A Surprising Alliance Between Rivals
In the hyper-competitive world of artificial intelligence, OpenAI and Anthropic are titans locked in a race for dominance. That’s what makes the recent joint calls for safety cooperation from their respective CEOs, Sam Altman and Dario Amodei, so significant.
This isn't just a friendly chat between competitors; it's a public acknowledgment of shared risk from two leaders with a complex history. Amodei, along with several other key researchers, left OpenAI in 2021 to found Anthropic, citing concerns that safety was not keeping pace with the rapid development of AI capabilities. This origin story makes their current alignment on safety a powerful signal that the industry's most influential players believe the risks of unchecked advancement are becoming too great to ignore.
The Core of the Cooperation Call
The calls for collaboration, recently echoed at a United Nations Security Council meeting, centre on managing the existential threats posed by increasingly powerful AI. Both Altman and Amodei have warned that as AI systems become more autonomous, humanity could lose control, with potentially catastrophic consequences. Amodei has proposed specific areas for agreement, such as prohibiting the use of AI to develop biological weapons and establishing shared evaluation standards to verify the safety of advanced models before they are released. The goal is to move beyond individual company promises and create a global framework that ensures safety measures evolve as quickly as the technology itself. Just recently, the two companies agreed to give the U.S. AI Safety Institute access to their models for testing, a concrete first step in this direction.
Why Talk Doesn't Equal a Treaty
Despite the unified messaging, turning these calls into a formal, binding agreement presents enormous challenges. A key issue is the inherent tension between cooperation and commercial competition. AI companies are in a race to build the most capable models, and a binding pact to slow down or share sensitive safety research could be seen as ceding a competitive edge. There are also significant legal hurdles, including potential antitrust concerns. In fact, a recent lawsuit argues that an agreement among major AI labs to slow development would be an illegal, anticompetitive practice. Furthermore, drafting a contract that is both specific enough to be enforceable and flexible enough to adapt to rapidly changing technology is a monumental task. The pace of AI innovation often outstrips the typical contracting process, making it difficult to 'future-proof' any agreement.
The Hurdles of a Binding Pact
Creating an effective industry-wide agreement requires consensus on a host of complex issues. Who defines what is 'safe'? Who serves as the independent auditor? And what are the penalties for non-compliance? An agreement would need to establish clear, measurable standards for safety evaluations, rules for reporting incidents, and a governance body to oversee the pact. Companies would need to grant unprecedented access to their proprietary models for third-party testing, raising intellectual property and data privacy concerns. Liability is another major sticking point; vendors are often reluctant to accept accountability for AI failures, a stance that would need to be addressed in any meaningful agreement. Getting global buy-in, particularly from nations that may see an opportunity to surge ahead if Western companies slow down, adds another layer of geopolitical complexity.
A Push to Pre-empt Regulation
This push for industry cooperation can also be seen as a strategic move to shape the future of AI governance. With governments around the world actively considering new regulations, the AI industry has a vested interest in demonstrating it can police itself. By proactively forming bodies like the proposed Standards Authority for Frontier AI (SAFA), companies like OpenAI, Anthropic, and Google hope to create a self-regulatory framework. This could allow them to set the terms of safety standards rather than having potentially more restrictive rules imposed upon them by lawmakers. The message is clear: the industry acknowledges the risks and wants to be a partner in creating the solution, not simply a subject of regulation.
