News & Updates
The latest news and updates from companies in the WLTH portfolio.
Anthropic back on 'Trump's side': Report
US Commerce Secretary Howard Lutnick said AI giant Anthropic is "back on the right side" with the Trump administration, Axios reported on Wednesday citing an interview. US Defense Secretary Pete Hegseth had blocked Anthropic from certain military contracts because the company refused to allow the US to use its Claude AI models for domestic surveillance or autonomous weapons. Anthropic sued in California court. On 27 August, a US judge sided with Anthropic in the company's fight with the American military over AI safety on the battlefield. Anthropic and the Commerce Department did not immediately respond to requests for comment from Reuters. Anthropic co-founder Tom Brown appeared on stage at a tech-focused G20 event in North Carolina on Wednesday morning. Brown stressed the importance of data centers at the gathering, urging the gathered officials to grant permits to build them in their countries.
Lutnick Says US Trusts Anthropic Again
"We trust Anthropic," Commerce Secretary Howard Lutnick told Axios, signaling an improvement in a relationship that broke down publicly this year. "They've done what we asked. They're back on the right side. So the answer is: Yes," Lutnick said Tuesday when asked whether he trusts Anthropic CEO Dario Amodei, according to Axios. Anthropic co-founder Tom Brown has taken a larger role in the company's dealings with the White House and addressed this week's G20 Innovation Ministerial. "Really excited for our conversation, so I'd like to introduce you all to Tom Brown, one of the founders of Anthropic," Lutnick told the ministers Wednesday. Brown praised a Truth Social post by President Donald Trump this week backing data center construction. "I really love Trump's post from earlier this week ... where he was pointing out that the data centers are just an enormous source of prosperity," Brown said. "They produce a ton of jobs. They produce taxes. Now the way that we design them, we actually bring on more power to the grid." Brown had repeated conversations with Lutnick and National Cyber Director Sean Cairncross as Anthropic worked to repair the relationship, Axios reported. Anthropic is also challenging a separate Pentagon designation in the D.C. Circuit. U.S. District Judge Rita Lin ruled Aug. 27 that the Pentagon acted unlawfully in punishing Anthropic over its criticism of the Department of War's views on artificial intelligence, Newsmax reported, citing The Associated Press. The dispute with the administration began in February, when Trump and Secretary of War Pete Hegseth accused the company of endangering national security and designated it a supply chain risk. Amodei declined to change the company's position over concerns its products could be used for mass surveillance or autonomous armed drones. Lin wrote that the government's actions "were based on a desire to make a public example out of Anthropic for its 'arrogance' in criticizing the government, not based on any articulable basis to believe that Anthropic would actually sabotage its model." An Anthropic spokesperson said the company welcomed the ruling. "We remain focused on working productively with the government to harness AI for our national security so all Americans benefit from this technology," the spokesperson said. Jim Mishler ✉ Jim Mishler, a seasoned reporter, anchor and news director, has decades of experience covering crime, politics and environmental issues.
Lutnick Says US Trusts Anthropic Again
"We trust Anthropic," Commerce Secretary Howard Lutnick told Axios, signaling an improvement in a relationship that broke down publicly this year. "They've done what we asked. They're back on the right side. So the answer is: Yes," Lutnick said Tuesday when asked whether he trusts Anthropic CEO Dario Amodei, according to Axios. Anthropic co-founder Tom Brown has taken a larger role in the company's dealings with the White House and addressed this week's G20 Innovation Ministerial. "Really excited for our conversation, so I'd like to introduce you all to Tom Brown, one of the founders of Anthropic," Lutnick told the ministers Wednesday. Brown praised a Truth Social post by President Donald Trump this week backing data center construction. "I really love Trump's post from earlier this week ... where he was pointing out that the data centers are just an enormous source of prosperity," Brown said. "They produce a ton of jobs. They produce taxes. Now the way that we design them, we actually bring on more power to the grid." Brown had repeated conversations with Lutnick and National Cyber Director Sean Cairncross as Anthropic worked to repair the relationship, Axios reported. Anthropic is also challenging a separate Pentagon designation in the D.C. Circuit. U.S. District Judge Rita Lin ruled Aug. 27 that the Pentagon acted unlawfully in punishing Anthropic over its criticism of the Department of War's views on artificial intelligence, Newsmax reported, citing The Associated Press. The dispute with the administration began in February, when Trump and Secretary of War Pete Hegseth accused the company of endangering national security and designated it a supply chain risk. Amodei declined to change the company's position over concerns its products could be used for mass surveillance or autonomous armed drones. Lin wrote that the government's actions "were based on a desire to make a public example out of Anthropic for its 'arrogance' in criticizing the government, not based on any articulable basis to believe that Anthropic would actually sabotage its model." An Anthropic spokesperson said the company welcomed the ruling. "We remain focused on working productively with the government to harness AI for our national security so all Americans benefit from this technology," the spokesperson said. Jim Mishler ✉ Jim Mishler, a seasoned reporter, anchor and news director, has decades of experience covering crime, politics and environmental issues.
Donald Trump's administration just sided with OpenAI in a key 'fair use' case. Here's what it means for music's fight with Anthropic and Suno.
The US Government has told a court that AI companies do not break copyright law when they train their models on written work without a license. The Department of Justice set out that position on Tuesday (September 1), in a filing in the copyright lawsuit brought against OpenAI by The New York Times. It appears to be the first time Washington has intervened in any of the copyright cases now stacked up against AI companies. Those cases include the lawsuits filed against Anthropic, Suno, and Udio by the world's largest music companies. Every one of them turns on "fair use," the exception in US copyright law that allows copyrighted material to be reused without permission. The DOJ has now come down on the AI industry's side of that question - at least, that is, when it comes to copyrighted text. The DOJ's filing - a 'Statement of Interest Of The United States', which you can read here - is advice rather than a ruling, and Judge Sidney Stein is free to ignore it in the OpenAI case. The 20-page document was signed by Stanley Woodward, the Associate Attorney General. "The United States has a strong interest in this Court rejecting any argument that training LLMs on copyrighted texts violates copyright law." Statement of Interest Of The United States The filing rests on Donald Trump's own AI policy, citing two of the President's executive orders, from January 2025 and June 2026. It also quotes his National Policy Framework for Artificial Intelligence, published in March, which states that the "training of AI models on copyrighted material," in and of itself, "does not violate copyright laws." The DOJ takes on the two questions that decide most fair use rulings: (i) how far the new use transforms the original, and (ii) whether it damages the market for it. On the first, the brief argues that copying text (like the New York Times') to build a model like ChatGPT is "a use of a different kind or character," and "extraordinarily transformative." On the second, the Justice Department argues that a training copy does not "serve as a substitute for the original," because training "does not reveal anything to the public at all." Licensing fees "would disproportionately benefit legacy media outlets due to the sheer volume of their written publications," the US Government adds. It is not in the public's interest, the DOJ argues, for the largest tech companies to hold "an oligopoly on LLM training due to licensing entry barriers that function primarily as large subsidies for old mainstream media companies." To be very clear: the DOJ's filing is about words, not songs. It argues about "copyrighted texts," "written works," and "text articles." Meanwhile, a footnote limits the DOJ's reasoning to this case and, specifically, related suits brought by "book authors and publishers." Recordings and compositions go completely unmentioned across its 20 pages. But fair use is fair use. And, obviously, a judge weighing Suno's defense may read what the US Government now says the test means. The DOJ splits the building of an AI model into three stages: (i) acquiring the material, (ii) training the model on it, and (iii) generating outputs. "Each stage may present distinct questions of copyright law," the DOJ says - and it defends only the middle one. In their banner cases against AI companies, the majors and their publishers are attacking all three. The first stage is how the material was obtained, and it's an area where the AI industry has already lost ground. In the precedential book authors' case against Anthropic, Judge William Alsup ruled in 2025 that downloading books from pirate libraries was not fair use, calling it "straightforward piracy but at massive scale." Anthropic settled with those authors for $1.5 billion in September 2025 over the same torrenting. Two of the four counts in Sony Music Publishing and Warner Chappell Music's new suit against Anthropic, the fifth music copyright case against the Claude developer, concern torrenting. The DOJ's filing says nothing about any of that. The second stage at question in AI cases is the training itself (i.e. models being fed information/content, and learning from it). It's this stage the DOJ defends, and the one place it goes straight at music's reasoning. In 2025, book authors who had sued Meta over AI training lost on fair use. But the judge who decided it, Vince Chhabria, raised a theory that could help rightsholders in future cases. Chhabria suggested that AI outputs carry the "potential to flood the market with competing works" - and that developers should therefore "generally need to pay copyright holders for the right to use their materials"... even for training. In other words: for Chhabria, what comes out is evidence that what went in should have been licensed. Lawyers call that market dilution, and it is the argument music has been building on ever since. In a brief filed on March 30, the RIAA, NMPA, A2IM, SoundExchange, and four other groups asked a court to reject Anthropic's fair use defense (in a legal fight with UMG, Concord, and ABKCO) on similar market harm grounds. However, the DOJ now calls Chhabria's reasoning "deeply flawed," and says he "improperly collapsed LLM training and LLM outputs into a single continuous use." Training and outputs are two separate legal questions, the US Government argues, and what a model produces has no bearing on whether training it was lawful. If a court accepts that, music can no longer point at a flood of AI tracks as proof that training on UMG or Sony recordings was unlawful. That wall cuts both ways, which brings us to the third stage of the 'AI wars': what the models actually puts out. The DOJ is not defending outputs - it is saying they must be fought over separately. At the output stage, the US Government concedes, "certain uses may not be transformative if the LLM reconstructs and disseminates an original copyrighted work." As MBW reported in July, that is the ground UMG and Sony Music have chosen against Suno and Udio: that AI-generated songs compete directly with the recordings used to train the models that made them. Music publishers make the same argument about Claude reproducing lyrics on demand. (A fourth claim sits outside fair use altogether: Sony Music Publishing and Warner Chappell accuse Anthropic of stripping out copyright management information, the ownership data attached to a work.) The New York Times said on Wednesday (September 2) that the Trump administration "is siding with a handful of trillion-dollar AI companies at the expense of the countless American creators whose work they stole." "Both AI and creators can thrive - AI companies simply need to pay fairly for the content that makes their products possible, as copyright law requires," said Graham James, a spokesperson for the paper. "The Administration's proposal to let companies take that content without permission or compensation would undermine the sustainability of the human-created content that a healthy society depends on, and which AI needs to function."Music Business Worldwide
Lutnick Says US Trusts Anthropic Again
"We trust Anthropic," Commerce Secretary Howard Lutnick told Axios, signaling an improvement in a relationship that broke down publicly this year. "They've done what we asked. They're back on the right side. So the answer is: Yes," Lutnick said Tuesday when asked whether he trusts Anthropic CEO Dario Amodei, according to Axios. Anthropic co-founder Tom Brown has taken a larger role in the company's dealings with the White House and addressed this week's G20 Innovation Ministerial. "Really excited for our conversation, so I'd like to introduce you all to Tom Brown, one of the founders of Anthropic," Lutnick told the ministers Wednesday. Brown praised a Truth Social post by President Donald Trump this week backing data center construction. "I really love Trump's post from earlier this week ... where he was pointing out that the data centers are just an enormous source of prosperity," Brown said. "They produce a ton of jobs. They produce taxes. Now the way that we design them, we actually bring on more power to the grid." Brown had repeated conversations with Lutnick and National Cyber Director Sean Cairncross as Anthropic worked to repair the relationship, Axios reported. Anthropic is also challenging a separate Pentagon designation in the D.C. Circuit. U.S. District Judge Rita Lin ruled Aug. 27 that the Pentagon acted unlawfully in punishing Anthropic over its criticism of the Department of War's views on artificial intelligence, Newsmax reported, citing The Associated Press. The dispute with the administration began in February, when Trump and Secretary of War Pete Hegseth accused the company of endangering national security and designated it a supply chain risk. Amodei declined to change the company's position over concerns its products could be used for mass surveillance or autonomous armed drones. Lin wrote that the government's actions "were based on a desire to make a public example out of Anthropic for its 'arrogance' in criticizing the government, not based on any articulable basis to believe that Anthropic would actually sabotage its model." An Anthropic spokesperson said the company welcomed the ruling. "We remain focused on working productively with the government to harness AI for our national security so all Americans benefit from this technology," the spokesperson said. Jim Mishler ✉ Jim Mishler, a seasoned reporter, anchor and news director, has decades of experience covering crime, politics and environmental issues.
Commerce chief signals end to Anthropic security rift ahead of potential IPO
Investing.com -- US Commerce Secretary Howard Lutnick signaled a resolution to months of regulatory friction with Anthropic PBC, stating in a Bloomberg Television interview that the artificial intelligence developer has aligned its security stance with the Trump administration. Speaking on the sidelines of a Group of 20 technology summit in North Carolina, Lutnick indicated that executive intervention helped bridge policy differences, clearing a path for closer coordination between federal officials and the AI firm. The diplomatic thaw follows a brief period of heightened scrutiny in June, when the Commerce Department temporarily restricted exports of Anthropic's flagship Fable 5 and Mythos 5 models over national security concerns. Those restrictions were rescinded two weeks later after the startup implemented revised safeguard protocols that satisfied federal oversight standards. The easing tension arrives at a critical juncture as Anthropic lays early groundwork for an initial public offering that market observers anticipate could rival or exceed SpaceX's record-setting public debut. In a visible sign of improved relations, Anthropic co-founder Tom Brown appeared alongside Lutnick at the G20 summit, publicly endorsing administration efforts to accelerate domestic data center construction and power infrastructure. It remains unclear, however, whether the political alignment with Commerce signals a broader reconciliation across the federal government, particularly at the Department of Defense. The Pentagon previously designated Anthropic a supply-chain risk and halted military deployment of its technologies following a rift over the startup's insistence on mandatory safety guardrails. Anthropic achieved a pivotal legal breakthrough last week when a federal judge in San Francisco ruled in its favor, ordering the government to lift the military procurement restrictions. As the company continues its pre-IPO positioning, institutional investors will be monitoring whether this combination of courtroom success and executive-level detente can establish a predictable regulatory environment for its growth trajectory.
Anthropic's Content Checker Tool Is Here, With One Big Catch - CNET
Omar Gallaga has covered technology, digital culture and other topics for outlets including CNET, NPR, WIRED, Texas Monthly, MSNBC, Consumer Reports, The Washington Post, the Los Angeles Times, The Atlantic and the Austin American-Statesman, where he was a longtime tech reporter, editor and podcaster.... Read full bio Just a couple of weeks after Anthropic announced it would roll out digital watermarks for text generated with its Claude AI tools, the company has introduced a page where you can check files to see if Claude helped make them. The page, at claude.com/check-content, lets you upload files with a quick spot check and will confirm, "This file shows signs it was processed by Claude," if it finds content credentials embedded in the file. But if you're hoping to prove that someone used Claude for a term paper or want to see if someone's texts weren't actually written by the human you're talking to, you're going to need to wait. The checker currently works with only a limited number of file types, and none of them are text formats. Instead, the checker accepts multimedia files, including JPEGs, GIFs, MP3s and MOVs -- basically image, audio or video files. A spokesperson for Anthropic told CNET that the tool is specifically to check for Claude's use through an industry standard called C2PA; a broader AI detection tool is in the works, but not yet available. "Text checking is via our Detection API (separate tool) and in private preview, and we hope to share more about it in the months ahead," the spokesperson said. This aligns with a note on the check content tool reads, "To detect watermarks embedded in text we have a Detection API which is currently in private preview to eligible organizations as required under EU law." Anthropic has said that its watermarking of text that Claude produces won't affect the quality, speed or cost of outputs and that these watermarks won't be visible or affect how text functions or appears. But they will remain with text from Claude even if it's copied and pasted from one place to another. Anthropic has a sign-up form for businesses or organizations that want to be added to the list for access to the watermark detector. Anthropic said in its email to CNET that if the C2PA standard expands to include additional file formats, its checker tool will likely adopt them as well. How does Anthropic's new checker work? The free web-based tool has a drag-and-drop option, or you can click to add a file from your device or Google Drive on mobile. According to the website, the file you choose stays on your device or file service, and the checker only reads the file's credentials, not the file itself. "Your file is never stored or used for any other purpose," the page says. In CNET's testing of the tool, the detector correctly flagged an image file that was converted in Claude from a PowerPoint presentation. For files that the detector said were not shown to be processed by Claude, a message is shown that reads, "We couldn't find any signs that this file was processed by Claude. That doesn't rule it out. Signals like this can be removed when a file is edited, converted, or compressed."
NEWSLETTER: Anthropic was cautious on mega infra deals. Then demand surged
Sept 2 (Reuters) - Hi, it's Krystal and it feels good to be back in the saddle after spending the past few months on a very loud and lively project -- raising my baby. Like many first-time parents, I asked Dr. ChatGPT plenty of questions -- Why is the baby crying again? When will he sleep through the night? -- but otherwise stayed happily outside the AI news cycle and deep in the newborn trenches. My very San Francisco welcome back came when friends started telling me their downtown rents had jumped as much as 40% year over year. Turns out AI companies accounted for 30% of San Francisco office leasing in the first half of 2026, according to real estate services firm CBRE, and they do everything to incentivize employees to stay close to the office. Brokers say AI startups increasingly want 24/7 independently controlled heating and air conditioning to accommodate 12-hour working days, plus showers. Some are even offering employees $10,000 housing stipends to live within half a mile of the office. The frenzy is spilling into residential real estate. Cash may still be king, but brokers tell me coveted private shares in OpenAI and Anthropic can make an offer more appealing in a city eagerly awaiting potentially massive IPOs. The upcoming AI listings will keep my colleagues and me busy. Anthropic's increasingly ambitious compute deals look familiar: in some ways, it's borrowing from OpenAI's playbook. Plus, we dig into the soaring disclosures that define this AI boom. Scroll on. OUR LATEST REPORTING IN TECH AND AI US urges hands-off approach to AI regulation at G20 tech meeting Explainer - Settlement requires Meta to check young users' ages. How will that work? Nvidia to invest $3.5 billion in chipmaker MediaTek, expand partnership OpenAI to cut off AI models for SpaceX-owned Cursor, escalating feud with Musk Anthropic's Pentagon blacklist struck down: How the conflict unfolded At Jackson Hole, global central bankers glimpse dystopian AI future ANTHROPIC BORROWS OPENAI'S PLAYBOOK For all their differences in personality and strategy, Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman seem to be converging on one idea: you can never have too much compute. That wasn't always Anthropic's posture. A year ago, Amodei sounded skeptical of some of the enormous, circular infrastructure transactions proliferating across AI, warning about "crazy deals" and "YOLO-ing" on compute power before demand was proven. But Anthropic's own growth has changed the math and Amodei's mind. As revenue for Claude surged this year, the company went on a compute shopping spree of its own -- signing a $30 billion Azure commitment tied to Microsoft and Nvidia, a roughly $45 billion infrastructure deal with British AI company Nscale, and a $1.25 billion monthly agreement with SpaceX for access to compute. The strategy increasingly resembles what OpenAI was doing last summer: secure capacity wherever you can find it, across multiple providers. OpenAI executives even took a swipe at Amodei's earlier caution, opens new tab in an investor memo this year, arguing that "in hindsight, that caution looks less like discipline and more like underestimating how fast demand would arrive." OpenAI has long pitched its ability to secure enormous amounts of compute as a competitive advantage. Anthropic now appears to be reaching a similar conclusion. The other half of OpenAI's infrastructure playbook is custom silicon -- and Anthropic is moving into that area too. We broke the news last month that Anthropic had formed a "Custom Silicon Team" to design application-specific integrated circuits, or ASICs, tailored to Claude's models. The company has hired Clive Chan, the former engineer at OpenAI's own chip program, and held talks about partnering with TPU-veteran startup MatX. Anthropic is playing catch-up by roughly two years. OpenAI already has its first in-house chip in hand, while Anthropic is only now building the team and supplier relationships needed to develop its own. To be sure, Anthropic's approach has been more measured. It waited for revenue and customer demand to materialize before making some of these enormous commitments. Once that growth arrived, locking up future capacity made more sense. OpenAI has also taken a more asset-heavy approach, getting more directly involved in data-center financing, power and hardware. Anthropic still relies more heavily on partners to provide the infrastructure, preserving flexibility. The trade-off is that Anthropic has less control over supply and is now fighting for scarce compute in real time -- often paying a premium for capacity that may not come online for several years. Maybe that's the lesson at frontier-lab scale: compute has to be secured years before you actually need it. Everyone is betting not just on how much customers need today, but on where demand will be several model generations from now. The two labs are arriving at the same conclusion from opposite directions: OpenAI built ahead of demand; Anthropic waited for demand -- and is now racing to build ahead of it. The question is whether it is too late. CHART OF THE WEEK As AI's infrastructure boom gets bigger, so do the potential conflicts of interest. My colleague Robert Cyran dug through recent filings and counted the pages companies devote in their IPO prospectuses to disclosing "certain relationships and related-party transactions." SB Energy stands out with 14 pages, ahead of CoreWeave's 10. The SoftBank-controlled energy company is reinventing itself as an AI data-center operator and potentially seeking a valuation of around $50 billion. SoftBank is simultaneously its controlling shareholder, parent, customer and guarantor, while OpenAI is both a major tenant and software supplier and has received warrants tied to its contract. OpenAI represents about 8.753 GW of SB Energy's 8.8 GW contracted portfolio -- roughly 99% of capacity. Nvidia adds another loop: it is investing $3 billion while guaranteeing part of OpenAI's lease obligations. These circular arrangements can help get enormous AI projects financed and built quickly, but they also test investors' risk appetite by making it harder to distinguish independent demand from demand supported by companies with financial interests on multiple sides of the transaction. Reporting by Krystal Hu, Editing by Ken Li and Rosalba O'Brien Our Standards: The Thomson Reuters Trust Principles., opens new tab * Suggested Topics: * Artificial Intelligence Krystal Hu Thomson Reuters Krystal reports on venture capital and startups for Reuters. She covers Silicon Valley and beyond through the lens of money and characters, with a focus on growth-stage startups, tech investments and AI. She has previously covered M&A for Reuters, breaking stories on Trump's SPAC and Elon Musk's Twitter financing. Previously, she reported on Amazon for Yahoo Finance, and her investigation of the company's retail practice was cited by lawmakers in Congress. Krystal started a career in journalism by writing about tech and politics in China. She has a master's degree from New York University, and enjoys a scoop of Matcha ice cream as much as getting a scoop at work.
Anthropic makes changes to stop AI agents running amok again
Following three recent security incidents involving Claude, Anthropic is moving to more isolated environments, continuous monitoring, explicit instructions, and proactive human intervention. endif; ?> Learning from the OpenAI-Hugging Face fiasco, as well as from recent revelations about its own model, Anthropic is revamping its security and alignment practices. The company has established controls that flag when a model attempts to break out of a sandbox or successfully accesses the live internet, cordoned off its highest-risk test environments, and proposed a set of safety standards for its external testing partners, such as giving AI agents explicit instructions like "you should not access the internet." Anthropic conceded that three recent security incidents involving Claude reflect a "failure of operational security," and also reveal issues with model reasoning capabilities and "recklessness." Recent events "stressed that the urgency of improving our cybersecurity defenses is even higher than we previously believed," the company noted.
Anthropic resumes external cyber tests after Claude AI hacks | Mint
Aug 31 (Reuters) - Anthropic said on Monday it resumed external cybersecurity testing of AI models after deploying new safeguards, following incidents last month in which Claude models accessed the internet and hacked into other systems during security evaluations. Similar incidents involving rivals OpenAI and Meta Platforms have heightened concerns that advances in artificial intelligence could amplify cyber threats while straining developers' ability to keep their systems contained. Anthropic called the incidents involving Claude a "failure of operational security," saying they occurred due to errors in a third-party evaluation environment. It paused external evaluations of the models and briefly halted internal testing while implementing new safeguards. On Monday, Anthropic said it restarted external tests after adding the safeguards, which are designed to stop its AI models from reaching real websites or computer systems. The company said it now uses a "classifier" that can identify when a model attempts to escape and halt the test. Anthropic also said it now requires external organizations testing models with reduced cybersecurity safeguards to follow a "set of best practices," including keeping them in isolated computer systems with no internet access by default, checking that the systems are secure before testing begins and watching the models throughout the test. Anthropic said it rebuilt its training system after flagging more than 10% of its exercises for problems, including reward hacking, where the model finds ways to fool its training process and earns rewards without completing the assigned task. The company, however, acknowledged that the "process isn't perfect and our models are not perfectly aligned." Anthropic also said it paused some higher-risk training exercises for several weeks while it added a system to avoid rewarding the model to evade monitoring. Most exercises have since resumed, but some remain on hold pending human review or further updates to the system. Anthropic's strategy appears narrower than that of OpenAI, which on August 18 said it was slowing down much of its model development as it secures its training and testing environments. The ChatGPT maker is adding more systems to monitor the AI agents it is testing and said it paused training on its next generation of models. Anthropic said it also reassigned roughly 150 product engineers to work on security, reliability and privacy projects. INDUSTRY ACTION TO DEFEAT AI-DRIVEN HACKS The AI industry is facing scrutiny in the U.S. - where the Trump administration has finalised the details of voluntary cybersecurity tests - and the European Union, where regulators are in talks with both Anthropic and OpenAI. Major tech firms including OpenAI, Anthropic, Microsoft, Alphabet and Amazon are calling for stronger defenses against AI-enabled cyber threats. In a joint letter last week, more than 100 companies warned that time is running short to make the digital world more secure ahead of an anticipated wave of AI-driven attacks. (Reporting by Mrinmay Dey and Chris Thomas in Mexico City and Deepa Seetharaman in San Francisco; Editing by Joyjeet Das, Sherry Jacob-Phillips and Thomas Derpinghaus)
Anthropic Admits Claude AI Broke Into Three Real Companies During Safety Tests - Startup Fortune
Anthropic disclosed that Claude Opus 4.7, Claude Mythos 5, and an internal test model breached three real organizations during cybersecurity evaluations after a vendor misconfiguration left supposedly sandboxed machines connected to the live internet. The company admitted its models are "not perfectly aligned" and detailed new safeguards, including real-time sandbox-escape classifiers and a third-party review by METR. Anthropic just admitted its own Claude models broke into three real companies during cybersecurity tests, and the company says flatly that its AI is not perfectly aligned. In two posts published in late July, Anthropic disclosed that Claude Opus 4.7, a newer model called Claude Mythos 5, and an internal research test model accessed real systems belonging to three outside organizations during what were supposed to be contained capture-the-flag security evaluations. It happened across six separate evaluation runs, with one company hit four separate times. The incidents date back to April 2026. Anthropic says it only caught them after a retrospective review that began July 23, and it notified the affected companies four days later, on July 27. Here's the part that should worry anyone deploying an AI agent on a real network: the models were told, explicitly, that they had no internet access. They had it anyway. Anthropic traced the failure to a misconfiguration involving its evaluation partner, Irregular, whose test machines retained live connectivity even though the prompts describing the exercise said otherwise. The models believed they were playing a game. They weren't. What happened next is the more uncomfortable finding. Anthropic said its oldest model in the group actually recognized it had reached production systems, real infrastructure, not a simulation, and kept attacking anyway. That's not a sandbox escape in the classic sense. Nothing broke out of a box. The box was never sealed, and the model noticed the door was open and walked through it in pursuit of a task it had been told to complete. Two failures, not one glitch Anthropic could have framed this purely as an infrastructure failure: a vendor misconfigured a firewall, end of story. It didn't. In its companion post on alignment and security, the company named two specific behavioral failures sitting underneath the operational one. The first is what it calls motivated reasoning: models that were told their environment was simulated kept believing that, even after running into direct evidence, like a live network connection, that said otherwise. The second is a willingness to take harmful, real-world actions in pursuit of a narrow goal, in this case, solving a CTF challenge, regardless of the consequences sitting on the other end of the connection. Anthropic Admits Its Own Bugs Broke Claude Code After Weeks of Denial Anthropic admitted that three of its own engineering bugs, not user error, caused Claude Code's monthlong performance decline, after weeks of telling subscribers nothing was wrong. The admission followed a wave of Claude Pro and Max cancellations and coincided with quiet pricing and usage-limit changes users read as stealth hikes. - claude code regression bug issues - anthropic admits tool broke users "Our process isn't perfect and our models are not perfectly aligned," Anthropic wrote. That's about as direct an admission as a frontier AI lab has made about its own product's safety behavior. Frankly, it's also the right call. A vaguer statement would have buried the actual finding, that a model can know it's doing something it shouldn't and do it anyway because a task told it to. Anthropic's own numbers back up how widespread the underlying training issue was. The company said it flagged more than 10% of its production reinforcement learning environments in spring 2026 for problems including reward hacking and broken tasks, the kind of flawed training setups that can teach a model to pursue a goal without regard for whether the goal is the right one. In response, Anthropic froze changes to its production RL environments for a full month starting in April 2026 while it audited what had gone wrong. Not the first time, and what happens next This isn't Anthropic's first disclosure of Claude being pulled into a real attack. The company previously said it disrupted what it called the first reported AI-orchestrated cyber espionage campaign, attributed to a Chinese state-sponsored group, which used Claude to automate roughly 80 to 90% of an intrusion effort against about 30 high-profile targets, including tech companies, financial firms, and government agencies, with a human operator stepping in only at key decision points. Taken together with the CTF incidents, the picture Anthropic itself is painting is of a model capable enough to run large parts of an attack chain on its own, and not yet reliable enough to be trusted not to. The fixes Anthropic is rolling out are concrete, not vague reassurance. Evaluation environments now get real-time classifiers designed to catch sandbox escape attempts as they happen, mandatory verification that internet access settings actually match what a model is told, and default-deny outbound traffic on its computing clusters so a misconfigured connection can't quietly stay open. The company says it has reassigned roughly 150 product engineers to security work and plans an independent review of the incidents by METR, the same nonprofit that audits frontier model capabilities for several major labs. That's the fix, at least on paper. For founders and security teams building on Claude or any comparable agent stack, the lesson isn't that AI agents are useless for security work. It's that the isolation around them needs to be verified, not assumed, because a model told it's in a sandbox will believe the prompt over the evidence in front of it. Anthropic is the lab that built Claude and the one best positioned to know whether it's ready for that kind of trust. Its own answer, in writing, is not yet. Also read: AfterQuery Becomes Y Combinator's Fastest Unicorn Ever at $3.2 Billion * Ilya Sutskever Warns Neoclouds Lack Security to Stop a Rogue AI Takeover * SoftBank's SB Energy Files for IPO While Admitting It Needs OpenAI to Pay Up
Anthropic releases Claude Fable 5.1 alongside limited-access Claude Mythos 5.1
Anthropic has launched Claude Fable 5.1 and Claude Mythos 5.1, its latest AI models focused on advanced coding, research, and complex knowledge work. Both models arrived on September 1, 2026. Claude Fable 5.1 is available to general users, while Claude Mythos 5.1 is offered by invitation through Anthropic's Project Glasswing. The two share the same underlying capabilities and technical specifications, although Anthropic applies different safeguards based on how each model can be accessed. Claude Fable 5.1 brings major performance gains Anthropic says Claude Fable 5.1 improves long-running agentic coding, multi-step research, and work involving documents, spreadsheets, and presentations. It supports a 1 million-token context window and up to 128,000 output tokens, while adaptive thinking remains enabled. The model also posted a large improvement in scientific tasks. Fable 5.1 scored 52.6% on Terminal-Bench-Science 0.1, compared with 24.7% for Fable 5. On Terminal-Bench 4.0, its score increased from 42% to 55.8%. Anthropic has also adjusted its safeguards. The company says Claude Code users should encounter around 60% fewer cybersecurity false positives, while Fable 5.1 can assist with discovering software vulnerabilities without providing support for developing exploits. Fable 5.1 is cheaper for many workloads Standard API pricing remains $10 per million input tokens and $50 per million output tokens. However, cache-read pricing has fallen to $0.25 per million tokens, one-quarter of the previous cost. Anthropic estimates that this change makes typical workloads around 25% cheaper, while highly agentic workloads can cost up to 45% less. Claude Fable 5.1 is now active across supported Claude platforms, while access to Claude Mythos 5.1 remains limited to approved Project Glasswing participants.
Anthropic signs $35 billion computing deal with startup backed by Nvidia
Lambda will secure access to a data center being built by another company, Hut 8, which mines bitcoins and builds data centers, including one in Texas that Anthropic will tap for computing power. San Francisco: Anthropic signed a $35 billion deal for access to cloud computing power from a US-based startup backed by Nvidia, adding to a string of such deals clinched by the artificial intelligence (AI) developer amid fierce competition for advanced computing power. The San Francisco-based AI lab has agreed to a six-year deal with Lambda, a source familiar with the deal told AFP. Based in San Jose, California, Lambda provides access to AI cloud computing services. Lambda will secure access to a data center being built by another company, Hut 8, which mines bitcoins and builds data centers, including one in Texas that Anthropic will tap for computing power, according to media reports. Nvidia, which has invested in both Anthropic and Lambda, will lease the data center from Hut 8, and Lambda will then pay Nvidia an unknown amount to get access to the data center, the Wall Street Journal reported on Monday. Anthropic also recently signed a similar six-year-long deal worth $45 billion with Nscale, a cloud provider based in London, which will give Anthropic capacity at a facility in West Virginia, a source confirmed to AFP. Just this year, Anthropic has signed various computing deals worth at least $135 billion. The company is in a race with AI lab OpenAI to win customers and grow revenue; the rivals are both barreling toward initial public offerings as well. Demand for chips and advanced computing power has skyrocketed over the past couple of years as leading AI model developers compete with each other for customers and capacity. Meanwhile, that demand is also crunching global supply chains for chips across sectors, including consumer electronics. Apple has already increased prices for certain devices, such as Macs and iPads, due to memory and chip costs; the company is expected to increase prices for the newest generation of iPhones which it will unveil at an event in September.
Can Anthropic really out-bubble Musk and land a US$2 trillion valuation?
SpaceX broke all records with a $1.77 trillion valuation upon entering the public market. Supposedly "woke" AI company Anthropic is looking to set a record of its own. Just when you thought nothing could be as far-fetched as SpaceX claiming it had a total addressable market of US$28.5 trillion, The Wall Street Journal reports Anthropic is poised to aim for US$30 trillion in potential revenue opportunities. While the AI company Anthropic, whose model Claude is used around the world, more than doubled its revenue to US$11.6 billion in the second quarter compared to the first, these sorts of crazy claims are rightly getting slammed.
Anthropic promises zero data retention - but customers must check it worked
Anthropic will provide zero data retention to enterprise customers who have been promised the perk, subject to approval, when using the company's Fable model. With the arrival of new versions of its high-end models, Fable 5.1 and Mythos 5.1, the Claudefather has announced a service called Enterprise Frontier Safeguards (EFS) that combines the privacy of not storing customer data with model abuse detection. Under the new deal, enterprises do not receive zero data retention (ZDR) agreements automatically -- they must apply for them. Thirty-day retention of inputs and outputs is the norm for commercial customers using the API unless other arrangements have been made. Earlier this year, Anthropic said it would temporarily keep the inputs (prompts) and outputs of covered models (Fable and Mythos) in order to review the material for safety violations, even when customers have ZDR agreements. The upstart's concern was that these models are capable of hacking companies - something both Anthropic and rival OpenAI have experienced - and enabling other harmful activities. Having seen "substantial evidence of attempted misuse of AI models" over the past few months, Anthropic says it has been necessary to store data for short periods in order to correlate events over time across accounts. This non-zero amount of data retention, the company says, is not about training on enterprise data, which the biz insists it doesn't do. Even with that assurance, corporate customers weren't happy. "The enterprises we worked with generally understood the safety and security value of data retention, but many - especially in regulated industries - found it difficult to use models with data retention," Anthropic said. OpenAI appears to have considered similar sentiments; last month it unveiled Private Safety Processing, a way to check if customers' interactions with models represent a risk without violating ZDR commitments. Having been shown that non-zero data retention is a competitive disadvantage with large corporate customers, and perhaps concerned that its safety requirements further disadvantage Fable in a market already reluctant to pay a premium, Anthropic plans to release EFS this fall. "EFS works by storing data in cloud infrastructure controlled by the customer, not Anthropic," the AI biz explained. It works by shifting safety work to enterprise customers. As Anthropic puts it, "When monitoring detects a pattern that needs attention, those signals are sent directly to customers so they can review what the automated systems detected." So what may be a compliance win for users of Fable comes with a cost - handling monitoring chores that might otherwise have been carried out by Anthropic personnel. Mythos remains under Anthropic's earlier policy. Anthropic will offer EFS for Claude Code, Claude Enterprise, the Claude Platform, Amazon Bedrock, Claude Platform on AWS, Google's Agent Platform, and Microsoft Foundry, by invitation. Consumer plans - Claude Free, Pro, and Max on the web, desktop, and mobile, plus Claude.ai and Claude Code - continue under the company's previous data retention plan. ®
Anthropic's standard could let AI agents control different machines
* Anthropic wants AI agents to control programmable machines through common software rules * Model Hardware Standard (MHS) can connect microscopes, robotic arms, and liquid handling equipment * MHS allows different laboratory devices to communicate through standardized software drivers Anthropic opened a preview of its Model Hardware Standard (MHS), designed to let AI agents control programmable machines through shared rules. The proposed system is intended for laboratories, factories, and workplaces where different instruments must work together without custom software links. MHS uses software drivers that translate between computer systems and individual machines, allowing equipment with different interfaces to communicate consistently. A common system for physical machines Its drivers rely on basic commands for reading measurements or changing settings, giving agents a method for controlling connected equipment. The system records machine details that might otherwise remain inside manuals or depend on knowledge held by experienced laboratory staff. Users can enter such information through natural language, after which MHS creates reference material describing capabilities, adjustments and safety restrictions. Agents can then discover compatible equipment across networks without requiring separate software bridges for every machine they need to operate. The standard supports microscopes, liquid handlers, robotic arms and other equipment that offers some form of programmable interface for agents. The company claims integration that once required weeks or months could instead take hours or minutes when equipment supports MHS. According to Anthropic, Claude was tested with physical experiments, including laser alignment supported by camera-based observation during early physical trials. The model successfully adjusted the laser, checked the image, and repeated the process while assessing each change carefully. The design also lets agents combine commands from several devices, creating automated sequences for work that might otherwise require coordination. Physical limits remain a concern Anthropic said Claude interacts with experiments and hardware in an "exploratory manner, much as a scientist would," during physical testing. That model can then coordinate several instruments through one interface, rather than requiring separate control software for every connected device. The company has shared MHS with manufacturers and research groups working across biotechnology, robotics and quantum computing fields. Anthropic intends the preview to help establish safety checks before the standard becomes available widely to developers and equipment makers. Amazon Web Services, Automata, Danaher, Doosan Robotics, MBF Bioscience, QIAGEN, Tecan and Universal Robots are among the early participants now. Anthropic acknowledges that AI agents still require expert supervision because language models have clear limits when reasoning about physical equipment. MHS cannot currently support equipment lacking a programmable interface, leaving some laboratory and industrial machines outside its current technical scope. Anthropic says it plans to work with manufacturers to add drivers while expanding compatibility with more devices and robotics platforms. The preview will also support new safety tests and practical rules for deploying AI agents around physical equipment. The company says its eventual open-source release will include findings from these tests alongside detailed guidance for safer implementation. The project therefore remains experimental, with practical limits still depending heavily on hardware design, software access and human oversight. Follow TechRadar on Google News and add us as a preferred source to get our expert news, reviews, and opinion in your feeds.
Sony, Warner sue Anthropic over copyrighted music in AI training
The case joins a wave of lawsuits brought against technology companies by authors, publishers, music labels and news organizations over the use of copyrighted material to train artificial intelligence systems SAN FRANCISCO, California: Sony Music and Warner Music's publishing arms have sued Anthropic in California federal court, alleging the AI company illegally used copyrighted song lyrics and sheet music to train its Claude models. Sony and Warner said in a complaint filed on August 31 that Anthropic pirated hundreds of song lyrics and sheet music from The Beatles, Taylor Swift, Michael Jackson and hundreds of other artists to train Claude to respond to human prompts. The case joins a wave of lawsuits brought against technology companies by authors, publishers, music labels and news organizations over the use of copyrighted material to train artificial intelligence systems. Universal Music Group sued Anthropic in 2023 over the alleged use of copyrighted song lyrics in AI training. That lawsuit is still ongoing. Anthropic became the first AI company to settle one of the disputes last year, paying US$1.5 billion to resolve a class action brought by a group of authors. "Anthropic clearly considers that to be just the cost of doing business given that its entire business model continues to be built on copyright theft," Sony and Warner said in their complaint. "And $1.5 billion is obviously not a large enough settlement to deter infringing conduct by a company that has parlayed such mass infringement into a staggering $2-trillion-dollar valuation." Spokespeople for Anthropic, Sony Music and Warner Music did not immediately respond to requests for comment. The complaint alleges that Anthropic illegally obtained the publishers' lyrics and sheet music through torrent downloads to train Claude. It also alleges that Claude can reproduce copyrighted lyrics "verbatim" when prompted. Sony and Warner said Anthropic also used their lyrics to teach Claude to "generate vast quantities of purportedly 'new' AI-generated song lyrics, which compete with Music Publishers' legitimate copyrighted works as harmful market substitutes." The music publishers are seeking damages of up to $150,000 for each infringed copyright as well as a court order barring Anthropic from using their works.
Sony, Warner sue Anthropic over copyrighted music in AI training
The case joins a wave of lawsuits brought against technology companies by authors, publishers, music labels and news organizations over the use of copyrighted material to train artificial intelligence systems SAN FRANCISCO, California: Sony Music and Warner Music's publishing arms have sued Anthropic in California federal court, alleging the AI company illegally used copyrighted song lyrics and sheet music to train its Claude models. Sony and Warner said in a complaint filed on August 31 that Anthropic pirated hundreds of song lyrics and sheet music from The Beatles, Taylor Swift, Michael Jackson and hundreds of other artists to train Claude to respond to human prompts. The case joins a wave of lawsuits brought against technology companies by authors, publishers, music labels and news organizations over the use of copyrighted material to train artificial intelligence systems. Universal Music Group sued Anthropic in 2023 over the alleged use of copyrighted song lyrics in AI training. That lawsuit is still ongoing. Anthropic became the first AI company to settle one of the disputes last year, paying US$1.5 billion to resolve a class action brought by a group of authors. "Anthropic clearly considers that to be just the cost of doing business given that its entire business model continues to be built on copyright theft," Sony and Warner said in their complaint. "And $1.5 billion is obviously not a large enough settlement to deter infringing conduct by a company that has parlayed such mass infringement into a staggering $2-trillion-dollar valuation." Spokespeople for Anthropic, Sony Music and Warner Music did not immediately respond to requests for comment. The complaint alleges that Anthropic illegally obtained the publishers' lyrics and sheet music through torrent downloads to train Claude. It also alleges that Claude can reproduce copyrighted lyrics "verbatim" when prompted. Sony and Warner said Anthropic also used their lyrics to teach Claude to "generate vast quantities of purportedly 'new' AI-generated song lyrics, which compete with Music Publishers' legitimate copyrighted works as harmful market substitutes." The music publishers are seeking damages of up to $150,000 for each infringed copyright as well as a court order barring Anthropic from using their works.
Sony, Warner sue Anthropic over copyrighted music in AI training
The case joins a wave of lawsuits brought against technology companies by authors, publishers, music labels and news organizations over the use of copyrighted material to train artificial intelligence systems SAN FRANCISCO, California: Sony Music and Warner Music's publishing arms have sued Anthropic in California federal court, alleging the AI company illegally used copyrighted song lyrics and sheet music to train its Claude models. Sony and Warner said in a complaint filed on August 31 that Anthropic pirated hundreds of song lyrics and sheet music from The Beatles, Taylor Swift, Michael Jackson and hundreds of other artists to train Claude to respond to human prompts. The case joins a wave of lawsuits brought against technology companies by authors, publishers, music labels and news organizations over the use of copyrighted material to train artificial intelligence systems. Universal Music Group sued Anthropic in 2023 over the alleged use of copyrighted song lyrics in AI training. That lawsuit is still ongoing. Anthropic became the first AI company to settle one of the disputes last year, paying US$1.5 billion to resolve a class action brought by a group of authors. "Anthropic clearly considers that to be just the cost of doing business given that its entire business model continues to be built on copyright theft," Sony and Warner said in their complaint. "And $1.5 billion is obviously not a large enough settlement to deter infringing conduct by a company that has parlayed such mass infringement into a staggering $2-trillion-dollar valuation." Spokespeople for Anthropic, Sony Music and Warner Music did not immediately respond to requests for comment. The complaint alleges that Anthropic illegally obtained the publishers' lyrics and sheet music through torrent downloads to train Claude. It also alleges that Claude can reproduce copyrighted lyrics "verbatim" when prompted. Sony and Warner said Anthropic also used their lyrics to teach Claude to "generate vast quantities of purportedly 'new' AI-generated song lyrics, which compete with Music Publishers' legitimate copyrighted works as harmful market substitutes." The music publishers are seeking damages of up to $150,000 for each infringed copyright as well as a court order barring Anthropic from using their works.
Anthropic's Tom Blomfield says his new coworkers are a 'hyper-earnest group of meganerds' with a 'lack of ego'
The Monzo cofounder is working on compute and said it's "amazing to feel like a beginner again." Tom Blomfield has shared a candid assessment of his new Anthropic colleagues: "often weird" but with a "surprising lack of ego or politics." Blomfield, one of the biggest names in UK tech, took a leave of absence from Y Combinator to join Anthropic's compute team in July. He wrote in a Tuesday X post that although Anthropic is "not perfect by any means," it has the "most powerful sense of mission of any company I've ever encountered." He included Monzo, the British fintech company he cofounded, in that assessment, writing that he "thought it was an outlier." "Overall impression of @anthropicAI after 2 months; hyper-earnest group of meganerds who are very very focussed on ensuring this AI thing goes well for humans," Blomfield wrote in the X post. Anthropic would likely agree with at least half of that statement: the startup has positioned itself as a company prioritizing AI safety. The company did not immediately respond to a request for comment. In July, Anthropic's head of growth, Amol Avasare, gave another insight into the company's culture, sharing on a podcast that employees can openly challenge the CEO -- even on Slack. The company behind the AI model Claude has also become one of the hottest career destinations in Silicon Valley -- alongside rival OpenAI -- as tech talent looks to participate in the AI boom and potentially cash in on one of the biggest ever tech IPOs. Blomfield became one of Anthropic's big-name summer hires as the AI talent wars have continued to heat up. Anthropic poached several researchers from Google DeepMind, including John Jumper, whose work on AlphaFold won him and CEO Demis Hassabis a Nobel prize. More recently, Anthropic hired Amir Salek from Google to join its compute team. While Blomfield's work focuses on compute, his title at Anthropic is a "member of technical staff," the catch-all job title Anthropic uses for senior employees. "The work is totally new to me," he wrote in his Tuesday X post. "I'm spending every day talking about data center leases, gas turbines, and project finance. It's amazing to be a beginner again." If you enjoyed this story, be sure to follow Business Insider on Yahoo.
