Good morning. Andrew here. While government and corporate leaders worry about the cybersecurity threats raised by Anthropic’s Claude Mythos Preview, there is another conundrum that has gone unexplained. I’ve been speaking to bank executives and A.I. researchers who tell me we are entering an era of the “reverse-engineered exploit,” or the “patch gap.” Once Mythos identifies a bug, a patch needs to be issued to fix it. But hackers have a window of time to study the fix and strike before it’s implemented. That’s always been the case. But in the past, only one or two patches were dealt with at a time. Now it’s likely to be hundreds or thousands of patches. So this becomes a race against time where the very act of repairing a system provides a blueprint for breaking it. Think about that. (Was this newsletter forwarded to you? Sign up here.)
Preparing for next-generation cyberthreatsThe rollout of Anthropic’s most powerful artificial intelligence model is continuing to reverberate not just in C-suites, but in the halls of government. Anthropic has warned that the model, Claude Mythos Preview, could uncover cybersecurity vulnerabilities — and potentially create new ones — at unprecedented speed and scale. That’s increasingly persuading officials in Washington and elsewhere that they need access to help guard against next-generation cyberthreats. The White House is working to make Mythos available to federal agencies, according to Bloomberg. (DealBook has heard the same thing.) Anthropic has limited access to Mythos to a few dozen companies so far, though it briefed government officials on its capabilities before the model’s release. That prompted Treasury Secretary Scott Bessent and Jay Powell, the Fed chair, to convene top banking executives in Washington last week to discuss the tool and its implications for their cyberdefenses. Now the Office of Management and Budget is setting up guardrails to let some agencies start using the A.I. model “in the coming weeks,” according to an email sent this week from Gregory Barbaccia, the federal chief information officer at the O.M.B. Mythos is scrambling the Trump administration’s approach to Anthropic. Remember that the Pentagon declared the A.I. company a supply chain threat, forcing government contractors to avoid using its tools in any military-facing work. Anthropic and the Defense Department remain locked in legal battles in two federal courts. (That said, Dario Amodei, Anthropic’s C.E.O., is set to meet today with Susie Wiles, the White House chief of staff, according to Axios.) But other federal agencies, including the Treasury and State Departments, have asked Anthropic for briefings on and access to Mythos, DealBook understands. Asked at a Semafor event this week about whether government agencies would be able to use Mythos, Sean Cairncross, the U.S. national cyberdirector said, “Sure,” adding that the government was working with tech companies to give agencies access to advanced A.I. “in a responsible fashion.” Some officials said that the Pentagon battle had made it more complicated for agencies to start using Mythos. But even critics of Anthropic’s insistence on tougher guardrails for A.I. technology acknowledge that its products are too good not to use. It isn’t just U.S. officials who are worried about Mythos:
One thing foreign officials and tech analysts are worried about is a lack of cooperation between countries, particularly the U.S. and China, on how to regulate increasingly powerful A.I. models. “Frontier A.I. capabilities are advancing faster than governance frameworks are being designed to control them,” Laila Khawaja, the research director of Gavekal Technologies, told Bloomberg. Time is ticking for governments to prepare for Mythos-level threats. Anthropic itself has cautioned that other developers, including U.S. adversaries, will quickly catch up to its new model’s capabilities, perhaps within two years.
U.S.-Iran peace talks may resume this weekend, President Trump says. Discussions between the two sides would take place in Pakistan, Trump said, claiming that Iran has become more open to unspecified concessions. Brent crude, the international benchmark for oil, is down today on that news and because of a new cease-fire between Israel and Lebanon. But at an event in Las Vegas, Trump played down higher fuel costs as “fake inflation.” Reed Hastings will step down from Netflix’s board in June. Hastings, who co-founded the company as a DVD subscription service in 1997 and turned it into a streaming colossus, said he would “focus on new things.” (He’s already expanding the Powder Mountain ski resort, where he owns a home.) Separately, Netflix disclosed that its latest quarterly earnings fell short of analyst expectations, prompting a sell-off in its shares in after-hours trading. U.S. officials may scrutinize other sports leagues over their media rights strategy. Brendan Carr, the chair of the Federal Communications Commission, said this week that he was concerned about leagues hurting consumers with how they sell some broadcasting rights to streaming services, raising prices. The Justice Department is already investigating the N.F.L. along similar lines. Separately, the White House is reportedly weighing Michael Murray, an official in the first Trump administration, as the Justice Department’s next antitrust chief. Crypto and artificial intelligence groups raise big war chests for the midterm elections. Political action committees tied to the industries have collected about $250 million over the past year, including from Elon Musk, the investor Marc Andreessen and the Wall Street brokerage Cantor Fitzgerald, according to an analysis by The Financial Times. Crypto and A.I. companies are seeking allies in Congress amid debates over how to regulate their businesses.
Readers respond: The pied-à-terre taxOn Wednesday, Andrew asked about Gov. Kathy Hochul’s proposal for a “pied-à-terre” tax on secondary residences in New York City valued at $5 million or more. It’s an initiative that has been floated before, and has already drawn support and alarm. “Critics warn that it could stifle big-ticket investments,” Andrew wrote, “while supporters argue that it is the fairest way to raise revenue.” A lot of readers responded. Here’s what you had to say. (Responses have been condensed and edited.)
“I’ve actually become Boeing’s number-two salesperson in the world, right after the president … unpaid, but it’s true,”— Paolo Zampolli, a longtime associate of President Trump who claims to have helped strike billions of dollars’ worth of business deals around the world as a presidential envoy. (Boeing did not confirm the accuracy of Zampolli’s statement.)
Talking A.I. with the C.E.O. of BCGEvery week, we’re asking a leader how he or she uses artificial intelligence. This week, Christoph Schweizer, who leads BCG, told DealBook’s Sarah Kessler that A.I. hadn’t changed the global consulting firm’s hiring patterns. The interview has been condensed and edited. How are you personally using A.I.? I was at this big conference the other day, and I saw I was seated next to this bank C.E.O. for lunch. I quickly used A.I. to say, hey, how did the last quarter go for that bank? How did the stock market react? What did analysts say? I had a totally different way into the conversation. I don’t use A.I. to summarize emails. For me, it’s really about making human interactions better. What have you told your team about how you want them to approach A.I.? We have given access to a suite of A.I. tools to every one of our 34,000 employees. And we have not prescribed how they should use it. We have said we urge you to embrace it. Now 81 percent of our staff say that A.I. is helping them generate more value in their work, and 73 percent say they find work more enjoyable thanks to A.I. This is about more joy and less toil. Has A.I. changed whom you hire? You read these headlines every day about how professional services of all sorts will be automated away by A.I. We have made the decision that we are not changing our recruiting, promotion or electing-to-partner patterns. The junior talent that comes out of college is incredibly A.I.-native. Nobody knows where the world will be in three or five years from now. But fundamentally, A.I. has been business-creative and not dilutive for us. That’s why we are convinced that our junior talent remains highly relevant. We hope you’ve enjoyed this newsletter, which is made possible through subscriber support. Subscribe to The New York Times.
Deals
Politics, policy and regulation
Best of the rest
|
