Welcome to Wednesday’s edition of Washington Secrets, where we are trying to finish up early so we can sneak off for a huge World Cup afternoon. But first, we talk to a tech insider who tells us that Anthropic’s doom-mongering about AI is actually a hype machine designed to help it corner the market. Plus, we have polling that shows what the American public thinks of AI and its impact on the American Dream…Anthropic sent shockwaves through the tech industry in April when it announced it had built a large language model so advanced, so awe-inspiringly powerful, so omnipotent, that it was simply too dangerous to release into the wild.Instead, the artificial intelligence giant set up Project Glasswing. It invited security experts from other companies, including rivals and U.S. intelligence agencies, to test the platform and find and fix vulnerabilities in their own critical infrastructure before the new model could be released. After weeks of probing, the result was Fable, a wings-clipped version of the original Mythos platform.
Even that was not enough. On June 12, the U.S. government imposed an export control directive suspending all access to Fable 5 and Mythos 5 by any foreign national, whether inside or outside the United States, for fear it could be used by malign forces to hack vital systems.But all is not as it seems, according to an industry figure who had access to Project Glasswing. The claims did not stack up.“Frankly what we saw with it was, I think, that some of the public statements were exaggerated,” he said.It was not right to say that the model could be sent off to hack into critical systems with just a command.“At the end of the day, it’s just a model that can do coding tasks better than the previous generation models,” the industry figure added.His inside view of Project Glasswing reflects a growing sense of where the debate over AI is headed. Anthropic is positioning itself against rivals OpenAI and Elon Musk’s xAI as the responsible company, the only one that can really be trusted to manage a frightening new technology (all while vacuuming up billions upon billions of dollars).They are the good guys at the frontier surveying a Wild West of rapidly developing technologies.“Look at us, we created a very scary thing, and at the same time we’re going to hold it back, you can trust us,” said the security expert. The result is a conversation driven by fear and anxiety at a time when the U.S. is poised to win big in the field.Nonsense, says Anthropic. There is a lag between any new platform being released and vulnerabilities being patched. Project Glasswing buys time to allow fixes to be made before any harm can be done.A spokeswoman pointed Secrets to a one-month update on the project, which said the company and about 50 partners had identified more than 10,000 of the worst vulnerabilities across “systemically important software” around the world.For example, it said Mozilla had found and fixed 271 issues in its browser, Firefox 150, while testing Mythos Preview, more than 10 times as many as it found in an earlier version using Claude Opus 4.6, Anthropic’s previous model.On Tuesday, the White House dropped its restrictions on Anthropic’s Mythos and Fable.But the industry source said the danger was that Anthropic had set a precedent for the entire sector with its hype. Its voluntary Project Glasswing could become a new standard.“We do run the risk of actually putting those under those regulations at a disadvantage when you look at international competition and where it’s evolving,” he said.Instead of discussions about how AI was solving problems in healthcare, creating new drugs and trials faster than ever before, the conversation was starting with the fears.Critics say it comes against a backdrop of a company crying wolf time and again, hyping unfounded fears over everything from the threat to jobs to the efficacy of export controls.In January, for example, Dario Amodei, its chief executive, set out the stakes in apocalyptic terms, claiming AI was advancing faster than the world’s ability to control it.“Whether we survive that test and go on to build the beautiful society described in Machines of Loving Grace, or succumb to slavery and destruction, will depend on our character and our determination as a species, he wrote in a blog post.In it, he warned that AI could halve the number of entry-level white-collar jobs.In fact, a recent study by corporate card firm Ramp and workforce analytics firm Revelio Labs found that companies investing in AI were expanding their workforces (although the data were drawn from large companies that may already have been growing).And he warned that allowing chipmakers to sell processors to China was “like selling nuclear weapons to North Korea.” In fact, Chinese executives have since said that they were grateful for the previous export control for helping their domestic industry to grow.HOW ANTHROPIC LOST A BATTLE BUT COULD WIN A WARThe result is a sense of frustration among security professionals inside the industry about a conversation dominated by extremes.“The polarizing ends of it don’t enable a good conversation about where we should be and where we should invest as an industry,” said the source.Quote of the day
