Anthropic disabled its new model Fable 5 late on Friday.<\/p>\n
NurPhoto via Getty Images<\/p>\n
Late on Friday, Anthropic issued a surprise announcement: It was disabling its new Fable 5 model, released just days earlier, after a directive from the U.S. government that cited national security concerns. The model is a safeguarded version of Anthropic\u2019s Mythos family of models that were deemed so powerful when first unveiled in April that it caused widespread alarm over cybersecurity threats. <\/p>\n
In a June 12 blog post<\/a>, Anthropic said it believes the government became aware of a \u201cjailbreaking\u201d method to bypass Fable 5\u2019s safeguards. While the company is complying with the directive, it pushed back against it. \u201cWe disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people,\u201d the blog post says. \u201cIf this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.\u201d <\/p>\n Just hours before the directive \u2014 Anthropic said it received the government\u2019s letter at 5:21 p.m. Eastern Time on Friday \u2014 Chief Commercial Officer Paul Smith discussed the \u201ctradeoffs\u201d when it comes to releasing such a powerful tool in an interview with Forbes. <\/p>\n \u201cYou have to make a judgment call on these things,\u201d he said. \u201cThe safest you can be is to not let people use something. And then it\u2019s totally safe. But then how is that helping the mission, and how is that helping people actually derive real value from Mythos-level intelligence?\u201d <\/p>\n Of course, he wasn\u2019t talking about the directive, which hadn\u2019t been issued yet. Instead, Smith\u2019s remarks were in response to complaints by some users that Anthropic\u2019s safeguards went too far, hamstringing the model with so many limitations, particularly with prompts related to biology (which could be used for bioweapons) and cybersecurity, that it rendered the model neutered<\/a>. \u201cOver the next few days and few weeks, we can tune and dial back those classifiers to make it a more smooth and seamless experience for users, but that was kind of like the tightrope we\u2019re trying to thread,\u201d he continued.<\/p>\n Still, Smith\u2019s sentiment is a helpful lens through which to view Anthropic\u2019s latest dustup with the government, which first sought to ban the use of Anthropic products for federal use in February and on Friday banned the use of Mythos by foreign nationals: Anthropic\u2019s goal is to get the tools out there, and do it as safely as possible, but acknowledges that there are always risks. <\/p>\n White House AI czar David Sacks doesn\u2019t agree. \u201cIn the past, Anthropic has always said that safety must be top priority and taken super seriously. In this case, Anthropic prioritized the continued offering of the consumer model over safety,\u201d he wrote on X<\/a> on Saturday.<\/p>\n Anthropic declined to comment on the situation beyond its blog post. <\/p>\n Meanwhile, the decision to disable the model has sent waves throughout the industry. It came as a surprise to some Anthropic customers. \u201cSounds like we\u2019re going to have [to] turn off access to Fable,\u201d Amjad Masad, CEO of Replit, an AI coding company that offers its users access to Anthropic\u2019s models, posted<\/a> on X late on Friday. According to a Reuters<\/a> report that cited a source close to the matter, several tech executives including Amazon\u2019s Andy Jassy had raised concerns to Trump over security risks tied to Anthropic\u2019s latest models.<\/p>\n In March, CEO Dario Amodei engaged in a high-profile standoff with the Department of War over the Pentagon\u2019s ability to use Anthropic\u2019s AI, particularly regarding mass surveillance and autonomous weapons. After the tense back-and-forth, the government labeled Anthropic a \u201csupply chain risk\u201d \u2014 a devastating blow to any company that does business with the military. Anthropic subsequently sued the DoW, and a judge put an injunction on the designation. <\/p>\n Aside from the Fable safeguards, Smith touched on other topics during his interview with Forbes. <\/p>\n Asked if the DoW situation would have been more complicated as a public company, Smith demurred. \u201cThat\u2019s a crystal ball thing,\u201d he said. \u201cThat was a very complicated situation. I think that will always be a very complicated situation [public company or not].\u201d<\/p>\n He declined to discuss a timeline for Anthropic\u2019s planned IPO, after the company confidentiality filed its S-1 prospectus with the SEC last month. But he insisted that the company\u2019s stated mission of AI safety wouldn\u2019t be at odds with being a public company, when it would be subject to shareholder pressure. \u201cThere is no question in terms of Dario and the other co-founders\u2019 commitment to safety, to trust, to keeping focused on that core mission,\u201d he said. \u201cCandidly, I think the number one question that’s going to be on a post IPO earnings call will be, how is AI diffusion going in the wider economy.\u201d <\/p>\n But he did discuss Anthropic\u2019s deal with Elon Musk\u2019s SpaceX, which went public Friday in the biggest IPO in history, to run its models on SpaceX\u2019s Colossus supercomputer. As part of the contract, Anthropic is paying SpaceX $15 billion a year through 2029, according to SpaceX\u2019s S-1. \u201cHow it came about is, we always want to diversify our sources of compute. And this was just the perfect intersection of SpaceX also wanting to diversify who was using that compute,\u201d Smith said. \u201cIt came together very, very, very, very quickly.\u201d <\/p>\n Still, Anthropic and Musk make unlikely bedfellows. In February, he slammed the company on X, calling it \u201cevil\u201d and discriminatory. \u201cFrankly, I don\u2019t think there is anything you can do to escape the inevitable irony of Anthropic ending up being Misanthropic,\u201d he wrote. \u201cYou were doomed to this fate when you chose your name.\u201d<\/p>\n Smith brushes off those comments now. \u201cI generally only react to what people say when they\u2019re actually in a room with you and in real conversations, versus what is posted on X and other forums,\u201d he said. \u201cSo rather, just deal with what’s in front of me.\u201d He clarifies he wasn\u2019t literally in the room when the deal was negotiated. That was Anthropic cofounder Tom Brown and CFO Krishna Rao, he says.<\/p>\n More From ForbesForbesFortunes Of Anthropic\u2019s Seven Cofounders More Than Double After Blockbuster FundraiseBy Richard Nieva<\/a>ForbesSpaceX\u2019s IPO Just Made Elon Musk The World\u2019s First TrillionaireBy Matt Durot<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"Anthropic disabled its new model Fable 5 late on Friday. NurPhoto via Getty Images Late on Friday, Anthropic…\n","protected":false},"author":2,"featured_media":73083,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[8],"tags":[40318,53,3261,6409,3623,7444,38107,353,40317,31697],"class_list":["post-73082","post","type-post","status-publish","format-standard","has-post-thumbnail","category-anthropic","tag-amjad-masad","tag-anthropic","tag-dario-amodei","tag-david-sacks","tag-department-of-war","tag-dow","tag-fable-5","tag-mythos","tag-paul-smith","tag-replit"],"_links":{"self":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/73082","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/comments?post=73082"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/posts\/73082\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media\/73083"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/media?parent=73082"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/categories?post=73082"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/ai\/wp-json\/wp\/v2\/tags?post=73082"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}