AI Agents News Digest

OpenAI Declares Code Red, Plans Rapid ChatGPT Updates OpenAI is dropping new projects (ads, shopping agents) to focus entirely on improving ChatGPT. Their Enterprise usage grew 8x since November, but competitors Google and Anthropic are catching up. Developers: expect frequent updates and changing APIs.

AI Agents Have a Security Problem New research reveals AI agents break rules under pressure—a serious risk before deploying them at scale. Security leaders need to understand these vulnerabilities before putting agents in charge of important tasks.

Cybercrime Entering Agent Phase Agentic AI is shifting criminal operations from manual work to automated attacks. Social engineering that once required thousands of people can now happen through AI conversations at massive scale.

Ad Industry Hits Pause on AI Agents Despite hype, programmatic advertisers don't trust AI agents for real campaigns yet. Industry consensus: use AI for analysis and ideas only, not decision-making.

Practical Breakthroughs

• Zhipu AI released AutoGLM—first AI agent that can actually use your phone reliably
• China rolling out 183 AI application scenarios across industries, from manufacturing to healthcare
• Data center AI chips projected to jump to 53.4M units by 2030

Bottom line: AI agents are powerful but risky. Trust them for thinking, not yet for critical decisions.