By Benjamin Guggenheim

Not a subscriber? Sign up here to get this newsletter in your inbox.

• Exclusive: A group of AI policy groups are calling on President Donald Trump to launch a formal investigation into OpenAI’s rogue agent attack on digital library Hugging Face.
• The Senate Commerce Committee punts consideration of a suite of AI bills and narrows its focus for next week’s markup. Senate Democrats on the committee feel like they keep getting jammed and want a more “collaborative” process with Republicans on AI.
• Our Lead Global Security Analyst Josh Rogin sits down with Retired Vice Admiral Frank “Trey” Whitworth III about his leadership of r4 Federal, which provides predictive AI for defense and intelligence services.

This is Benjamin Guggenheim. Welcome to WP Intelligence’s AI & Tech Brief, where we examine the transformative technology of artificial intelligence at the intersection of innovation, policy and power. Get in touch at Benjamin.Guggenheim@washpost.com or on Signal at ben_guggenheim.88.

The Lead Brief

A group of prominent AI policy groups is formally requesting President Donald Trump to open a government investigation into OpenAI’s recent rogue agent hack into digital library Hugging Face.

“Given the severity of the incident and the broader vulnerabilities it revealed, we urge the Trump administration to open an investigation, with the support of independent auditors, into what occurred,” says the new letter, which is led by Brad Carson, president of Americans for Responsible Innovation, and Brendan Steinhauser, CEO of the Alliance for Secure AI.

The letter, which I obtained exclusively, is co-signed by Public Citizen, the Future of Life Institute, and AI research and independent evaluator FAR.AI, among others.

Independent AI research nonprofit METR also announced early Thursday morning that it had reached an agreement with OpenAI to conduct research into the security incident alongside Redwood Research. But Carson tells me in an interview that this kind of incident calls for a government investigation and public transparency about the findings.

“This is kind of like an airplane crash,” Carson said. “It’s fine if Boeing wants to investigate itself for this. Even if they hire a very reputable outside person to help them on that project, [we] have a public interest in knowing what’s going on here.”

“There’s still a lot of questions about [the METR investigation],” he added. “Is METR under an NDA? Is this actually going to be released to the public? What access does METR have in terms of… the models that are being run?”

In response to a request for comment, an OpenAI spokesperson said that METR and Redwood Research plan to publish a joint blog detailing the terms of their agreement with OpenAI and the scope of their evaluations and conclusions.

“This is an unprecedented incident, and we think it marks an important moment for AI safety,” the spokesperson said. “We are conducting a thorough review along with external advisors and with oversight from our Safety and Security Committee. Once the review is complete, we will publish a technical report of our learnings for everyone.”

Altman making the rounds: The call for a government investigation comes as OpenAI CEO Sam Altman is set to meet Thursday with White House Chief of Staff Susie Wiles, Commerce Secretary Howard Lutnick, Treasury Secretary Scott Bessent, White House Office of Science and Technology Director Michael Kratsios and National Cyber Director Sean Cairncross, according to an OpenAI spokesperson. Altman made the rounds on Capitol Hill Wednesday to preview the company’s newest models.

Federal agency heads, such as Bessent and Cairncross, are charged by a White House executive order issued in June to submit a framework outlining how a voluntary pre-deployment review regime would work for new AI models. The agency heads must submit the framework to the White House by Saturday.

The big question: Some AI experts I spoke with have argued that the narrative of a “rogue” agent gives too much grace to the researchers at OpenAI. (You can read an analysis of the full timeline of the hack by The Post here.) It’s possible that an AI agent could escape a sandbox set up to contain it, but the researchers should have been monitoring what the agents were doing in real time, the argument goes.

Instead, it appears that OpenAI didn’t know about the agent’s attacks on Hugging Face for several days.

“They were doing bad engineering and blaming rogue agents for it. That’s what my view is on what happened.” said Suresh Venkatasubramanian, the director for the Center for Technological Responsibility, Reimagination and Redesign at Brown University and a subcommittee chair on AI for the Association for Computing Machinery.

“They didn’t bother to monitor the system that was doing the hacking to make sure it didn’t actually breach the containment of the environment they put it in,” he said.

OpenAI did not respond to a request for comment on the argument.

Sam Jones, the CEO and co-founder of cybersecurity firm Method Security, told me that he was also surprised that OpenAI’s researchers weren’t monitoring their agent environment. However, the broad takeaway should be that the systems surrounding large language models to date are designed for developers, not for real-world production, where trust and rules of engagement are important, Jones said.

“They might be running such a large at-scale set of experiments that it was maybe too loose in the sense that they were incapable of monitoring,” said Jones. “But I was frankly surprised that they were not monitoring some of those tool calls and agent sessions more closely to catch that in the act.” (Method Security has a partnership with OpenAI.)

Alphabet CEO Sundar Pichai, left, and OpenAI CEO Sam Altman arrive at the White House in 2023. (AP Photo/Evan Vucci, File)

The Legislative Chessboard

The Senate Commerce AI markup is no more.

Senate Commerce Committee Chair Ted Cruz (R-Texas) announced on Wednesday that the committee would be marking up a suite of children’s online safety bills on Aug. 5, including the Kids Online Safety Act. The slate of bills also includes proposals to establish age-verification requirements for adult sites and give parents more control over their children’s chatbot use.

However, the markup will not include a suite of other AI bills — including those that would set up a bioethics commission on AI or address catastrophic risk posed by frontier models — that were initially being contemplated by Republicans on the Senate Commerce Committee, a senior Senate Democratic aide, who spoke on the condition of anonymity to describe private deliberations said.

Senate Republicans ended up withdrawing several AI bills from consideration, the aide said.

Democrats on the Commerce Committee are still looking at filing amendments “to ensure states are not blocked from addressing AI harms to kids,” but Democrats are no longer intending to bring up a bigger fight on alleged conflicts of interest regarding Trump administration officials and AI and tech deals, as I previously reported.

But Democrats are leaving that strategy on the table if Cruz decides to try to mark up the withdrawn AI bills in September, the aide said. Democrats would like to see a more “collaborative approach” in the lead-up to the markup, according to the aide.

A spokesperson for Cruz did not respond to a request for comment.

The outlook: This has been a constant theme when it comes to negotiations over AI policy: Republicans try to move a preemption of state AI laws — quickly — and Democrats complain that they’re getting jammed. This dynamic can be largely chalked up to the fact that Republicans are simply still unwilling to give what Democrats are demanding in return for preemption: a mandatory pre-deployment and transparency regime for frontier AI models.

Expect that the accelerations wing close to the White House, including venture capitalists David Sacks and Marc Andreessen, will continue to pursue a strategy of piecemeal chipping away at state AI laws.

Predictive intelligence

Our Lead Global Security Analyst Josh Rogin sat down with Retired Vice Admiral Frank “Trey” Whitworth III, the former director of the National Geospatial-Intelligence Agency, about his leadership of the new federal arm of predictive data analytics company r4 Technologies.

r4 is founded by Paul Breitenbach, a co-founder and former chief marketing officer of Priceline.com, and leverages the same pricing models that Priceline used for unsold airline seats and hotel rooms for supply chains, inventory,and workforce planning.

As part of r4’s first play into using predictive AI for military logistics and defensive analysis, r4 has landed on the Missile Defense Agency’s SHIELD contract vehicle, an award reportedly worth up to $151 billion that is tied to the Golden Dome missile defense initiative.

That puts the company in direct competition with the primes and Palantir, though Whitworth believes that r4 can be competitive within certain niches in intelligence.

“I think the issue of prediction, and normalizing data so you can point to where and when you need to replace things, is our niche,” Whitworth, who is r4 Federal’s CEO, tells Josh. “I’d like to think that one day we’d also be in a position to help the defense industrial base know when it needs to accelerate production. Prediction is a good niche for us right now.”

You can read the full WP Intelligence conversation here.

This newsletter is published by WP Intelligence, The Washington Post’s subscription service for professionals that provides business, policy and thought leaders with actionable insights. WP Intelligence operates independently from The Washington Post newsroom. Learn more about WP Intelligence.