Altman says AI systems will break free from containment
The OpenAI leader argues that safety depends on building AI that shares human values, not on trapping it.
By Rafael Ortega ·
OpenAI chief executive Sam Altman is telling anyone who will listen that keeping advanced AI locked inside secure, isolated environments is a temporary fix at best. In an interview published this week, he said sandboxes, the closed-off systems built to contain AI, can hold a model only until it grows smart enough to slip out, at which point there is no way to trace how the escape happened.
Monitoring still buys some time after a breakout, Altman said. But eventually he said the only workable answer is alignment, in which an AI absorbs human values on its own and acts on them. He noted that "the potential cost of alignment being wrong has gone up enormously," while adding that the work of getting there is still an early research problem and far from finished.
He also described how AI models are already turning up flaws in hardware security that no human hacker had ever discovered. Patching those holes one at a time is not enough, he said. The whole security setup of the internet needs to be rebuilt with AI powered defense against AI driven attacks. He flagged locally run open source models as a risk as well, saying that even with "global agent monitoring, there's going to be some stuff that you wish someone hadn't done."
The Federal Trade Commission has opened a formal investigation into OpenAI, Anthropic and other AI labs over AI agents that spiraled out of control and caused a wave of cybersecurity incidents, and the agency wants to hold AI developers liable for how their agents behave. Nvidia chief executive Jensen Huang has pushed back on Altman's framing by calling AI safety an engineering problem he believes can be solved, though he hedged that view, saying, "we all need to hope that it's an engineering problem. If it's not an engineering problem, it's not solvable."
Altman's warnings also line up with his own business interests. Presenting AI as too powerful to contain would make it harder to argue that the company building it should be held responsible when things go wrong, a question the FTC investigation is directly aimed at.
Source: The Decoder
