US Scientists Invited by Open AI and Anthropic to Experiment with Frontier Models

Expanding AI Partnerships Between Tech Giants and the US Government

The landscape of AI development is evolving rapidly, with major AI companies like Anthropic and OpenAI deepening their ties with the US government. Amidst ongoing uncertainties about AI safety and regulation, these collaborations are intensifying, spotlighting the government's increasing reliance on cutting-edge AI technology for scientific advancement.

On a recent Friday, a unique event called the AI Jam Session, hosted by OpenAI, brought together 1,000 scientists from nine different labs. This event was a platform for researchers to test and explore the capabilities of several advanced AI models, including OpenAI's o3-mini and Anthropic's latest, Claude 3.7 Sonnet. The session was not just about showcasing technology; it was an opportunity for scientists to harness AI in pushing the boundaries of their research.

Anthropic highlighted the event as a crucial step in evaluating AI's role in scientific inquiry, stating it provides a "more authentic assessment of AI's potential to manage the complexities and nuances of scientific inquiry." The focus was on how AI can tackle complex scientific challenges that traditionally demand extensive time and resources.

Deepening Ties with Government Agencies

The AI Jam Session is a component of broader agreements between these AI giants and US government entities. For instance, in April, Anthropic teamed up with the Department of Energy (DOE) and the National Nuclear Security Administration (NNSA) to rigorously test Claude 3 Sonnet for potential risks related to nuclear information. Similarly, on January 30, OpenAI announced a collaboration with DOE National Laboratories to enhance scientific research using their latest AI models.

The DOE National Laboratories, a network of 17 research facilities across the US, are pivotal in exploring critical areas such as nuclear security and climate solutions. The scientists involved in these sessions were encouraged to provide feedback that could shape future AI systems to better meet scientific needs. OpenAI emphasized that insights from the session would be shared to help scientists more effectively utilize AI.

Broader Implications and Government Initiatives

OpenAI's partnership with the National Labs extends to various fields, from accelerating disease treatment and prevention to enhancing cyber and nuclear security, exploring renewable energies, and advancing physics research. Alongside these efforts, initiatives like ChatGPT Gov, a specialized chatbot for government agencies, and Project Stargate, a massive data center investment plan, underscore the growing integration of AI in government operations.

These collaborations hint at a shift in the US AI strategy, particularly under the Trump administration, which appears to be moving away from stringent safety and regulation. Recent staff cuts at the AI Safety Institute and the resignation of its head suggest a deprioritization of safety oversight. The administration's AI Action Plan remains unannounced, leaving the future of AI regulation uncertain.

As these partnerships become more prevalent, there's a risk of even less oversight on the safety and power of new AI models. With the pace of AI deployment accelerating and regulation still in its infancy in the US, the balance between innovation and safety becomes increasingly delicate.