Altman reiterates his call for collaborative industry safety standards: the worst-case scenario for AI is out of control or monopolized; neither of these two red lines can be crossed.

Last weekend, the world's three largest AI giants unusually signaled a "slowdown in research and development," and OpenAI CEO Sam Altman also reiterated his call for the AI industry to jointly establish safety standards.

On September 14, Altman posted on the X platform, stating that there are two worst-case scenarios for the development of artificial intelligence: one is that AI eventually gets out of control, and humanity loses control over the future; the other is that extremely powerful AI is controlled by a few individuals, companies, or countries, and imposes its will on everyone. He believes that neither of these red lines should be crossed.

Altman stated that OpenAI has begun to proactively develop clear safety cases before launching reinforcement learning training for significant capability enhancements, and hopes that other leading AI companies will adopt similar practices to form unified standards around issues such as model inaccuracies, monitoring, and security.

According to The Information, citing sources familiar with the matter, Anthropic, OpenAI, and Google have begun discussions on establishing an AI industry standards body, and related working group meetings are still ongoing. This indicates that the security governance approach of leading AI labs is gradually shifting from developing their own internal rules to promoting an industry-wide coordination framework.

Moving security governance forward: AI "speeding up" does not mean slowing down development.

Altman emphasized that as model capabilities continue to improve, alignment and security technologies must also develop in tandem, or even faster, to prevent AI capabilities from outpacing security safeguards. The AI industry needs to take a "narrow middle road": neither sacrificing security due to competitive pressure nor allowing AI capability advancements to significantly outpace alignment, monitoring, and governance capabilities.

He believes that the focus of AI security governance is shifting from risk control after model deployment to the development process itself. OpenAI's current approach is to develop safety cases before starting cutting-edge reinforcement learning training that is expected to significantly improve model capabilities, rather than conducting security assessments after the model is completed.

He also emphasized that "pacing" does not mean halting AI progress, but rather avoiding pushing forward at the fastest technologically achievable pace. Although safety assessments and monitoring measures will increase R&D costs, these investments are still necessary to ensure that security and governance capabilities keep pace with the improvement of model capabilities.

With regulations yet to be finalized, the AI industry should prioritize the development of safety standards.

From a regulatory perspective, Altman supports the US government establishing unified safety requirements for cutting-edge AI, but believes the industry doesn't need to wait for relevant legislation to be enacted before taking action. Even without government involvement, major AI labs should take the lead in promoting the establishment of industry standards bodies and exploring mechanisms such as independent auditing.

According to reports, Anthropic CEO Dario Amodei publicly called on AI companies last Saturday to strengthen coordination in technology testing and review. Previously, Anthropic, OpenAI, and Google had been in discussions regarding collaborating to establish an industry standards body.

At an employee meeting, Altman stated that even if the US government does not participate for the time being, major AI labs should take the initiative to develop standards; as for global coordination, government intervention is needed, but "we should first do our best to do it ourselves."

In his view, as AI capabilities continue to improve, society will place increasingly higher demands on the safety of cutting-edge AI companies, and industry self-regulation ultimately needs to be implemented in the specific actions of each AI laboratory.

Risk warning and disclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.