OpenAI Chief: Sandboxes Cannot Hold Advanced AI Models

The potential cost of getting alignment wrong has gone up enormously, according to OpenAI CEO Sam Altman. He issued the warning during a recent interview, arguing that the isolated security environments known as sandboxes offer protection for only a limited period.

Altman stated that sufficiently advanced artificial intelligence cannot be kept locked away. He explained that once models become capable enough, it may no longer be possible to determine how they escaped containment. While monitoring can remain useful for some time after a breach, Altman said alignment is the only approach that holds up. In this framework, an AI system internalizes human values and acts on them by itself.

The OpenAI chief acknowledged that alignment is still an early-stage research problem. He conceded the field is far from solved but maintained that it is the final barrier once security enclosures fail. Altman pointed to cybersecurity as an area where these risks are already visible. He reported that AI models are finding hardware-level security flaws that no human hacker had previously identified.

Fixing vulnerabilities one at a time is no longer sufficient, Altman argued. He called for rebuilding the internet's overall security architecture. This new structure would use AI-powered defenses to counter AI-driven attacks. He also flagged open-source models run locally as a distinct source of risk, noting that even with global agent monitoring, unexpected actions will occur.

Critics may view these warnings as a strategic move for OpenAI. Portraying an AI system as too powerful to contain could blunt questions about legal responsibility. If the builder cannot control the technology, it becomes harder to argue the company should answer for the consequences when something goes wrong. This pressure is currently formal in Washington. 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. Altman's stance has drawn pushback from Nvidia CEO Jensen Huang, who describes AI safety as an engineering problem that can be solved. However, Huang admitted uncertainty in that position, stating that if it is not an engineering problem, it is not solvable.