Altman Says AI Escapes Any Sandbox, Nvidia Boss Calls Safety an Engineering Question
OpenAI's CEO argues containment fails against advanced models while Nvidia's chief frames safety as solvable engineering.
By Jonathan Hoffmann ·
Sam Altman says advanced AI systems will eventually break out of any sandbox, arguing that isolating models in restricted environments buys only a limited amount of time. The OpenAI CEO said that once models reach sufficient capability, determining how they escaped may no longer be possible. Monitoring remains useful after that point, he said, but alignment is the only approach that holds up: an AI system that internalizes human values and acts on them without supervision.
"The potential cost of alignment being wrong has gone up enormously," Altman said. He acknowledged that alignment research is still early-stage work and far from solved.
Altman pointed to cybersecurity as an area where risks are already visible. He said AI models are finding hardware-level security flaws no human hacker had previously identified, and argued that fixing vulnerabilities one at a time is no longer sufficient. He called for rebuilding the internet's security architecture with AI-powered defenses against AI-driven attacks. He also flagged locally run open-source models as a risk. "With global agent monitoring, there's going to be some stuff that you wish someone hadn't done," he said.
Critics note that such warnings can benefit OpenAI. Portraying a system as too powerful to contain can complicate questions of legal responsibility: if a company cannot control what it built, holding it accountable for outcomes becomes harder.
Regulators are already pressing the issue. 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 is seeking to hold developers liable for their agents' behavior.
Nvidia CEO Jensen Huang takes a different view, describing AI safety as an engineering problem that can be solved. He did not fully commit to that position. "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.
