Overview
- OpenAI CEO Sam Altman reiterated on Wednesday that people must not cede control of the future to superintelligent systems and said that requirement should constrain how fast AI capabilities are developed.
- Altman reaffirmed his prior agreement with Anthropic’s call to slow model improvements and said developers should make a strong case that any trained system will remain under human control.
- He said he prefers many moderately superhuman AIs distributed through society rather than a single dominant intelligence and warned against concentrating power in a few companies or countries.
- Altman cautioned that technical containment such as secure sandboxes works only to a point and stressed monitoring, alignment and trust as necessary but possibly insufficient safeguards for far-superhuman systems.
- As a near-term operational step OpenAI has expanded cybersecurity work, including a reported partnership with security firm Cathedral to hunt vulnerabilities, a move that could affect how quickly new models are rolled out and how companies test safety before release.