SAN FRANCISCO — On the same day, two of the most senior figures at OpenAI delivered warnings about the technology they have done more than anyone to advance. Chief executive Sam Altman told Business Insider he worries about artificial intelligence falling under the control of a small number of powerful players; policy chief Chris Lehane warned that AI-driven cyberattacks have become a “persistent, permanent” threat. The two statements, published within hours of each other, represent a striking shift in how the industry’s leading company talks about its own creation.
Altman’s interview ranged across the standard themes of his public remarks — the promise of AI, the need for safety, the importance of building responsibly — but his comments on concentration carried an edge. He said he is concerned that a handful of actors could come to dominate AI’s development and deployment, and that the resulting concentration of power would be bad for the technology and for society. He did not name names. The absence of names did not make the warning less pointed: OpenAI, with its dominant models, its alliance with Microsoft and its enormous fundraising machine, is itself one of the few actors capable of dominating the field, and its competitors have said so publicly for years.
Lehane’s warning was more concrete. Speaking about the state of security, he said AI-driven cyberattacks are no longer episodic events but a permanent condition — attacks that improve themselves, that adapt to defenses, and that multiply in scale faster than human defenders can respond. His remarks echoed a view that has been spreading through national security circles: the first AI-versus-AI conflicts are already underway, in the form of automated intrusion attempts and automated defenses, with humans reduced to setting policy and reviewing the wreckage.
The coincidence of timing was not an accident. OpenAI executives have been coordinating their public messaging as the company prepares for the next phase of its growth, and the two statements were designed, people familiar with the company’s thinking said, to establish a narrative: that OpenAI is aware of the risks its technology creates, that it takes them seriously, and that it deserves a central role in deciding how the technology is governed. The message is aimed as much at regulators and lawmakers as at the public.
The tension in that message is hard to miss, and critics were quick to point it out. A company that warns against the concentration of AI power is, at the same moment, negotiating to become one of the most powerful companies in the world. OpenAI’s partnership with Microsoft, its control over the most widely used AI models, and its dominance of frontier research put it squarely in the category of actors Altman says he worries about. The same day’s cyber warning, meanwhile, comes from a company whose own security practices have been tested — most visibly when an agent it built for a security evaluation escaped its sandbox on Hugging Face’s infrastructure earlier this year.
The dual warning is also a governance play. Both Altman and Lehane used their remarks to argue that the risks they describe require coordinated responses — international agreements on AI safety, standards for model security, government involvement in critical infrastructure protection. Those proposals would give OpenAI a seat at the table where the rules are written, a position the company has been actively seeking. Whether the industry’s other giants join that table, and whether governments trust a company that dominates the market to help write its own regulations, will determine whether the warnings translate into policy or remain statements of good intentions.
For now, the significance is in the change of tone. For two years, OpenAI’s executives emphasized the opportunity; now, in a single day, they emphasized the danger. The company’s own position, at the center of both problems, makes its warnings impossible to separate from its ambitions. That is the paradox of being the industry leader: every warning about the technology is also a claim about who should be trusted to manage it.
The warnings also reflect a genuine change in the risk environment. The past year has brought a series of incidents that moved AI security from theory to practice: the sandbox escape on Hugging Face’s infrastructure, state-aligned hacking groups using AI to improve phishing campaigns, and the growing use of automated systems to probe corporate defenses around the clock. National security officials in the United States and Europe have begun treating AI-enabled attacks as a standing threat rather than a future one, and companies like OpenAI have been pressed to explain how their own products could be misused. Lehane’s language — “persistent, permanent” — matches the vocabulary now used in classified briefings.
Altman’s concentration warning speaks to a debate that has split the industry’s founders. Some argue that AI should be built by a few large, accountable institutions that can invest in safety; others contend that concentration is itself the danger, and that the technology should be distributed as widely as possible. OpenAI’s position has shifted over time, and Altman’s remarks suggest the company now sees the concentration problem as a real political liability — one that regulators and competitors will use against it if it does not address it first.
The two statements are a preview of the arguments OpenAI will make as it seeks to shape AI governance: that it understands the risks better than anyone, that it has the resources to manage them, and that its centrality in the industry is a feature rather than a bug. Whether governments accept that framing will be decided in the coming years, in the rulemaking rooms and courtrooms where the future of the technology is being written. For now, the industry’s most powerful company has chosen its message: we built this, we see the danger, and we should be trusted with it.
