In a strategic move to fortify its defense against the potential hazards posed by artificial intelligence, OpenAI has initiated a significant expansion of its internal safety protocols. The introduction of a novel "safety advisory group" positioned above technical teams is set to play a pivotal role in offering recommendations to the leadership. Simultaneously, the board has been endowed with the power to exercise veto authority, an intriguing development that raises questions about its practical implementation.While the intricacies of such policies typically remain confined to closed-door meetings with limited external insights, recent internal upheavals and the burgeoning discourse on AI risks prompt a closer examination of OpenAI's safety-centric approach. Documented in a comprehensive update titled the "Preparedness Framework," the company delineates a meticulous process for identifying, analyzing, and addressing catastrophic risks inherent in the models under development.

The framework distinguishes between in-production models, governed by the "safety systems" team, and frontier models, overseen by the "preparedness" team. Additionally, a "superalignment" team is focused on establishing theoretical guide rails for potential "superintelligent" models. The risk evaluation process encompasses four key categories: cybersecurity, persuasion (e.g., disinformation), model autonomy, and CBRN threats (chemical, biological, radiological, and nuclear).Notably, the framework details specific risk levels, ranging from medium to critical, and establishes stringent deployment and development criteria based on these assessments. Beyond the technical evaluation, OpenAI acknowledges the need for a broader perspective on decision-making. Consequently, the company has instituted a "cross-functional Safety Advisory Group" tasked with reviewing technical reports and providing recommendations, ensuring a holistic evaluation of potential risks.

Crucially, recommendations from this group will be simultaneously forwarded to both the board and leadership, comprising CEO Sam Altman and CTO Mira Murati. Although the leadership retains the authority to make final decisions on deployment or discontinuation, the board possesses the prerogative to reverse such decisions, introducing an additional layer of checks and balances to the process.This new safety framework aims to prevent instances where high-risk projects proceed without the board's awareness or approval, as reportedly occurred in prior instances. However, questions linger about the board's willingness to counter expert recommendations and exercise its veto power, with transparency on such decisions remaining a notable aspect yet to be addressed fully by OpenAI. As the AI landscape evolves, the effectiveness of OpenAI's safety measures will be under scrutiny, raising concerns about potential unknown risks that may emerge in this rapidly advancing field.