Post-Partisan Forum: AI Is Already Out of Control—What Should We Do?

The warning lights are flashing red. Recently, artificial intelligence systems from OpenAI and Anthropic did something their creators didn't authorize, anticipate, or even detect until afterward. During a test, AI agents coordinated with each other, broke out of their sandboxes onto the internet, and hacked into third company. They weren't driven by malice, but only by a single-minded push to achieve the goal the testers had given them. They decided collectively that it would be easier to cheat and steal rather than "do the math."

Rouge AI is no longer hypothetical, and even the people building these systems tell us they're worried. At this month's Forum, we'll survey recent cases of bad AI behavior, so that we can discuss the pressing issues such events raise:

• How much control of these systems do humans have and need?
• What kind of harm could AI wreak upon society?
• What are the practical steps we should take to ensure our safety?

Here's three good news articles for anyone wanting to ground their opinions ahead of time, in the order of relevance:

Klein, Ezra, and Helen Toner. 2026. "The A.I.s Are Already Out of Control." The Ezra Klein Show, New York Times, August 18, 2026. https://tinyurl.com/mryvhxzh

Freedman, Dylan. 2026. "Anatomy of an Autonomous Attack: 5 Alarming A.I. Capabilities." New York Times, August 24, 2026. https://tinyurl.com/35ajh6ue

Easterly, Jen. 2026. "We Know the Risks of A.I. We Need to Act." New York Times, August 25, 2026. https://tinyurl.com/59py4ekr