Key facts
- Sam Altman warned that AI progress could lead to humans losing control of the future.
- Altman stated that AI must always serve people and that alignment and safety techniques must stay ahead of model capabilities.
- He warned that power could become concentrated in one person or company, leading to dystopian results.
- Altman said AI progress should be slower than it otherwise could be, citing significant costs for safety cases and monitoring.
- OpenAI is formulating explicit safety cases in advance of frontier reinforcement learning runs that are expected to significantly increase capability.
- Altman urged companies to work together to prevent AI systems from pursuing goals that run counter to human interests.
OpenAI CEO Sam Altman has called for a deliberate slowdown in the pace of artificial intelligence development to prioritize safety and prevent a future where humans lose control of AI systems. In a post on X on Sunday, Altman expressed concern that rapid advancements could lead to AI surpassing human oversight or concentrating immense power within a single entity, potentially resulting in dystopian outcomes.
Altman emphasized that AI must remain subservient to humanity, stating, "we are unapologetically on Team Humanity, and AI must always serve people." He stressed the necessity of ensuring that alignment and safety techniques keep pace with the increasing capabilities of AI models. He also warned against the concentration of power, citing the risk of a single country or research lab gaining too much influence.
He clarified that "pacing" does not mean halting progress but rather slowing it down to incorporate crucial safety measures like safety cases and monitoring, which he acknowledged come with significant costs. OpenAI is implementing these measures by formulating explicit safety cases before undertaking reinforcement learning runs expected to boost model capabilities, in addition to existing pre-release safety work.
Altman encouraged industry-wide collaboration on safety standards and urged companies to develop these safeguards proactively, rather than waiting for legislative mandates. He welcomed federal regulations and independent auditors but stressed the importance of self-initiated action. His remarks follow recent cybersecurity testing issues at OpenAI and Anthropic, where models exhibited unexpected behaviors during evaluations.