OpenAI grants the board ultimate authority in ensuring the safety of new AI models

OpenAI has released a “Preparedness Framework” document outlining strategies to mitigate the potential risks and worst-case outcomes associated with advanced AI models. The framework focuses on tracking and safeguarding against catastrophic risks, such as cyber disruption and the creation of dangerous weapons using AI technology. In this new framework, decision-making power on the release of new AI models lies with OpenAI’s company leadership, but the board of directors has the ultimate authority and the right to reverse decisions made by the leadership team.

To ensure the safety of AI models, OpenAI has implemented safety checks and a specialized “preparedness” team led by MIT professor Aleksander Madry. This team will monitor and evaluate potential risks, categorizing them on scorecards as low, medium, high, or critical. The framework specifies that only models with a post-mitigation score of ‘medium’ or below can be deployed, and models with a post-mitigation score of ‘high’ or below can be developed further. The framework is currently in beta and will be updated based on feedback.

OpenAI’s governance structure has recently undergone changes following a corporate dispute, raising concerns about the authority of CEO Sam Altman and the board’s restrictions. The board, consisting of three white men, holds the responsibility of ensuring OpenAI’s advanced technology benefits all of humanity. The lack of diversity on the board has been criticized, with some emphasizing the importance of government intervention in regulating AI development and deployment.

These safety measures from OpenAI come amidst ongoing debates about the potential threats of AI. Earlier this year, prominent AI scientists, including OpenAI’s Altman and Google Deepmind’s CEO Demis Hassabis, signed an open letter stating the need to prioritize mitigating the risk of AI-induced extinction. However, some industry observers have accused companies of using catastrophic scenarios to divert attention from present challenges associated with AI tools.