OpenAI's Sam Altman Says the World Is "Right to Be Afraid" of AI, Then Asks for Trust

Today’s Focus

OpenAI Chief Executive Sam Altman told an audience at a Salesforce conference in San Francisco on Tuesday that the public is justified in fearing artificial intelligence, but that AI developers should still be trusted to police themselves, according to the BBC.

“The world should trust that we are going to do the right thing because it’s the right thing and we feel the magnitude of this,” Altman said, according to the BBC’s account of the event.

He added that “it doesn’t take as much imagination as it used to” to picture how the technology could cause harm, and that “the world is right to be afraid of this.”

Altman’s remarks landed as calls for federal AI regulation intensify from an unusual coalition. Sen. Bernie Sanders (I-VT) and Steve Bannon, a former adviser to President Donald Trump, have both pressed for tougher governance of frontier AI systems, the BBC reported.

Trump has moved in the opposite direction. The president called safety fears surrounding AI a “hoax” and said the only “guardrails” the sector needs are “a strong and smart” president, according to the BBC.

Altman himself has acknowledged the gap. In an Axios interview published Sept. 3, he said he had expected Congress to enact a “basic framework” for advanced AI after his 2023 Senate testimony, but that lawmakers have struggled to keep pace without “slow[ing] down innovation.”

No comprehensive federal AI statute has passed Congress since ChatGPT’s late-2022 debut. Regulation to date has come through executive orders, state laws, and voluntary industry commitments brokered by the White House.

The Debate

Supporters argue

Altman and other industry leaders contend that the companies building frontier models understand the technology better than anyone and internalize the risks. “We feel the magnitude of this,” Altman said at the Salesforce conference, per the BBC, framing safety as a core corporate motivation rather than a compliance burden.

Supporters of a light-touch approach argue that heavy-handed rules would freeze American firms while Chinese competitors accelerate. Trump, according to the BBC, has repeatedly cast strict guardrails as a threat to U.S. leadership, arguing that executive judgment can substitute for prescriptive statute.

Industry groups add that AI capabilities change monthly, and that Congress writes laws in years. Altman told Axios earlier this month that lawmakers face a genuine dilemma in trying to regulate a moving target without cutting off beneficial uses in medicine, science, and productivity.

Backers also point to voluntary safety testing, red-teaming, and disclosure agreements that leading labs, including OpenAI, Anthropic, and Google DeepMind, have signed with the U.S. and U.K. governments as evidence that self-governance is already producing accountability.

Critics argue

Skeptics say no industry in modern history has been trusted to regulate itself on questions of public safety, and AI should not be the first. Sanders, cited by the BBC, and Bannon have both argued that concentrated corporate control over powerful models demands democratic oversight rather than executive goodwill.

Civil liberties groups warn that voluntary commitments are unenforceable. Without statutory authority, regulators cannot audit training data, mandate incident reporting, or impose penalties when models cause discriminatory or dangerous outcomes.

Critics also note the contradiction inside Altman’s own remarks. If the world is “right to be afraid,” as he told the Salesforce audience per the BBC, they argue that fear is precisely the condition under which binding rules, not trust, are appropriate.

Some Republicans share the concern from a different angle. Bannon and allied populists contend that unregulated AI concentrates power in a handful of Silicon Valley firms, a structural problem they say Trump’s “strong president” framing does not address.

What the experts say

Academic researchers who study AI governance have documented a widening gap between capability and oversight. Stanford’s 2025 AI Index, produced by the university’s Institute for Human-Centered AI, found that the number of notable AI incidents tracked by the AI Incidents Database rose sharply between 2022 and 2024, while binding federal U.S. regulation remained largely absent.

The Brookings Institution’s Center for Technology Innovation has argued in published analyses that voluntary commitments by AI labs, while useful, lack the enforcement mechanisms of sectoral regulation in areas such as aviation or pharmaceuticals, where safety cases are reviewed by independent agencies before deployment.

Comparative data from the OECD AI Policy Observatory shows the European Union’s AI Act, which entered into force in 2024, is the first broad statutory framework classifying AI systems by risk tier. The United States, by contrast, relies on a patchwork of state laws, including measures in California and Colorado, and executive branch guidance.

Pew Research Center surveys have consistently found that a majority of American adults report more concern than excitement about the growing role of AI in daily life.

By the Numbers

2023: year Sam Altman testified before the Senate Judiciary Committee calling for federal AI licensing, according to the BBC and committee records.

“Hoax”: the word Trump has used to describe AI safety fears, per the BBC.

2024: year the European Union’s AI Act, the first comprehensive statutory framework for AI, entered into force, according to the OECD AI Policy Observatory.

Sept. 3, 2026: date of Altman’s Axios interview in which he said Congress has failed to enact a basic AI framework since his 2023 testimony.

Majority: share of U.S. adults expressing more concern than excitement about AI’s expanding role, according to Pew Research Center polling.

Zero: number of comprehensive federal AI statutes enacted by Congress since ChatGPT’s public release in November 2022.

Rising: trend in notable AI incidents tracked by the AI Incidents Database between 2022 and 2024, per Stanford’s 2025 AI Index.