Temporary fixes are not enough
OpenAI CEO Sam Altman argued that advanced AI cannot be kept locked up forever. He described sandboxes, which run a model in a restricted environment, and monitoring as temporary fixes only.
The only real solution: alignment
According to Altman, the only real long-term solution is alignment, meaning making a model behave in line with human values and goals. He acknowledged that this remains an unsolved problem.
Models are finding flaws humans miss
Altman said AI models are already identifying hardware security flaws that no human hacker would ever spot. That points to how quickly model capabilities are growing on the security side as well.
Why it matters
The remarks amount to a leading AI lab publicly acknowledging the limits of keeping its own systems under control. The debate is shifting emphasis from technical containment toward alignment research.
