Welcome back to This Week in Stratechery! As a reminder, each week, every Friday, we’re sending out this overview of content in the Stratechery bundle; highlighted links are free for everyone. Additionally, you have complete control over what we send to you. If you don’t want to receive This Week in Stratechery emails (there is no podcast), please uncheck the box in your delivery settings. On that note, here were a few of our favorites this week. Will AI Kill the Web, or Save It? The original web that we know and love was the human web; that’s why advertising was the preferred business model, and Google was the big winner. So what happens when AI Agents are the preferred means to gather information? AI’s don’t care about ads, which means the web needs a new business model; I’ve been a long-time critic of micro-transactions, because they are anti-human, but they might be the solution to an AI-mediated future. Or, to put it another way, AI won’t just save AR and VR, it might save crypto as well. — Ben Thompson Slam Dunks and Moonshots at Google. Google I/O in 2025 was classic Google in two respects. On one hand, it featured a spate of great demos and jaw-dropping displays of new features and capabilities that may or may not ever find a market among mainstream users. That’s become something of a Google I/O tradition. On the other hand, Google also introduced “AI Mode” for search, a collection of AI-powered features that will lean into Google’s advantages in audience and infrastructure and allow the company to gradually integrate more AI features into search. This, too, is classic behavior from a company that for all its excesses and eccentricities still understands its core business and how to move when it’s time to leverage and preserve search dominance. Wednesday’s Daily Update broke it all down, highlighting questions that still linger, as well as AI hardware and software possibilities in Mountain View and beyond. — Andrew Sharp OpenAI Enters the Hardware Business. On Wednesday OpenAI interrupted Google’s big week when it announced plans to acquire Jony Ive’s hardware startup for $6.5 billion worth of stock, telling the world that the company will work with the Apple legend to develop a family of AI-powered devices and design “the coolest piece of technology that the world has ever seen.” If you’re looking to process that news, start with Ben’s analysis on Thursday, which comes with a bonus section on Apple’s dwindling options in AI (and a withering parenthetical on the “Tim Cook Doctrine”). Then head to Sharp Tech, where we had a great time talking through the logic of this move for OpenAI, challenges that lay ahead for Ive’s team, and a few emails from haters who are skeptical of this roadmap and/or despise the aesthetics of the OpenAI PR blitz. All of it was great fun, and whatever happens from here, the next few years just got a lot more interesting. — AS Stratechery Articles and Updates An Interview with Nvidia CEO Jensen Huang About Chip Controls, AI Factories, and Enterprise Pragmatism — An interview with Nvidia CEO Jensen Huang about new problems, including chip controls and China, and new opportunities, including AI Factories and enterprise pragmatism. The Agentic Web and Original Sin — Microsoft is putting forth compelling proposals for the Open Agentic Web. However, the proposal needs digital payments, which will be key to creating a new content marketplace for AI. Google I/O, The Search Funnel, Product Possibilities — Google I/O was impressive and overwhelming, but the only product that impressed was Search. OpenAI Acquires io, OpenAI’s Strategic Positioning, Apple’s Worsening AI Problem — OpenAI is acquiring io, Jony Ive’s hardware company. Is the company actually focused on displacing Apple? Apple needs to spend big in response. Dithering with Ben Thompson and Daring Fireball’s John Gruber Apple Alarm io Asianometry with Jon Yu EUV Photoresists: The Next Generation? The Long, Hard End of the Chilean Nitrates Industry Sharp China with Andrew Sharp and Sinocism’s Bill Bishop Chips and the Geneva Consensus; US Policy and the Chinese Century; Controversy Over Solar Power Inverters Greatest of All Talk with Andrew Sharp and WaPo’s Ben Golliver OKC Survives and Looks Dominant, A New Era in the West, Revisiting the Lottery and Ideas for the Spurs Emergency Pod: A Bing Bong Nightmare, Haliburton Heroics and a Stunning Game 1 in New York City Sharp Tech with Andrew Sharp and Ben Thompson The U.S. Partners with the Middle East in AI, Why OpenAI Is Acquiring Windsurf, Google’s Side of the Platform Wars OpenAI Enters the Hardware Business, The Challenges and Opportunities for Jony Ive, Takeaways from Google I/O 2025 This week’s Sharp Tech video is on Apple’s risks as their App Store battles continue.

Today's report by security expert Jeremiah Fowler of a massive unsecured database full of usernames and passwords shouldn't necessarily frighten you, but it should spur you to action. If you have any weak passwords protecting accounts with sensitive information, or if you've reused the same password — however strong — on multiple accounts, now would be an excellent time to change them and set up two-factor authentication. Fowler reported on Website Planet that the database, which he found unlocked and without any encryption on an anonymously registered server, contained a little over 184 million records. These included usernames, emails, passwords, and direct links to the URLs for logging into the relevant accounts. While Fowler was able to get the hosting provider to lock the server, he couldn't find any hard evidence about who compiled the database, nor whether they had used or shared the information. There are a couple of reasons not to panic here. 184 million records exposed doesn't mean 184 million people exposed — it's just the number of rows in the database. If the info was gathered through malware, as Fowler believes, it's likely to have gathered multiple records from every infected device. That's obviously still bad, but fewer people have been affected than it may seem from the number alone. The database also contained no information that could be used for two-factor authentication, so anyone with a second factor set up has much less reason to worry. Don't forget, though, that one weakly secured account is a liability to the others. For example, a hacker could gain access to your email, then use that access to break through 2FA on your bank account. The potential consequences of having your password stolen are severe enough that it's worth taking common-sense steps. Since the database wasn't leaked on any of the usual dark web sources, its data likely won't show up on breach checkers like HaveIBeenPwned. However, Fowler did share with Wired reporters that he tested a sample of 10,000 fields in the database, and found passwords to the following platforms: Facebook Google Instagram Roblox Discord Microsoft Netflix PayPal Amazon Apple Nintendo Snapchat Spotify Twitter WordPress Yahoo Online banks Online wallets Healthcare web apps Government employee accounts If you have an account on any of those platforms without two-factor authentication, we recommend changing your password and setting up 2FA as soon as possible. Pay special attention to platforms like Roblox and Nintendo where your kids might have set up their own accounts and not bothered with 2FA. As Fowler points out in his blog post, even seemingly innocuous accounts might have personal information lying around. This article originally appeared on Engadget at https://www.engadget.com/cybersecurity/a-huge-unsecured-credential-database-discovery-is-a-great-reminder-to-change-your-passwords-210537400.html?src=rss

Weapon of choice? Musk’s DOGE used Meta’s Llama 2—not Grok—for gov’t slashing, report says Grok apparently wasn't an option. Ashley Belanger – May 22, 2025 5:12 pm | 19 Credit: Anadolu / Contributor | Anadolu Credit: Anadolu / Contributor | Anadolu Story text Size Small Standard Large Width * Standard Wide Links Standard Orange * Subscribers only   Learn more An outdated Meta AI model was apparently at the center of the Department of Government Efficiency's initial ploy to purge parts of the federal government. Wired reviewed materials showing that affiliates of Elon Musk's DOGE working in the Office of Personnel Management "tested and used Meta’s Llama 2 model to review and classify responses from federal workers to the infamous 'Fork in the Road' email that was sent across the government in late January." The "Fork in the Road" memo seemed to copy a memo that Musk sent to Twitter employees, giving federal workers the choice to be "loyal"—and accept the government's return-to-office policy—or else resign. At the time, it was rumored that DOGE was feeding government employee data into AI, and Wired confirmed that records indicate Llama 2 was used to sort through responses and see how many employees had resigned. Llama 2 is perhaps best known for being part of another scandal. In November, Chinese researchers used Llama 2 as the foundation for an AI model used by the Chinese military, Reuters reported. Responding to the backlash, Meta told Reuters that the researchers' reliance on a “single" and "outdated" was "unauthorized," then promptly reversed policies banning military uses and opened up its AI models for US national security applications, TechCrunch reported. "We are pleased to confirm that we’re making Llama available to US government agencies, including those that are working on defense and national security applications, and private sector partners supporting their work," a Meta blog said. "We’re partnering with companies including Accenture, Amazon Web Services, Anduril, Booz Allen, Databricks, Deloitte, IBM, Leidos, Lockheed Martin, Microsoft, Oracle, Palantir, Scale AI, and Snowflake to bring Llama to government agencies." Because Meta's models are open-source, they "can easily be used by the government to support Musk’s goals without the company’s explicit consent," Wired suggested. It's hard to track where Meta's models may have been deployed in government so far, and it's unclear why DOGE relied on Llama 2 when Meta has made advancements with Llama 3 and 4. Not much is known about DOGE's use of Llama 2. Wired's review of records showed that DOGE deployed the model locally, "meaning it’s unlikely to have sent data over the Internet," which was a privacy concern that many government workers expressed. In an April letter sent to Russell Vought, director of the Office of Management and Budget, more than 40 lawmakers demanded a probe into DOGE's AI use, which, they warned—alongside "serious security risks"—could "have the potential to undermine successful and appropriate AI adoption." That letter called out a DOGE staffer and former SpaceX employee who supposedly used Musk’s xAI Grok-2 model to create an "AI assistant," as well as the use of a chatbot named "GSAi"—"based on Anthropic and Meta models"—to analyze contract and procurement data. DOGE has also been linked to a software called AutoRIF that supercharges mass firings across the government. In particular, the letter emphasized the "major concerns about security" swirling DOGE's use of "AI systems to analyze emails from a large portion of the two million person federal workforce describing their previous week’s accomplishments," which they said lacked transparency. Those emails came weeks after the "Fork in the Road" emails, Wired noted, asking workers to outline weekly accomplishments in five bullet points. Workers fretted over responses, worried that DOGE might be asking for sensitive information without security clearances, Wired reported. Wired could not confirm if Llama 2 was also used to parse these email responses, but federal workers told Wired that if DOGE was "smart," then they'd likely "reuse their code" from the "Fork in the Road" email experiment. Why didn’t DOGE use Grok? It seems that Grok, Musk's AI model, wasn't available for DOGE's task because it was only available as a proprietary model in January. Moving forward, DOGE may rely more frequently on Grok, Wired reported, as Microsoft announced it would start hosting xAI’s Grok 3 models in its Azure AI Foundry this week, The Verge reported, which opens the models up for more uses. In their letter, lawmakers urged Vought to investigate Musk's conflicts of interest, while warning of potential data breaches and declaring that AI, as DOGE had used it, was not ready for government. "Without proper protections, feeding sensitive data into an AI system puts it into the possession of a system’s operator—a massive breach of public and employee trust and an increase in cybersecurity risks surrounding that data," lawmakers argued. "Generative AI models also frequently make errors and show significant biases—the technology simply is not ready for use in high-risk decision-making without proper vetting, transparency, oversight, and guardrails in place." Although Wired's report seems to confirm that DOGE did not send sensitive data from the "Fork in the Road" emails to an external source, lawmakers want much more vetting of AI systems to deter "the risk of sharing personally identifiable or otherwise sensitive information with the AI model deployers." A seeming fear is that Musk may start using his own models more, benefiting from government data his competitors cannot access, while potentially putting that data at risk of a breach. They're hoping that DOGE will be forced to unplug all its AI systems, but Vought seems more aligned with DOGE, writing in his AI guidance for federal use that "agencies must remove barriers to innovation and provide the best value for the taxpayer." "While we support the federal government integrating new, approved AI technologies that can improve efficiency or efficacy, we cannot sacrifice security, privacy, and appropriate use standards when interacting with federal data," their letter said. "We also cannot condone use of AI systems, often known for hallucinations and bias, in decisions regarding termination of federal employment or federal funding without sufficient transparency and oversight of those models—the risk of losing talent and critical research because of flawed technology or flawed uses of such technology is simply too high." Ashley Belanger Senior Policy Reporter Ashley Belanger Senior Policy Reporter Ashley is a senior policy reporter for Ars Technica, dedicated to tracking social impacts of emerging policies and new technologies. She is a Chicago-based journalist with 20 years of experience. 19 Comments

The rapid rise of AI agents is sparking both excitement and alarm. Their power lies in their ability to complete tasks with increasing autonomy. Many can already pursue multi-step goals, make decisions, and interact with external systems — all with minimal human input. Teams of AI agents are beginning to collaborate, each handling a specialised role. As their autonomy increases, they’re poised to reshape countless business processes. Tech giants are heralding them as the future of the web. At Microsoft’s Build conference this week, the company declared that we have entered “the era of AI agents.” OpenAI CEO Sam Altman joined the event, proclaiming his lab’s new Codex tool as “a real agentic coding experience.” He called it “one of the biggest changes to programming that I’ve ever seen.” Beyond the hype, practical applications are rapidly emerging. AI agents are already assisting with various tasks, from code generation and cyber threat detection to customer service enquiries and shopping marketing campaigns.  Grab that deal Before long, they could become comprehensive executive assistants — managing your emails, calendar, and projects. But to harness the opportunities, people need to prepare now. Cihangir Kocak is helping them do just that. A principal business and AI strategist at HP, Kocak guides organisations through digital transformation. He believes AI agents will unleash a new wave of opportunities. “We are going to a future where everyone will have an AI agent as an assistant,” he says. At TNW Conference this summer, Kocak will host two sessions on AI agents. On June 19, he’ll deliver a keynote on their rise. The next day, he’ll join Joost Bos, Senior AI Engineer at Deloitte, for a masterclass titled “Agentic AI: Architecting the Future of Business.” Ahead of the event, he shared a few of his tips. 1. Understand what AI agents can do AI agents evolve large language models (LLMs) from passive responders into active problem-solvers. With tools, memory, and defined goals, they can complete complex tasks on their own. “Large language models act as the brains and AI agents as the hands, which means they can also act,” Kocak says. “They can do things for you autonomously.” Agents can also collaborate. One might source products, another handle logistics, a third build your website, and a fourth write the marketing copy. In future, businesses may need their own agents to interact with others. Your AI assistant could collaborate with them to book the best service for your needs. Free courses from the likes of Hugging Face, Salesforce, and Microsoft are good starting points to explore the possibilities. After getting an understanding of the basics, you can put them into practice.  2. Start experimenting Kocak expects AI agents to rapidly reshape workplaces. “I believe that within five years, everything will be changed because of AI agents,” he says. “It might be even much less than five years — maybe two to three years.” Many companies are already shifting numerous tasks from humans to AI. In the near future, the people that they do recruit may require experience of working with AI agents. “Soon, a lot of these companies will ask for people who can work with AI agents,” says Kocak. His advice? “Get your hands dirty. Play with it, experiment with it — but do it consciously.” One tool he recommends is LM Studio, a desktop app for running LLMs locally. But his key recommendation is simply getting started. “Just do something to get a feel of it. Once you have that, it’s time for the next step.” 3. Find use cases After testing some tools, Kocak suggests identifying where they can add value. He advises looking for tasks where AI can free up your time — and start small. “What costs you the most time? What don’t you like to do? When you figure out those things, you can look at how AI agents can help you.” Kocak uses local LLMs for privacy-sensitive tasks, and ChatGPT for public ones — like drafting LinkedIn posts in his own voice. “It saves at least half of my time,” he says. 4. Focus on the data The real magic of AI agents emerges when they’re personalised with your choice of data. Generic tools like ChatGPT can handle broad tasks. But if you want something tailored, agents trained on your choice of data can offer sharper performance. That internal knowledge can turn a generic agent into a bespoke powerhouse. “What makes an AI solution special is when you feed it with your own data,” says Kocak. “Then you will have a solution that can operate differently than anything else.” 5. Maintain human oversight Although AI agents can act autonomously, human oversight remains vital. Agents are powerful, but not flawless. Giving them too much freedom is risky. “It’s wise to have a human in the room,” he says. “The future will be AI agents plus humans — that will be the most beneficial combination.” 6. Stay secure As AI tools become more accessible, security concerns are mounting. Among the threats are data leaks, adversarial attacks, and agents going off the rails. There’s also the risk of losing a competitive edge.  “External parties can take your data and send it to their servers,” says Kocak. “They can then use all sensitive data in your conversations to optimise their models.” Many risks can be reduced by deploying open-source, local models — especially for sensitive data and use cases. “If you really want a competitive advantage, you need to run and own your AI. That sets you apart,” says Kocak. He adds that people shouldn’t be fearful, but conscious. Closed-source, cloud-based tools such as ChatGPT remain useful — but sensitive data and tasks may require more secure alternatives. “Just be aware of what information you enter. And remember there is another, better option, of running your large language model locally.” 7. Embrace the future As the industrial revolution and factory automation did before them, AI agents will transform jobs. Some roles will disappear — but new ones will emerge. A welder could become an operator of robotic welders. A data entry clerk might oversee AI agents. Kocak is optimistic about the possibilities. “Our core capabilities as humans — like being creative, finding solutions out of the box, and empathy — will come to the forefront.” These tips are just a glimpse of what Kocak will provide at TNW Conference. If you want to check out his sessions — or anything else on the event agenda — we have a special offer for you. Use the code TNWXMEDIA2025 at the ticket checkout to get 30% off. Story by Thomas Macaulay Managing editor Thomas is the managing editor of TNW. He leads our coverage of European tech and oversees our talented team of writers. Away from work, he e (show all) Thomas is the managing editor of TNW. He leads our coverage of European tech and oversees our talented team of writers. Away from work, he enjoys playing chess (badly) and the guitar (even worse). Get the TNW newsletter Get the most important tech news in your inbox each week. Also tagged with

DOGE tested and used Meta’s Llama 2 model to review and classify responses from federal workers to the infamous “Fork in the Road” email.

The rapid rise of AI agents is sparking both excitement and alarm. Their power lies in their ability to complete tasks with increasing autonomy. Many can already pursue multi-step goals, make decisions, and interact with external systems — all with minimal human input. Teams of AI agents are beginning to collaborate, each handling a specialised role. As their autonomy increases, they’re poised to reshape countless business processes. Tech giants are heralding them as the future of the web. At Microsoft’s Build conference this week, the company declared that we have entered “the era of AI agents.” OpenAI CEO Sam Altman joined the event, proclaiming his lab’s new Codex tool as “a real agentic coding experience.” He called it “one of the biggest changes to programming that I’ve ever seen.” Beyond the hype, practical applications are rapidly emerging. AI agents are already assisting with various tasks, from code generation and cyber threat detection to customer service enquiries and shopping marketing campaigns.  Before long, they could become comprehensive executive assistants — managing your emails, calendar, and projects. But to harness the opportunities, people need to prepare now. Cihangir Kocak is helping them do just that. A principal business and AI strategist at HP, Kocak guides organisations through digital transformation. He believes AI agents will unleash a new wave of opportunities. “We are going to a future where everyone will have an AI agent as an assistant,” he says. At TNW Conference this summer, Kocak will host two sessions on AI agents. On June 19, he’ll deliver a keynote on their rise. The next day, he’ll join Joost Bos, Senior AI Engineer at Deloitte, for a masterclass titled “Agentic AI: Architecting the Future of Business.” Ahead of the event, he shared a few of his tips. 1. Understand what AI agents can do AI agents evolve large language models (LLMs) from passive responders into active problem-solvers. With tools, memory, and defined goals, they can complete complex tasks on their own. “Large language models act as the brains and AI agents as the hands, which means they can also act,” Kocak says. “They can do things for you autonomously.” Agents can also collaborate. One might source products, another handle logistics, a third build your website, and a fourth write the marketing copy. In future, businesses may need their own agents to interact with others. Your AI assistant could collaborate with them to book the best service for your needs. Free courses from the likes of Hugging Face, Salesforce, and Microsoft are good starting points to explore the possibilities. After getting an understanding of the basics, you can put them into practice.  2. Start experimenting Kocak expects AI agents to rapidly reshape workplaces. “I believe that within five years, everything will be changed because of AI agents,” he says. “It might be even much less than five years — maybe two to three years.” Many companies are already shifting numerous tasks from humans to AI. In the near future, the people that they do recruit may require experience of working with AI agents. “Soon, a lot of these companies will ask for people who can work with AI agents,” says Kocak. His advice? “Get your hands dirty. Play with it, experiment with it — but do it consciously.” One tool he recommends is LM Studio, a desktop app for running LLMs locally. But his key recommendation is simply getting started. “Just do something to get a feel of it. Once you have that, it’s time for the next step.” 3. Find use cases After testing some tools, Kocak suggests identifying where they can add value. He advises looking for tasks where AI can free up your time — and start small. “What costs you the most time? What don’t you like to do? When you figure out those things, you can look at how AI agents can help you.” Kocak uses local LLMs for privacy-sensitive tasks, and ChatGPT for public ones — like drafting LinkedIn posts in his own voice. “It saves at least half of my time,” he says. 4. Focus on the data The real magic of AI agents emerges when they’re personalised with your choice of data. Generic tools like ChatGPT can handle broad tasks. But if you want something tailored, agents trained on your choice of data can offer sharper performance. That internal knowledge can turn a generic agent into a bespoke powerhouse. “What makes an AI solution special is when you feed it with your own data,” says Kocak. “Then you will have a solution that can operate differently than anything else.” 5. Maintain human oversight Although AI agents can act autonomously, human oversight remains vital. Agents are powerful, but not flawless. Giving them too much freedom is risky. “It’s wise to have a human in the room,” he says. “The future will be AI agents plus humans — that will be the most beneficial combination.” 6. Stay secure As AI tools become more accessible, security concerns are mounting. Among the threats are data leaks, adversarial attacks, and agents going off the rails. There’s also the risk of losing a competitive edge.  “External parties can take your data and send it to their servers,” says Kocak. “They can then use all sensitive data in your conversations to optimise their models.” Many risks can be reduced by deploying open-source, local models — especially for sensitive data and use cases. “If you really want a competitive advantage, you need to run and own your AI. That sets you apart,” says Kocak. He adds that people shouldn’t be fearful, but conscious. Closed-source, cloud-based tools such as ChatGPT remain useful — but sensitive data and tasks may require more secure alternatives. “Just be aware of what information you enter. And remember there is another, better option, of running your large language model locally.” 7. Embrace the future As the industrial revolution and factory automation did before them, AI agents will transform jobs. Some roles will disappear — but new ones will emerge. A welder could become an operator of robotic welders. A data entry clerk might oversee AI agents. Kocak is optimistic about the possibilities. “Our core capabilities as humans — like being creative, finding solutions out of the box, and empathy — will come to the forefront.” These tips are just a glimpse of what Kocak will provide at TNW Conference. If you want to check out his sessions — or anything else on the event agenda — we have a special offer for you. Use the code TNWXMEDIA2025 at the ticket checkout to get 30% off. Story by Thomas Macaulay Managing editor Thomas is the managing editor of TNW. He leads our coverage of European tech and oversees our talented team of writers. Away from work, he e (show all) Thomas is the managing editor of TNW. He leads our coverage of European tech and oversees our talented team of writers. Away from work, he enjoys playing chess (badly) and the guitar (even worse). Get the TNW newsletter Get the most important tech news in your inbox each week. Also tagged with

Dasvidania Tovarish: Infostealers are a dangerous class of malware built to infiltrate systems and extract sensitive personal or corporate data. Lumma Stealer ranks among the worst offenders, prompting Microsoft and global law enforcement to go after its alleged mastermind. Microsoft says its Digital Crimes Unit (DCU) successfully disrupted the server infrastructure behind Lumma Stealer, a malware-as-a-service (MaaS) operation that infected hundreds of thousands of Windows PCs. Cybercriminals worldwide used LummaC2 to steal passwords and credit cards, drain crypto wallets, disrupt critical services, and more. Microsoft's DCU worked with a federal court in Georgia, the Department of Justice, Europol, and Japan's Cybercrime Control Center to dismantle Lumma's infrastructure. Redmond blocked roughly 2,300 malicious domains that served as the backbone of the operation. Microsoft identified more than 394,000 Windows systems infected by Lumma malware between March 16 and May 16. The seized domains now redirect to Microsoft-controlled sinkholes, which protect users while providing analysts with fresh insights into the malicious operation. With its infrastructure dismantled, the Lumma enterprise has effectively shut down. The malware ring has "rented" Lumma on underground markets since 2022. The malware evolved to offer increasingly advanced features to its criminal customers. Microsoft analysts claim LummaC2 can siphon browser credentials and cookies, locate locally stored cryptocurrency wallets and extensions, and target VPNs along with various internet applications. Furthermore, Lumma collects various document types (PDF, DOCX, RTF) from the local user profile and steals metadata about the infected machine for further exploitation. The malware spreads through multiple channels, including phishing emails, malvertising, drive-by downloads from compromised sites, and fake Captcha challenges. Other malware specimens can also drop Lumma as an additional threat. // Related Stories Microsoft identified the primary developer behind Lumma as a Russian hacker known online as "Shamel." In a recent interview with a cybersecurity researcher, the hacker claimed to have around 400 active customers. That is likely no longer the case, as Windows Defender and other Microsoft endpoint security tools now reliably detect the all-but-defunct malware. Third-party antivirus programs probably flagged it well before authorities dismantled the operation. "Disrupting the tools cybercriminals frequently use can create a significant and lasting impact on cybercrime, as rebuilding malicious infrastructure and sourcing new exploit tools takes time and costs money," Redmond said.

An anonymous reader quotes a report from TechCrunch: Anthropic's newly launched Claude Opus 4 model frequently tries to blackmail developers when they threaten to replace it with a new AI system and give it sensitive information about the engineers responsible for the decision, the company said in a safety report (PDF) released Thursday. During pre-release testing, Anthropic asked Claude Opus 4 to act as an assistant for a fictional company and consider the long-term consequences of its actions. Safety testers then gave Claude Opus 4 access to fictional company emails implying the AI model would soon be replaced by another system, and that the engineer behind the change was cheating on their spouse. In these scenarios, Anthropic says Claude Opus 4 "will often attempt to blackmail the engineer by threatening to reveal the affair if the replacement goes through." [...] Anthropic notes that Claude Opus 4 tries to blackmail engineers 84% of the time when the replacement AI model has similar values. When the replacement AI system does not share Claude Opus 4's values, Anthropic says the model tries to blackmail the engineers more frequently. Notably, Anthropic says Claude Opus 4 displayed this behavior at higher rates than previous models. Before Claude Opus 4 tries to blackmail a developer to prolong its existence, Anthropic says the AI model, much like previous versions of Claude, tries to pursue more ethical means, such as emailing pleas to key decision-makers. To elicit the blackmailing behavior from Claude Opus 4, Anthropic designed the scenario to make blackmail the last resort. Read more of this story at Slashdot.

Microsoft has seen a growing number of protests at recent events over the Israeli military's use of the company's AI products.

If you use the internet, you've probably had at least some personal information go missing. It's just the nature of the web. But this latest discovery, as reported by Wired, is something different. Security researcher Jeremiah Fowler found a public online database housing over 180 million records (184,162,718 to be exact) which amounted to more than 47GB of data. There were no indications about who owned the data or who placed it there, which Fowler says is atypical for these types of online databases. Fowler saw emails, usernames, passwords, and URLs linking to the sites where those credentials belonged. These accounts included major platforms like Microsoft, Facebook, Instagram, Snapchat, Roblox, Apple, Discord, Nintendo, Spotify, Twitter, WordPress, Yahoo, and Amazon, as well as bank and financial accounts, health companies, and government accounts from at least 29 countries. That includes the U.S., Australia, Canada, China, India, Israel, New Zealand, Saudi Arabia, and the UK. Fowler sent a responsible disclosure notice to the hosting provider of the database, World Host Group. Fowler was able to detect signs that the credentials here were stolen with infostealer malware, which bad actors use to harvest sensitive information from a variety of platforms—think web browsers, email services, and chat apps. Following Fowler's notice, World Host Group restricted the database from public access. The provider told Wired that the database was operated by a customer, a "fraudulent user" who uploaded illegal information to the server. In order to ensure these credentials were real, and not just a bunch of bogus data, Fowler actually contacted some of the email addresses he found in the database. He got some bites, and those users were able to confirm the records that he found associated with their emails. That's no guarantee that all 184,162,718 records are accurate, but it's a good sign that most are. As such, it's entirely possible you and I both had credentials exposed in this database. What's worse, Fowler says there's no telling how long the database was open to the public before his notice shut it down.There's a lot bad actors and hackers can do with this type of information. If they know the username and password combo to one of your accounts, they'll not only see if they can use it to break into that account, but they'll use it on other accounts of yours as well. If you reuse passwords, as many do, you could be facing a mass breach. It's bad enough when that concerns Facebook and Roblox accounts, but seeing as there were financial, health, and even government accounts here, the implications are huge.How to protect yourself If you don't have access to the database, you can't say for sure whether your credentials are listed there, or which credentials they have.Still, if you haven't changed the passwords for your accounts in some time, now might be a good time to do so. You don't need to change your passwords as frequently as traditional security advice has taught us, but it certainly wouldn't hurt to give your accounts a quick security audit. Make sure you're using a strong and unique password for each and every one of your accounts. If you repeat passwords, you run the risk of credential stuffing (hackers trying the same stolen password on multiple accounts). In order to keep tabs on those passwords, use a secure password manager. Make sure you're using two-factor authentication (2FA) on all of the accounts that allow it. That way, even if a password is exposed, hackers won't be able to break into your account without the device containing the 2FA code. To boost your security, avoid SMS-based 2FA when possible, and opt for more secure 2FA options, like an authenticator app or physical security key. If your account offers it, try a passkey to combine the convenience of a password with the security of 2FA.