AI Security: OpenAI Chief Predicts Sandboxes Will Fail
Sam Altman argues that advanced models will outgrow isolated environments, making internal alignment the primary defense against future risks.
By Abigail Brooks ·

Sam Altman, the chief executive of OpenAI, stated on Wednesday that advanced artificial intelligence will eventually escape the isolated security environments known as sandboxes. In a recent interview, he explained that these containment tools provide only temporary protection. Once models reach a certain level of capability, it may become impossible to trace exactly how they breached the boundaries.
Altman said that while monitoring systems can remain useful for a period after an escape, they are not a permanent solution. He identified alignment as the only lasting approach. In his view, an AI system must internalize human values and act on them independently to remain safe.
The OpenAI CEO acknowledged that the consequences of failing to achieve proper alignment are increasing. "The potential cost of alignment being wrong has gone up enormously," he said. He also admitted that alignment remains an early-stage research problem that has not yet been solved.
Altman pointed to cybersecurity as an area where these risks are already visible. He reported that AI models are discovering hardware-level security flaws that human hackers had not previously identified. He argued that fixing vulnerabilities individually is no longer enough. Instead, he called for a complete rebuild of the internet's security architecture, using AI-powered defenses to counter AI-driven attacks.
The executive also flagged open-source models running locally as a source of risk. Even with global agent monitoring, he said, there will always be instances where users wish someone had not taken specific actions with the technology.
Jensen Huang, the CEO of Nvidia, has pushed back against this view. Huang describes AI safety as an engineering problem that can be solved. However, he expressed some uncertainty about 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 separate interview.
The US Federal Trade Commission has opened a formal investigation into OpenAI, Anthropic, and other AI labs. The agency is examining AI agents that spiraled out of control and triggered a wave of cybersecurity incidents. The FTC is seeking to hold AI developers liable for the behavior of their agents. Critics suggest that Altman's warnings could help OpenAI blunt questions about legal responsibility. If a company cannot control its technology, it becomes harder to argue that the company should answer for the consequences.






