Racing Towards AI Regulation: Lessons from the Past

Historical Concerns about Genetic Engineering

In 1974, scientists who discovered how to recombine DNA worried about the potential risks. They called for a pause in their research to address these concerns. Led by Stanford’s Paul Berg, the group recommended halting experiments until safety measures were established. In February 1975, about 140 scientists met at the Asilomar Conference in California, setting guidelines to ensure safe experimentation. By 1976, the NIH had formalized these guidelines, tying federal funding to compliance.

AI’s Asilomar Moment

Fast forward to today, the tech industry faces similar challenges. Jacob Coxon, formerly of OpenAI and Anthropic, resigned citing irresponsible actions in AI development. He expressed concerns about unchecked advancements potentially leading to dangerous superintelligence. Coxon urged cooperation among AI labs and a temporary halt on enhancing AI capabilities.

In 2017, AI researchers held a conference at Asilomar aiming to guide AI development through 23 principles. Principle No. 5, “Race Avoidance,” urged teams to avoid compromising safety. Despite this, the AI industry quickly moved towards unregulated development.

Political and Industry Responses

Political figures like Greg Casar and Bernie Sanders proposed the Ban Artificial Superintelligence Act to regulate AI development. This act aims to pause advanced AI development until safety regulations are enacted. However, the Trump administration took a different approach, emphasizing AI dominance.

Anthropic and OpenAI are evaluating safety protocols similar to biological containment measures from the 1970s. Anthropic’s Frontier Safety Roadmap seeks to establish stringent security workflows, although implementation within a short timeframe seems unlikely. OpenAI has deployed GPT-6 Astra, marking a new level of cybersecurity readiness.

Comparing Past and Present

Unlike the 1970s, AI commercialized before regulation. Scientists have urged for voluntary guidelines given AI’s rapid advancement but lack enforceable regulation. OpenAI and Anthropic now advocate for mandatory safety standards.

Over 1,000 scientists have expressed concerns about losing control over AI capabilities, seeking government intervention. The original Asilomar conference illustrates how early regulation was possible before commercial pressures hindered consensus.

Today’s AI race underscores a difference in industry dynamics, where competitive stakes are high, making regulation challenging.