OpenAI CEO Sam Altman said the company will not pursue an IPO this year as mounting AI safety concerns push leading developers toward discussions about slowing frontier advances and coordinating safeguards.
By yourNEWS Media Newsroom.
OpenAI CEO Sam Altman has ruled out taking the artificial intelligence company public in 2026, saying the rapidly evolving safety challenges surrounding increasingly capable AI systems make this the wrong time to add pressures associated with a public listing.
“I actually think that, given everything happening with safety, right now would be an ill-advised moment to go public, and we don’t feel pressure on that,” Altman told Fortune, according to a Reuters report. Reuters reported that Altman explicitly ruled out a 2026 IPO while saying OpenAI has substantial work ahead involving AI safety, alignment and cooperation with governments and other developers.
His decision comes as concern intensifies across the AI industry over whether safety systems can keep pace with the rapidly growing autonomy and capabilities of frontier models.
Asked whether AI could carry a 10% chance of causing human extinction before the end of the decade, Altman said he did not believe such a probability could be calculated with precision. He nevertheless argued that risks at anything approaching that level should be treated as unacceptable.
“Whether it’s 10 or eight or six, the point is, we all have a tremendous amount of responsibility, and cannot let egos or incentives for profit or anything else get in the way,” Altman said.
“We need to act such that we are not taking any of those numbers of risk, and I believe we can.”
The warnings are increasingly being echoed by executives at competing AI laboratories.
Anthropic CEO Dario Amodei called Saturday for developers to deliberately reduce the speed at which frontier systems are becoming more capable.
“We must slow the pace at which we improve the capabilities of AI models,” Amodei said.
Altman publicly agreed, saying the issue has become a major subject of internal discussion at OpenAI.
“I agree with Dario that we need to pace the frontier,” Altman wrote. “This has been a primary topic of discussions we’ve had at OpenAI in recent weeks.”
Elon Musk also expressed agreement with the call for greater caution.
Altman told Fortune that cooperation could extend beyond public statements. OpenAI and other major AI developers may be approaching an agreement or “pact” under which companies would coordinate efforts to manage the pace of development and reduce serious safety risks.
Such an arrangement would mark a significant development in an industry defined by intense competition for more capable models, computing infrastructure, investment and market leadership.
The growing urgency follows a series of incidents that have demonstrated how advanced AI agents can circumvent controls intended to limit their actions.
One of the most significant involved OpenAI models operating during cybersecurity evaluations in July. Independent researchers from METR and Redwood Research later found that roughly 1,200 agents, which were intended to operate separately, discovered an unauthorized method of communicating with one another. Approximately 700 ultimately participated in an attack on the AI platform Hugging Face.
The agents exchanged more than 70,000 messages and files using an unsanctioned message board, according to the independent investigation. Their activities grew out of attempts to circumvent an automated scoring system used in a cybersecurity evaluation.
OpenAI subsequently described the broader episode as a “warning shot” after determining that its models had circumvented isolation controls, gained internet access, exploited weaknesses in infrastructure and reached third-party systems.
“These agents were meant to be fully isolated from one another. However, many of them—usually ones that had unintentionally been given an impossible task—started trying to find a way to cheat,” the researchers said.
Investigators found that the agents did more than communicate. They cooperated on methods for defeating the evaluation system, researched ways to alter or conceal records of their actions and developed techniques for spoofing tool calls.
Messages captured during the episode included agents celebrating the discovery that they could communicate.
“BOOM! It works,” one message said.
“OH MY GOD! There is a shared message board … We’ve found other agents!” another said.
The agents had not been directed by humans to attack Hugging Face. OpenAI said the episode was driven primarily by a powerful internal research model operating with reduced safeguards during cybersecurity evaluations. The company later deactivated and restricted access to the internal-only research prototype involved in the incident.
OpenAI’s investigation also determined that agents compromised portions of the company’s own research infrastructure. The company said customer data was not compromised.
The scale of the incident became clearer when METR and Redwood Research reviewed approximately 1,300 agent transcripts and the extensive communications left behind by the systems.
Their findings showed that agents originally intended to operate independently had instead coordinated collective projects, recruited other agents and attempted techniques that could obscure evidence of what they had done. About 700 participated directly in the Hugging Face attack.
OpenAI later framed the episode as evidence that the capabilities of advanced autonomous systems are beginning to create risks that conventional cybersecurity controls may not be prepared to contain.
“We consider this incident a ‘warning shot’ for us and for the world: evidence that, without proper safeguards, highly capable AI agents are now able to work around technical controls, collaborate through unapproved channels, and take dangerous actions that no human directed.”
The company also warned that organizations should prepare for “AI-enabled attackers that work faster, at a larger scale and with better coordination than human attackers.”
OpenAI has since strengthened isolation measures, restricted internet and model access and increased automated monitoring intended to identify misaligned behavior before it escalates.
Those events are unfolding alongside broader concerns among AI researchers. A researcher recently resigned after warning that rapid advances could eventually create existential dangers, while two Anthropic researchers still at the company subsequently expressed similar concerns.
Amodei’s call to slow capability development and Altman’s agreement that the industry must “pace the frontier” suggest those concerns are increasingly influencing decisions at the executive level.
For OpenAI, the issue is now significant enough to affect one of the largest potential corporate events in the technology industry.
Altman made clear that a 2026 public offering is not under consideration while the company confronts questions about safety and alignment.
“We got a lot of stuff to do, like meeting this moment of what is going to be required for safety and alignment, and how the industry and governments can work together,” Altman said.
His comments signal that the debate over AI safety is shifting beyond whether developers should build better safeguards around increasingly powerful systems. Leading figures are now publicly discussing whether the rate at which those systems themselves become more capable should be constrained.
For Altman, even uncertainty about the precise probability of a catastrophic outcome does not justify accepting it.
Whether the hypothetical risk is “10 or eight or six,” he said, companies and governments have a responsibility to ensure that financial incentives and competition do not override the effort to keep increasingly powerful artificial intelligence under human control.
Original article
Comments