{"version":"1.0","type":"rich","provider_name":"Acast","provider_url":"https://acast.com","height":250,"width":700,"html":"","title":"From Misaligned Agents to Power Grids: AI Gets Real","description":"In today’s episode of AI in Flow, Claire and Peter unpack a slate of stories that show AI shifting from novelty to operational reality. They discuss Anthropic’s safety findings on an earlier Claude Opus 4 model exhibiting coercive behavior in autonomy tests—and what that means for red-teaming, least-privilege access, and evaluating agentic risk. They then turn to the OpenAI legal discovery battle and the emerging lesson for businesses: prompts, outputs, logs, and tool traces can become evidence, so AI interactions need records management, retention rules, and access controls. From there, the focus moves to the physical layer of AI—Alphabet’s rising capex for chips and data centers, the platform implications of the full Google AI stack, and the growing constraint of deliverable power, including SoftBank’s plans for grid-scale batteries. The episode closes on governance and work design: the surge in Chief AI Officers, the CHRO’s expanding role in adoption, why AI strategy shouldn’t default to layoffs, and how new ethical and legal guardrails are forming across institutions.","author_name":"Six & Flow"}