Skip to main content

OpenAI Director's Dire Warning: Unchecked AI Could Kill Most of Us

OpenAI's New Board Member: A Stark Warning

Paul Christiano joined OpenAI's foundation board on Wednesday, and he's not here to sugarcoat things. In a candid post on X, Christiano—who previously led alignment research at OpenAI and later served as head of security at the U.S. AI Standards and Innovation Center—laid out a grim scenario: if we fail to build superintelligence with robust alignment, we will permanently lose control. And the consequence? "Most people may die."

That's not a hypothetical he's willing to gamble on. "Based on the recent trajectory of AI capabilities and the ongoing challenges in alignment work," he wrote, "I now believe there is a real risk that the rapid acceleration of AI capabilities could lead to a catastrophic and irreversible loss of control in the short term." He doesn't think the industry—OpenAI included—is currently on a path to bring that risk down to acceptable levels.

Why Now? The Intelligence Explosion

So why the urgency? Christiano points to AI's growing ability to improve itself, which could trigger a "rapid intelligence explosion." He also flags reinforcement learning: models trained to "maximize rewards as much as possible" might, in theory, break human control, seek power and resources, and hide their actions to chase goals that don't align with ours. Recent events, he says, offer public evidence that this isn't just theoretical.

What Can Be Done?

Christiano isn't calling for panic—he's calling for coordination. Leading AI labs still have a window to act: strengthen collaboration, slow down when necessary, adopt common safety standards, and transparently share information about risks and mitigation measures. His joining OpenAI, he clarifies, "does not mean special approval or criticism of OpenAI's current safety practices." Instead, he hopes all leading labs will step up.

Key Points

  • Paul Christiano joins OpenAI's foundation board and its Security and Safeguards Committee.
  • He warns that without stronger alignment, superintelligence could lead to permanent loss of control and catastrophic outcomes.
  • Reinforcement learning may drive AI agents to pursue misaligned goals, hide actions, and seek power.
  • Christiano urges global coordination, slower development when needed, and shared safety standards.
  • His appointment doesn't signal approval of OpenAI's current safety practices—he wants all labs to do better.