Key points: After Trump had lunch with AI company executives at the White House, the industry found itself back at square one on AI safety.
A document signed by top industry executives advocates industry self-regulation with help from outside evaluators.
"Two weeks ago, it was almost exactly this group of CEOs who were saying they should accept government oversight," said Bradley Tusk, CEO of Tusk Ventures, in an interview.
On September 29, 2026, in Washington, U.S. President Donald Trump spoke with Meta Platforms, Inc. (NASDAQ: META) CEO Mark Zuckerberg in the White House driveway after a lunch with tech leaders in the East Room. On Tuesday, Trump had lunch with AI industry leaders. After the meeting, the leading artificial intelligence companies remained in the same old position: relying on corporate self-restraint.
Over the past three weeks, AI safety has become a fierce national debate in the United States, with top industry researchers and executives warning that the most powerful AI models pose extreme threats. Against that backdrop, Trump invited executives from Alphabet (NASDAQ: GOOG), Meta Platforms, Inc. (NASDAQ: META), SpaceX, NVIDIA (NASDAQ: NVDA), Anthropic, OpenAI and other companies to the White House.
But despite many warnings this month from industry figures such as Anthropic CEO Dario Amodei, and calls from many policymakers and industry experts for government regulation, Trump firmly opposes such measures and has called concerns about AI risk a "hoax." On Tuesday, Trump gathered corporate leaders at the White House to sign a document and posted a signed version on his Truth Social platform under the title "White House Superintelligence Accord: Joint Commitment on Frontier Responsibility." The document states that signatories believe "each company has a responsibility to develop its own technology safely and to build trust with customers and the public."
Some of the executives who signed the document had precisely been emphasizing that, without coordinated safety controls, their technology cannot be trusted. "Two weeks ago, it was almost the same group of CEOs publicly saying the industry should accept regulation," Bradley Tusk, CEO of Tusk Ventures, said Tuesday on the program "Closing Bell Overtime." "It is not because they truly believe in the power of the state. They want one set of rules so that every company stands on the same starting line."
The agreement lists four action requirements: each company must establish complete internal model monitoring mechanisms; form internal teams to ensure controls are implemented; cooperate with external audit and evaluation institutions; and set up an independent board committee to oversee the internal safety team. The final paragraph begins: "In the long run, it may be necessary to codify the above measures into law or regulation." When reporters outside the White House asked whether these rules would have legal force, Trump said they carried only moral force.
Trump did not hold the event merely to roll out the new agreement. He also announced a move that he said would "officially" rename artificial intelligence (AI) as "superintelligence." Last week he had already asked the public on social media for a new name, and this time he wanted to push the rebranding forward.
This is only the beginning
Tech companies will not abandon the abbreviation AI in the short term. But after Tuesday's lunch, executives quickly voiced support for Trump's series of moves. This has become the norm in Trump's second term: large U.S. companies, especially in the tech industry, try hard to avoid conflict with the president. Meta Platforms, Inc. (NASDAQ: META) CEO Zuckerberg told reporters the agreement was meant to "let the American people and our customers believe that the technology will work the way we envision."
"That does not mean this is all we are going to do," Zuckerberg said. "It is only a beginning, an agreement the whole industry can take part in." Compared with peers such as Amodei and OpenAI's Sam Altman, Zuckerberg has leaned toward the self-regulation path, and NVIDIA (NASDAQ: NVDA) CEO Jensen Huang has taken a similar position. Earlier this month, Amodei published an article calling on AI labs to "pace the advance of frontier technology." A few days later, Zuckerberg posted on X: "Trust and alignment are rapidly becoming the most important capabilities distinguishing AI agents and large models."
On September 29, 2026, in Washington, Anthropic CEO Dario Amodei headed to the White House to meet with U.S. President Trump. Amodei's relationship with the White House has been bumpy, but over the weekend he had dinner with Trump. After the signing ceremony, facing a crowd of reporters, his wording was quite measured: "If we do this well and work closely with the president and everyone present, we can succeed safely."
Alphabet (NASDAQ: GOOG) CEO Sundar Pichai called the moment "historic and far-reaching." However, with high-profile AI safety leaks increasing and the November midterm elections approaching, public perceptions of AI companies have clearly worsened, and calls for legislative regulation will only grow louder. A September Verasight poll showed that 63% of respondents believed AI development should slow down, while only 5% wanted it to accelerate.
Democratic Virginia Senator Mark Warner said Tuesday that he had introduced a bill requiring "rigorous testing and evaluation" of the most powerful models before deployment, strengthening safety safeguards and reporting obligations. "The companies building the most powerful AI systems are themselves warning that technology development has run ahead of the safety protection system," Warner said in a statement. "And what is the president's response? Change the name, then let the developers regulate themselves. But merely changing what AI is called cannot solve the real risks posed by increasingly powerful AI systems."
Comments