OpenAI CEO Sam Altman has outlined three possible approaches to controlling advanced AI, including international oversight, cooperation between leading laboratories and concentrated computing power.
OpenAI CEO Sam Altman has outlined three potential approaches to managing the risks posed by increasingly powerful artificial intelligence (AI) systems. His proposals include establishing an international oversight organisation, encouraging cooperation among leading AI developers and a scenario involving one developer gaining enough economic power to restrict competitors’ access to computing resources.
Altman believes OpenAI must determine which approach it supports within the next year. The discussion comes as concerns grow over AI security, the potential misuse of advanced models and the need for stronger safeguards across the technology industry.
As AI systems become more capable, policymakers and technology companies are exploring ways to prevent their misuse and manage risks that could extend beyond individual organisations.
Altman has identified three possible approaches that could shape the future of AI governance. Each represents a different way to manage powerful AI technologies, from international regulation to cooperation between companies and the concentration of computing resources.
The proposals raise important questions about how AI developers should be supervised, who should establish safety standards and how the industry can balance innovation with public protection.
According to the provided report, Altman told OpenAI employees that the company should decide within the next year which direction it wants to support.
Altman’s first proposal involves establishing an international organisation responsible for overseeing advanced AI systems. The concept draws a comparison with the International Atomic Energy Agency (IAEA), which works with countries on nuclear safety and safeguards.
A similar organisation for AI could help establish common international standards, coordinate safety assessments and monitor compliance with agreed rules.
Because advanced AI technologies can be developed and deployed across national borders, international cooperation could help governments address risks that cannot be managed effectively by one country alone.
However, creating such an organisation would require agreement on its authority, responsibilities and enforcement powers. Countries would also need to determine how sensitive technical information could be shared without compromising national security or commercial confidentiality.
The effectiveness of this approach would ultimately depend on participation by major AI developers and governments.
Altman’s second approach involves greater cooperation among leading AI companies. Instead of allowing organisations to address every major safety challenge independently, major developers could coordinate their efforts to manage the risks associated with advanced AI.
Companies such as OpenAI, Google, Anthropic and Meta are developing increasingly sophisticated AI models. Cooperation could help them share safety practices, identify common vulnerabilities and develop consistent standards for testing powerful systems.
A coordinated approach could also reduce duplication in safety research and help establish clearer expectations for responsible AI development.
Nevertheless, cooperation between competitors presents challenges. Companies may be reluctant to share proprietary technology, confidential research or commercially sensitive information. Differences in business strategies and views about acceptable risks could also complicate collective decision-making.
Any joint initiative would therefore need clear governance, accountability and safeguards against anti-competitive behaviour.
The third scenario involves a single AI developer acquiring enough economic influence to restrict competitors’ access to computing resources.
Advanced AI development depends on substantial computing infrastructure, including specialised processors, data centres and large-scale energy resources. Control over these resources can influence which organisations are capable of training and operating advanced models.
In theory, a developer with overwhelming economic power could use its position to limit the ability of rivals to build powerful AI systems.
However, this approach raises serious concerns about market concentration, competition and the distribution of technological power. Restricting access to computing resources could disadvantage smaller companies and research organisations while placing substantial influence in the hands of one entity.
The supplied report indicates that Altman is not in favour of this scenario, distinguishing it from the other approaches under consideration.
The discussion comes amid increasing attention to security risks associated with advanced AI systems. Major developers, including OpenAI, Google, Anthropic and Meta, face growing pressure to ensure their technologies cannot easily be misused or exploited.
Potential concerns include unauthorised access, information leaks, misuse of powerful AI capabilities and failures in safeguards designed to prevent harmful outcomes.
One initiative mentioned in the report is the proposed Standards Authority for Frontier AI, which several leading AI laboratories are working towards establishing. The organisation is expected to launch by early 2027, according to the supplied information.
A common standards framework could help improve consistency in how advanced AI systems are evaluated. However, its practical impact would depend on the standards adopted, the organisations participating and the mechanisms available to assess compliance.
Alongside industry-led initiatives, policymakers are also considering stronger regulatory measures to address potentially dangerous AI capabilities.
The report refers to a draft bill that could allow the US Commerce Department to halt AI models considered capable of causing large-scale harm.
Such a proposal would represent a more direct government intervention in AI development. Depending on the bill’s final wording, it could provide officials with additional authority to respond to models that pose serious risks.
However, questions would remain about how harmful capabilities would be defined, what evidence would be required and how decisions could be reviewed.
The balance between preventing serious harm and preserving technological innovation is likely to remain a central issue in debates over AI regulation.
Altman’s three approaches illustrate the different ways governments and technology companies might attempt to manage advanced AI.
International oversight could encourage consistent rules across borders, cooperation between developers could improve shared safety practices, and concentrated computing power represents a more controversial scenario involving economic dominance.
Each approach comes with potential benefits and significant limitations. A workable framework may ultimately require a combination of technical safeguards, independent evaluations, government oversight and international cooperation.
OpenAI’s decision about which approach to support could help shape its future position on AI governance. The wider challenge will be developing rules that reduce serious risks without unnecessarily limiting competition or beneficial innovation.
Conclusion
Sam Altman’s proposals highlight the growing urgency of establishing effective safeguards for advanced artificial intelligence. International oversight, cooperation between leading AI laboratories and the concentration of computing resources offer very different approaches to managing potential risks.
As industry initiatives and proposed legislation develop, the effectiveness of these measures will depend on transparent standards, accountability and meaningful oversight. The choices made by OpenAI and other major AI developers could influence how advanced AI is governed in the years ahead.