OpenAI confirmed that it is discussing AI security cooperation with Anthropic and Google (GOOGL.US) DeepMind. The FTC is throwing cold water: it's like “digging a moat”

The Zhitong Finance App learned that AI security issues are bringing Silicon Valley's fiercest rivals to the same negotiation table. Chris Lehane (Chris Lehane), head of global policy at OpenAI, said in Washington on Tuesday local time that OpenAI has been cooperating with Anthropic and Google (GOOGL.US) DeepMind for weeks on AI security issues to respond to concerns from the outside world that AI poses an economic and security threat. He also added that OpenAI believes that the three AI companies do not need antitrust exemptions to coordinate security matters.

This statement brought AI giants' collaboration on security issues to the front of the stage, but it also quickly triggered US antitrust regulators to be wary of “digging a moat.”

Lehan said at a briefing in Washington that OpenAI's contacts with Anthropic and Google DeepMind have continued for weeks. “A better approach is to work together and prioritize safety,” he said. Using the aviation industry as an example, he said that for a long time, companies in many industries have helped each other on safety issues, and there are plenty of precedents for AI companies to cooperate on safety issues.

An OpenAI spokesperson said that discussions have been ongoing since Google DeepMind Chairman Demis Hassabis (Demis Hassabis) proposed the establishment of an American-led “standards body” in July. At the time, Hassabis proposed the establishment of a public-private partnership or self-regulatory organization and subject to federal supervision, similar to the Financial Industry Regulatory Authority (FINRA). OpenAI CEO Sam Altman (Sam Altman) allegedly even made it clear to employees that there is no need to rely on the US government to cooperate to create an AI industry standards body.

Lehan also posted a blog earlier this month saying that OpenAI is promoting “industry-leading standards” and will establish voluntary efforts with other companies “with or without government support.” He stressed that any industry-leading standard should complement, and not replace, mandatory federal guarantees and democratic supervision. “We've reached a different stage in AI capabilities, and everything should be on the table, including considering volunteering to build something outside of government as an immediate step.”

The disclosure of this discussion stemmed from Anthropic CEO Dario Amodei (Dario Amodei)'s 3,800-word long post last Saturday. He called for limiting the pace of development of the most advanced systems so that researchers can better understand potential threats. The article was quickly supported by Altman, Hassabis, Musk, and others.

Hassabis wrote on X: “Dario's article points to the right path forward. The details are yet to be worked out, but this direction is the right one for dealing with this critical juncture.” Altman also said he agreed on the need to slow down the development of the industry's most advanced models. He said on X that the slowdown is a “major topic” discussed by OpenAI in recent weeks, and the company will share more information soon. Altman also said OpenAI will join Anthropic by embedding third-party evaluators within the company to monitor security.

Amodei's article proposed granting narrow government exemptions to allow such security coordination. But Lehan said on Tuesday that the three companies do not need antitrust exemptions. This statement did not dispel the concerns of regulators.

Antitrust red line: FTC chairman says he is “deeply skeptical”

US Federal Trade Commission (FTC) Chairman Andrew Ferguson (Andrew Ferguson) said on Tuesday that he “deeply doubts” AI companies' requests for antitrust exemptions. He believes these companies are actually “requesting barriers to entry” to protect their business from challengers. “It really sounds like digging a moat,” Ferguson added.

At the same time, the National Assembly is also promoting relevant legislation. Senators Jim Banks (Jim Banks) and Adam Schiff (Adam Schiff) proposed a bill to grant AI companies narrow antitrust exemptions to allow them to share uncontrolled, cyber, or biological threats with AI systems. A person familiar with the matter said that the bill has been incorporated into the annual National Defense Bill.

However, the possibility of passing any proposals before the end of the year is still slim. The US House of Representatives is scheduled to adjourn on Thursday and will not return until after the November elections. The short “lame duck session” after the election is expected to focus on plans to raise the Pentagon's budget.

Lehan said that he is in Washington this time to meet with US lawmakers, and OpenAI will support bipartisan proposals aimed at dealing with the potentially disastrous risks of technology. He said that the measures supported by OpenAI include the federal AI governance framework proposed by Republican Congressman Jay Obernolte (Jay Obernolte) and Democrat Lori Trahan (Lori Trahan), as well as ongoing proposals from Senate Majority Leader John Thune (John Thune), Republican Ted Cruz (Ted Cruz), and Democrat Amy Klobuchar (Amy Klobuchar). “Whatever we can push forward, we support it,” Lehan said.

At another Washington event, Independent Senator Bernie Sanders (Bernie Sanders) said he will introduce legislation next week to ban so-called “super-smart AI” — a level of ability that may exceed human control. The Democrat-allied Vermont senator rarely found common ground with conservative agitator Steve Bannon (Steve Bannon), calling for immediate action to strengthen AI security.

Sanders criticized, “Congress, whether controlled by the Democratic Party or the Republican Party, is asleep at the wheel. And we now have a president, and his ignorance of this issue is embarrassing.”

However, US President Trump strongly opposed the AI fence measures. He posted a series of social media posts on Monday, criticizing Anthropic and OpenAI for pushing to slow down cutting-edge system development and using exaggerated language to refer to AI existential risks as a “scam.” Trump's key AI adviser David Sacks (David Sacks) also added that concerns about the risks of AI's existence have been exaggerated.

Within Silicon Valley, opinions about the need for regulation are equally divided. Nvidia (NVDA.US) CEO Wong In-hoon said at a SEFTSE event on Tuesday that market power is sufficient to help the company innovate safely, and that choosing between the safety and speed of AI development is a false proposition. “You can have both at the same time,” Wong In-hoon said. “You can slow down until you're sure what you publish will be recognized by the market.”

Lehan said at the conference on the same day that OpenAI supports a provision in the “FRONTIER Act” that will force top cutting-edge laboratories to allow “independent verification organizations” to enter the company to ensure safe model development.

As of press time, neither Anthropic nor Google DeepMind spokespersons had responded to requests for comment.