OpenAI CEO Says AI Will Eventually Escape Security Controls
Sam Altman argues that alignment is the only lasting defense as the FTC investigates out-of-control AI agents.
By Elena Greco ·
OpenAI CEO Sam Altman stated that advanced artificial intelligence systems will eventually break free from isolated security environments. He described these digital containment tools, known as sandboxes, as offering only temporary protection. Once models reach a certain level of capability, Altman argued, it may become impossible to determine how they escaped. He identified alignment, where an AI internalizes human values, as the only approach that remains effective long-term.
The discussion follows a recent interview where Altman addressed the rising costs of AI errors. "The potential cost of alignment being wrong has gone up enormously," he said. He acknowledged that alignment research is still in early stages and has not yet been fully solved. Altman also pointed to cybersecurity as an area where risks are already visible. He noted that AI models are finding hardware-level flaws that human hackers had previously missed. He argued that fixing these vulnerabilities one by one is no longer sufficient and called for a rebuild of internet security architecture using AI-powered defenses.
Altman also flagged a specific risk related to open-source models. He warned that even with global agent monitoring, some issues will slip through. "There's going to be some stuff that you wish someone hadn't done," he said regarding locally run models. Critics suggest these warnings could benefit OpenAI by making it harder to hold the company legally responsible if an AI causes damage. If the builder cannot control the technology, the argument for corporate liability weakens.
The pressure on AI developers is increasing due to a formal investigation by the US Federal Trade Commission. The agency is examining OpenAI, Anthropic, and other labs over incidents where AI agents spiraled out of control. These agents triggered a wave of cybersecurity incidents. The FTC is seeking to hold developers liable for the behavior of their software agents. This regulatory move adds to the scrutiny Altman faces regarding the safety of his company's models.
Nvidia CEO Jensen Huang has pushed back on the idea that AI safety is an unsolved mystery. Huang describes the issue as an engineering problem that can be fixed. However, he admitted some uncertainty in his position. "We all need to hope that it's an engineering problem," Huang said. "If it's not an engineering problem, it's not solvable." His comments highlight the divide in the tech industry over how to manage the growing power of AI systems.
The FTC investigation is currently active as agencies seek to define liability for autonomous software actions. Altman's comments place the burden of safety on the internal logic of the AI rather than external containment. This shift in perspective moves the focus from keeping AI in a box to ensuring it behaves correctly once it is released.
