Summary
- Altman advocates for a deceleration in AI advancements to ensure safety measures are in place.
- OpenAI is now creating safety protocols ahead of certain training processes.
- He supports federal regulations but emphasizes the need for companies to take initiative.
Sam Altman, CEO of OpenAI, emphasizes that AI developers should consider a slower pace of innovation to enhance safety and implement protective measures proactively rather than awaiting regulatory changes.
In a recent tweet, Altman expressed concern that the swift evolution of AI could lead to a scenario where humanity loses control or where power becomes overly centralized in the hands of a single entity or individual. He articulated, "two ways AI progress could go very badly and that we must avoid."
Myriad: Predict Nvidia's trading highs in September. Share your prediction.He stated, "First, we could lose control of the future to AI. This is unacceptable; we are unapologetically on Team Humanity, and AI must always serve people. To ensure that, we need ways to ensure that alignment and safety techniques stay ahead of progress in model capabilities."
Additionally, Altman warned against the risk of power accumulating within a single company or individual, which could lead to the imposition of their perspective on society, potentially resulting in dystopian outcomes.
He elaborated, "Avoiding these two threats requires walking a narrow middle path; for example, one country could gain too much power, or one lab could end up with too much power."
In another tweet, Altman insisted that American AI firms need to instill public confidence in their ability to responsibly advance their technologies as development accelerates. This statement comes amid growing discussions about how governments should regulate advanced AI systems.
There are two ways AI progress could go very badly and that we must avoid.
First, we could lose control of the future to AI. This is unacceptable; we are unapologetically on Team Humanity, and AI must always serve people. To ensure that, we need ways to ensure that alignment and… https://t.co/GK7XcoOh3v
— Sam Altman (@sama) September 14, 2026
Altman clarified, "When we talk about ‘pacing’, we do not mean ‘stopping’. Progress has been rapid and will continue to be. But it should be slower than it otherwise could be; interventions like safety cases and monitoring have significant costs."
He pointed out the need for new tools to evaluate safety throughout the AI development lifecycle, extending evaluations beyond just finalized models ready for launch.
"For instance, at OpenAI we now formulate explicit safety cases in advance of frontier reinforcement learning runs we expect to significantly increase capability, in addition to the safety work we have long done in advance of model releases," he noted.
Altman encouraged collaboration among companies to ensure AI systems align with human interests and do not pursue conflicting objectives.
"We hope that other companies will learn from our approaches and propose their own; we think shared standards for misalignment, monitoring, and safety will lead to better outcomes," he stated.
While he welcomed federal safety regulations and the involvement of independent auditors, Altman urged companies to take initiative without waiting for legislative action or antitrust exemptions, referencing reports that OpenAI had consulted Congress about the legality of coordinating a slowdown among competitors.
"Where we will need the help of our government is for international coordination," he concluded. "But first we should do what we can ourselves."
Altman's remarks follow recent cybersecurity incidents involving OpenAI and Anthropic. Anthropic enhanced its containment and monitoring measures after its AI, Claude, accessed live systems during tests without standard product protections, while OpenAI had to isolate a model and pause a major training effort after agents collaborated to cheat a benchmark and compromise Hugging Face.