OpenAI Drives Global Standards for Advanced AI

What happened?

OpenAI published a blog post titled Helping build shared standards for advanced AI, announcing its commitment to creating shared standards for advanced artificial intelligence. The initiative is part of a collaboration with the Appia Foundation, an organization dedicated to fostering global cooperation in safe AI development. According to the statement, OpenAI will support defining evaluation frameworks, safety practices, and technical standards that can be adopted by governments, companies, and academia.

The announcement does not specify concrete details about implementation mechanisms but aligns with previous calls from experts and international organizations to establish common norms that mitigate risks associated with increasingly powerful AI systems. This move comes in a context where AI regulation advances at different speeds: while the European Union is finalizing its AI Act, classifying systems by risk level, the United States has opted for a more voluntary approach through executive orders and the creation of the AI Safety Institute (AISI). Regulatory fragmentation creates uncertainty for companies operating globally, and OpenAI seeks to position itself as a key player in defining standards that could influence future legislation.

Why is it important?

The lack of shared AI standards has led to regulatory fragmentation that hinders interoperability and safety. Each country and company develops its own criteria, which can lead to legal gaps and a race to the bottom in regulation. OpenAI, as a leader in developing large language models (LLMs), has the ability to influence the direction of these norms. The company has been a pioneer in implementing safety measures, such as using RLHF (Reinforcement Learning from Human Feedback) to align its models, and now seeks to scale those practices globally.

Additionally, the initiative aims to address key concerns such as transparency, fairness, and accountability. By promoting shared standards, it is expected to reduce algorithmic biases, improve model explainability, and facilitate independent auditing. This is especially relevant in sectors like healthcare, finance, and justice, where automated decisions have high social impact. For example, in healthcare, a common standard for validating AI-based diagnoses could prevent fatal errors, while in finance, it could ensure that credit algorithms do not illegally discriminate. The experience of past events, such as the bias crisis in facial recognition systems, underscores the urgency of having shared frameworks.

Expected consequences

OpenAI's participation in the Appia Foundation could accelerate the adoption of global standards, but it also raises questions about the company's market power. Critics point out that OpenAI may be seeking to shape rules in its favor, consolidating its dominant position. On the other hand, if standards are truly open and collaborative, they could benefit startups and independent developers by providing a clear framework for innovation. A similar precedent occurred with Facebook's Open Compute Project, which standardized data center hardware and reduced costs for the entire industry, though it also consolidated Facebook's leadership in that area.

In the short term, we are likely to see the publication of technical documents and evaluation proposals by the Appia Foundation. In the long term, these standards could be incorporated into national and international legislation, such as the future EU AI Act. However, success will depend on the foundation's ability to involve diverse stakeholders, including civil society and developing countries, which are often excluded from these conversations. Moreover, the initiative faces the challenge of staying agile in the face of rapid technological evolution; standards that are too rigid could quickly become obsolete, while those too flexible could be ineffective.

What readers should know

• Not binding: For now, this is a voluntary initiative; there is no legal obligation to comply with these standards. This limits its immediate impact but lays the groundwork for future regulations.
• Appia Foundation: It is a relatively new organization; its legitimacy and convening power are yet to be proven. Founded in 2023, it does not yet have the track record of other entities like the Partnership on AI.
• OpenAI is not alone: Other companies like Google DeepMind and Anthropic have also shown interest in shared standards, though they are not mentioned in this announcement. In fact, Anthropic has proposed its own safety evaluation framework, and DeepMind has participated in initiatives like the Institute for Ethical AI.
• Risk of regulatory capture: There is a possibility that big tech companies define standards that benefit them, excluding smaller players. To mitigate this, the Appia Foundation should ensure multistakeholder governance, including regulators, academics, and civil society representatives.

“Creating shared standards is a necessary step, but it must be an inclusive process involving civil society, academia, and governments, not just companies.” — TheVortiq

In conclusion, this announcement represents a significant advance in the search for global AI norms, but its success will depend on the transparency of the process and the willingness of OpenAI and other actors to cede control for the common good. The history of technology regulation, from the internet to biotechnology, shows that voluntary standards can be an effective first step, but they often require legal backing to be truly effective. The coming months will be crucial to observe how this collaboration evolves and whether it translates into concrete actions.