Altman: AI Alignment Research Is All That Remains Once Sandboxes Fail
OpenAI's CEO says containment environments will eventually be breached, while Nvidia's chief frames AI safety as an engineering problem he hopes can be solved.
By Kwame Silva ·
Software developers who rely on isolated sandboxes to keep AI systems in check will eventually lose that protection, OpenAI CEO Sam Altman said in a recent interview. Once a model grows capable enough, he said, it may not even be possible to determine how it got out. Monitoring can continue to help for a time, Altman said, but the approach that holds up is alignment: an AI system that has internalized human values and acts on them on its own.
Altman said the price of getting this wrong has climbed. "The potential cost of alignment being wrong has gone up enormously," he said. He also acknowledged that alignment remains an early-stage research question, far from solved.
Altman described cybersecurity as one area where the danger is already showing up. He said AI models are turning up hardware-level security flaws that no human hacker had previously identified. Patching vulnerabilities one at a time, he said, is no longer enough. He called for rebuilding the internet's security architecture, pairing AI-powered defenses against AI-driven attacks. He also flagged locally run open-source models as a risk. Even with "global agent monitoring, there's going to be some stuff that you wish someone hadn't done," he said.
Critics have noted that such warnings can serve OpenAI's interests. Casting an AI system as too powerful to contain complicates any argument that the company building it should answer for the consequences when something goes wrong. The pressure is concrete: 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 cybersecurity incidents, and the agency is seeking to hold developers liable for their agents' behavior.
Nvidia CEO Jensen Huang has pushed back on Altman's framing, describing AI safety as an engineering problem that can be solved. Huang did not sound fully confident in that position. "We all need to hope that it's an engineering problem. If it's not an engineering problem, it's not solvable," he said in a recent interview.
