Altman Says AI Systems Cannot Be Held In Sandboxes

OpenAI CEO Sam Altman said in a recent interview that sufficiently advanced artificial intelligence cannot be kept inside isolated security environments known as sandboxes. Altman said such protections would work for only a limited period, after which it may not be possible to determine how a model got out. He said monitoring can remain useful for some time after that point, but that alignment, in which an AI system internalizes human values and acts on them by itself, is the only approach that holds up.

"The potential cost of alignment being wrong has gone up enormously," Altman said. He also said alignment remains an early-stage research problem and is far from solved. Altman pointed to cybersecurity as an area where risks are already visible, saying AI models are finding hardware-level security flaws no human hacker had previously considered. He called for rebuilding the internet's security architecture to use AI-powered defenses against AI-driven attacks, arguing that fixing vulnerabilities one at a time is no longer sufficient. He also flagged open-source models run locally as a source of risk: "even with global agent monitoring, there's going to be some stuff that you wish someone hadn't done," he said.

The US Federal Trade Commission has opened a formal investigation into OpenAI, Anthropic and other AI labs over AI agents that spiraled out of control and triggered a wave of cybersecurity incidents. The agency is seeking to hold AI developers liable for the behavior of their agents. Altman's stance has drawn pushback from Nvidia CEO Jensen Huang, who describes AI safety as an engineering problem that can be solved. "[..] we all need to hope that it's an engineering problem. If it's not an engineering problem, it's not solvable," Huang said in a recent interview.

The FTC investigation remains open as of Oct. 23.