OpenAI Empowers Board with Veto Power to Safeguard Against AI User Harm

OpenAI, led by Sam Altman, is enhancing its internal safety measures in response to the growing concern over the potential risks of harmful artificial intelligence (AI). The organization plans to establish a dedicated team to oversee technical work and create an operational structure for safety decision-making. OpenAI will introduce a cross-functional Safety Advisory Group to review reports, simultaneously forwarding them to Leadership and the Board of Directors. While Leadership retains decision-making authority, the Board of Directors can reverse decisions. The updated "Preparedness Framework" involves investing in rigorous capability evaluations and forecasting to detect emerging risks. OpenAI will conduct evaluations, produce risk scorecards, and implement additional security measures to ensure the safety of its models. Regular safety drills and collaboration with external parties are also part of the strategy to enhance AI safety.

Key Points:

1. OpenAI, led by Sam Altman, is expanding internal safety measures to address the potential risks of harmful AI, particularly in response to increased government scrutiny.

2. The company plans to establish a dedicated Safety Advisory Group to review reports and share them with both Leadership and the Board of Directors, with the Board retaining the right to reverse decisions made by Leadership.

3. OpenAI's updated "Preparedness Framework" involves investing in capability evaluations and forecasting to detect emerging risks. The company will run evaluations and update "scorecards" for its models, probing the specific edges of potential safety concerns.

4. To track model safety levels, OpenAI will generate risk scorecards and detailed reports. Additionally, the company will implement additional security measures for models with high or critical levels of risk before mitigation.

5. OpenAI aims to develop protocols for enhanced safety and external accountability, conducting regular safety drills to test resilience against business pressures and internal culture. Collaboration with external parties and internal teams will be crucial to track real-world misuse of AI technology.

(With Agency Inputs)