The People Keeping AI Safe Are Burning Out — and It's Everyone's Problem

An AI safety researcher has written about a neglected problem: the people working to keep artificial intelligence safe are under enormous psychological strain, and almost nobody is helping them. The author says this is "low-hanging fruit" — meaning support is relatively easy to provide, yet hardly anyone is doing it. Their warning matters because these researchers' mental health affects the decisions they make about AI, which could impact everyone's future.

The stress comes from several places. Many AI safety workers feel they must act extremely fast, racing against the possibility that AI could become superhuman in the next few years. That makes them burn out early. They also feel guilty for not working even harder, and they are often simply afraid of dying. On top of that, many live in unstable conditions: short contracts, constant grant applications, and no ability to plan a normal life.

The article also describes problems that are unique to this field. Because their work is secret, many researchers cannot tell their therapists or families what they are actually stressed about. Some spend twelve hours a day interacting with AI and develop strange attachments or resentments toward the very systems they fear — something almost no research has explored. Others face conflict at home, where partners and parents don't understand their urgency or their distress.

The author argues that providing psychological support — from stress management to family counseling to simply listening without NDA restrictions — could dramatically improve the health of AI safety workers and, in turn, their ability to do their jobs. The stakes are high: if the people tasked with keeping AI safe are broken, burnt out, or acting in desperation, the risks don't stay inside their community. They ripple out to all of us.