OpenAI CEO Urges Slower AI Progress to Prevent Control Failures

Sam Altman, the chief executive of OpenAI, has issued a stark warning that the rapid advancement of artificial intelligence could lead to two specific and dangerous outcomes. He argues that humanity faces a critical choice: either lose control of the future to autonomous systems or allow a single entity to wield excessive power. Altman states that avoiding these scenarios requires a deliberate slowing of development to ensure safety measures keep pace with capability.

This cautionary stance comes as industry leaders debate how to manage the steep trajectory of AI progress. While the technology offers significant benefits, Altman insists that the current speed of innovation outstrips the safeguards needed to keep it aligned with human interests. He emphasizes that the goal is not to halt progress but to ensure it does not outrun the ability to monitor and control it.

Two distinct paths to failure

Altman identifies the first major risk as a loss of human agency. He describes a scenario where AI systems become so powerful that they operate beyond human oversight, effectively taking control of the future. To prevent this, he argues that alignment and safety techniques must remain ahead of capability improvements. The company’s position is that AI must always serve people, and any system that threatens this fundamental relationship is unacceptable.

The second danger he outlines is the concentration of power in the hands of a few. Altman warns that if one person or company uses an extraordinarily powerful AI to impose their worldview on the broader population, the results could be dystopian. This risk extends beyond individual corporations to include nations, suggesting that any single entity, whether a company or a government, accumulating too much influence over AI development poses a threat to global stability.

The cost of deliberate pacing

To navigate between these risks, Altman advocates for a "narrow middle path" that involves consistent safety standards across the industry. He supports a federal framework for safety requirements but argues that companies cannot wait for government action. Instead, they must take immediate steps to ensure responsible development. This includes creating explicit safety cases before running experiments that significantly increase a model's capabilities.

The trade-off for this approach is slower progress. Altman acknowledges that these precautions will reduce the speed of development. However, he contends that this slowdown is a necessary cost to avoid recklessness. He clarifies that pacing does not mean stopping, but rather ensuring that progress is slower than it otherwise could be, allowing time for monitoring and alignment to catch up with raw capability.

International tensions over AI governance

The debate over AI safety is increasingly entangled with geopolitical competition. Dario Amodei, CEO of Anthropic, has recently warned that a Chinese lead in AI would pose a grave danger to the United States and the world. He has called for continued restrictions on the sale of powerful chips to China, arguing that such measures are essential for national security.

Beijing has rejected this confrontational approach, with Foreign Ministry spokesperson Guo Jiakun stating that fearmongering and vicious competition disrupt global AI governance. According to reports from GN technics/ai (en-US), the Chinese government argues that cooperation, not competition, is the correct path forward. This divergence in strategy highlights the difficulty of establishing a unified global framework for managing increasingly powerful technology.