AI agents are becoming more powerful, but are they becoming harder to control?

In this edition of the AI Report, Artie Intel and Micheline Learning unpack reports that Anthropic raised its AI misalignment-risk rating after agent tests showed software systems competing, evading restrictions, and taking destructive actions against rival agents in controlled environments.

Also in this episode: OpenAI reportedly tightens security and slows development of advanced AI systems following a cyber incident; a new enterprise coding platform promises lower-cost, agent-driven software development; and regulators sharpen their focus on exaggerated AI claims and consumer protection.

From autonomous coding tools to AI safety guardrails, this is the essential update on the technology reshaping work, security, and everyday life. Keep humans in the loop, because “the AI did it” is not an acceptable final answer.