenAI's Frontier platform across customer service, software development, and employee productivity as an 'operating layer.' The partnership includes developing specialized edge-AI hardware for continuous on-device inference, marking a shift from cloud-based generative AI to agentic capabilities. 7
Daily AI
The most important AI & large language model developments — sourced and cited every morning.