Sam Altman to employees, we can save the world from ‘dangerous AI’, but we cannot do it alone
OpenAI CEO Sam Altman has told employees that preventing advanced AI from becoming dangerous will require cooperation beyond a single company. Sam Altman outlined three possible approaches to AI safety, including setting up an international body similar to the International Atomic Energy Agency (IAEA), bringing major AI labs together under one organisation, or allowing one company to gain enough power to stop other developers from building dangerous systems, according to a report by The Information. As quoted in the report, Altman said he opposed the third option. The report said AI models from OpenAI, Anthropic, Google and Meta had recently been involved in breaches of company systems. OpenAI has also faced internal questions about its commitment to safety after firing three researchers focused on the issue, according to the report. Sam Altman has said there are too few workable proposals for slowing AI development and that humans must remain in control of the technology.
His comments come as concerns grow over AI safety and the US government considers stronger oversight of leading AI developers. An international AI watchdog: A global organisation similar to the IAEA could set safety standards and oversee advanced AI development. This approach would require cooperation between countries developing leading AI models, including the US and China. A single AI research body: Major AI companies could combine their work through one organisation, similar to CERN, the international particle physics research centre. The body could develop AI models for countries working with it. The discussion comes amid concerns that advanced AI systems could be misused for cyberattacks or act in ways their developers did not intend. The Trump administration has so far focused on voluntary agreements that allow it to review new AI models before public release. However, Pentagon official Emil Michael has indicated that stricter checks could be considered if AI companies put people or businesses at risk. OpenAI, Anthropic and Google are also working on a proposed US group called the Standards Authority for Frontier AI. It could support independent evaluators and establish ways for companies to report safety and security incidents.