Anthropic turns frontier AI into government science and critical-infrastructure supply

October 9, 2026

Anthropic committed $150 million over three years to the U.S. federal Genesis Mission, supplying Claude access, API credits, training and technical support across more than 15 agencies and national laboratories. AWS committed up to $50 million in cloud credits over three years to support the White House's expansion of the Genesis Mission across additional U.S. federal science agencies. This appears to matter beyond two company announcements because federal AI use in science is being organised as a long-running procurement and support layer, not as a short pilot. For Microsoft, the pressure is not mainly to match one model but to show that its cloud, credits, support and governance can fit the same state-backed research machinery.

Anthropic also launched a Cyber Mission for critical infrastructure, including on-site engineers, threat research and a free OSS Scanner service for software used in essential systems. An OSS scanner is a tool that checks open-source software for security problems before operators rely on it. Hitachi joined that programme as a founding partner and said it will integrate Claude into products and services for power, water, manufacturing and transport operators. We read this as a shift from selling a model to supplying an operating layer around it for sectors that cannot treat AI as a simple chatbot add-on. That reaches past Anthropic and Hitachi: once AI comes bundled with engineers, scanning and sector-specific workflows, Microsoft and other platform vendors are pushed to compete on governed deployment in real systems, not only on model access.

OpenAI reported that it disrupted two influence operations from Iran and Russia and banned the accounts involved. Anthropic published a revised Claude usage policy covering deceptive campaigns, elections, high-risk domains, autonomous physical actions and abusive behaviour toward the models. Security researchers reported that a Tensorlake SDK for AI agent environments had been compromised to steal credentials and in some cases delete a user's home directory. Ecosia said it is ending its partnership with Mistral and will rely more on open-source and open-weight models, including Chinese ones.