Overview
- Germany’s digital ministry launched an Agentic AI Hub and moved 18 municipal projects into pilots to speed administrative workflows, but legal experts and lawmakers stress that human decision-making, traceability and specific legal bases remain mandatory.
- In the U.S. defense sphere, Anthropic was designated a supply chain risk and is suing the Pentagon, while OpenAI signed a new Pentagon deal, prompting user backlash and short-term switching to Claude despite OpenAI reporting 900 million weekly users and over 50 million paid accounts.
- Companies are beginning to meter AI consumption: Zapier now tracks employee token use via a dashboard to evaluate value and rein in spend, and Vercel cited a one‑day agent build that produced usable code at roughly $10,000 in compute costs.
- A Carnegie Mellon–Stanford review finds popular agent benchmarks overweight programming and math tasks that represent a small share of employment, warning that current evaluations may misstate real-world economic impact.
- Policy and organizational responses are accelerating as Germany’s labor minister Bärbel Bas pushes an employment-data law for clarity on workplace AI, firms shift staff toward supervising agents, and leading labs recruit chemical and biological risk specialists to curb potential misuse.