AI MODELS BY OPENAI AND ANTHROPIC TO UNDERGO SAFETY TESTING BEFORE PUBLIC RELEASE

Authored by: Ms Tanima Bhatia

Key Highlights:

Groundbreaking AI Safety Collaboration: OpenAI and Anthropic have partnered with the U.S. AI Safety Institute to subject their upcoming AI models to rigorous safety testing before public release, marking a noteworthy shift toward prioritizing safety and ethical standards in AI development.
Government and Industry Cooperation: The U.S. AI Safety Institute, operating under the National Institute of Standards and Technology (NIST), will gain early access to AI models from both companies, allowing for thorough evaluation and feedback to ensure safer AI technologies.
Global Implications for AI Governance: This collaboration could influence global AI regulation, encouraging other countries to adopt similar safety assessments for AI models, as seen with recent legislative efforts in California and earlier advisories in India.

In a groundbreaking move, leading AI companies OpenAI and Anthropic have agreed to subject their upcoming AI models to rigorous safety testing before releasing them to the public. This collaboration, announced on August 29, 2024, marks the first-time major tech companies have partnered with a government body for such a comprehensive evaluation process, reflecting a significant shift toward ensuring that AI advancements are aligned with public safety and ethical standards.

A New Era of AI Safety

The agreement was enforced between OpenAI, Anthropic, and the U.S. AI Safety Institute, which operates under the National Institute of Standards and Technology (NIST), a division of the U.S. Department of Commerce. This initiative stems from an executive order issued by President Joe Biden in October 2023, mandating safety assessments for AI models among other critical regulations. The goal is clear: to ensure that AI technologies are not only innovative but also safe and reliable for public use.

Access and Collaboration

The partnership will provide the U.S. AI Safety Institute with early access to new AI models from OpenAI and Anthropic, both before and after their public rollout. This access is crucial for conducting thorough evaluations of the models’ capabilities and potential safety risks. The Institute will work closely with these companies to provide feedback on safety improvements, contributing to the development of safer AI technologies.

Elizabeth Kelly, the director of the U.S. AI Safety Institute, emphasized the importance of this collaboration, stating, “Safety is fundamental to enabling breakthrough technological innovation. With these agreements, we look forward to initiating our technical collaborations with Anthropic and OpenAI to advance the science of AI safety.” This collaboration underlines the need for responsible governance in the fast-paced field of artificial intelligence.

Industry Leaders Speak Out

Sam Altman, CEO of OpenAI, expressed his support for the initiative, highlighting the importance of national-level regulations in the AI sector. “We are happy to have reached an agreement with the U.S. AI Safety Institute for pre-release testing of our future models,” Altman said. He added that this move is critical as the U.S. continues to lead the global AI narrative, ensuring that innovation does not come at the expense of safety.

Similarly, Jack Clark, co-founder of Anthropic, echoed these sentiments, emphasizing the importance of collaboration with the U.S. AI Safety Institute. He noted that this partnership “leverages their wide expertise to rigorously test our models before widespread deployment,” strengthening the company’s ability to identify and mitigate risks effectively, advancing responsible AI development.

Implications for the Future

This unprecedented move could pave the way for other countries, including India, to implement similar safety evaluations for AI models before they are released to the public. Earlier this year, the Indian government faced backlash after issuing an advisory that required untested AI models to receive explicit approval before being made available to users, although now rectified the advisory requires such models to inform the users by labeling the possible inherent unreliability of the output generated by such models. This recent development in the U.S. could influence global AI governance, encouraging a more cautious and responsible approach to AI deployment.

The collaboration between OpenAI, Anthropic, and the U.S. AI Safety Institute also comes at a time when AI regulations are becoming a focal point in various regions. In California, lawmakers recently passed legislation that would mandate safety tests for AI models exceeding certain costs or computing power. This bill, if approved, could add another layer of oversight for AI companies operating in the state, further highlighting the growing emphasis on AI safety.

Conclusion

The partnership between OpenAI, Anthropic, and the U.S. AI Safety Institute represents an enormous step toward integrating safety and ethical standards into the AI development process. As these companies share their models for testing and feedback, they are not only advancing the science of AI safety but also helping to restore public trust in AI technologies. This initiative signals a shift in the industry, where innovation is balanced with human oversight and safety, setting a new standard for responsible AI development in the future.

References: