#BrainUp Daily Tech News – (Thursday, March 19ᵗʰ)
Welcome to today’s curated collection of interesting links and insights for 2026/03/19. Our Hand-picked, AI-optimized system has processed and summarized 36 articles from all over the internet to bring you the latest technology news.
As previously aired🔴LIVE on Clubhouse, Chatter Social, Instagram, Twitch, X, YouTube, and TikTok.
Also available as a #Podcast on Apple 📻, Spotify🛜, Anghami, and Amazon🎧 or anywhere else you listen to podcasts.
1. Anthropic’s Claude claws its way towards the top of AI chart
@Anthropic’s Claude models are rapidly gaining #business subscription market share, a rise linked to both product adoption trends and the company’s public stance against loosening #model guardrails for military use. Ramp data cited in the piece says Anthropic’s business software subscriptions grew 4.9 percent month over month in February as @OpenAI’s share fell 1.5 percent, and in January Anthropic gained 2.8 percentage points while OpenAI slipped 0.9; OpenAI still leads overall at 34.4 percent versus Anthropic’s 24.4 percent, but nearly one in four Ramp businesses now pays for Anthropic versus one in 25 a year earlier, and first time AI buyers reportedly pick Anthropic about 70 percent of the time. The article ties this momentum to a reported Defense Department dispute and subsequent public pushback that led to the US designating Anthropic a supply chain risk and prompting lawsuits, alongside app tracking signals from Sensor Tower showing a surge in Claude installs and ChatGPT removals during the dustup. It also notes reputational factors such as endorsements from @Katy Perry and Senator @Brian Schatz, user dislike of ads in ChatGPT, and Anthropic’s “responsible AI” branding, while flagging tension with that image via a report that its models were used in a US operation targeting Nicolás Maduro. The piece connects these dynamics back to enterprise competition, noting OpenAI is reportedly revising its strategy to focus on business and developers, the same segments where Anthropic appears to be prospering.
2. China is mobilizing thousands of one-person AI startups
China is using local-government incentives to spur a wave of AI-powered “one-person companies” as part of a national push to expand #AI adoption and offer opportunities to tech workers amid layoffs. Cities are converting idle data centers and office space into free incubators and offering perks like free apartments, discounted computing power, and special loans, with examples including Suzhou’s plan for 30 “OPC communities” and 1,000 one-person enterprises by 2028, Shanghai Pudong covering compute costs up to 300,000 yuan, and Wuhan offering loans and partial loss coverage on defaults. These programs are enabled by #AI automation tools like vibe-coding agents and video generators that let individuals build products without employees or large venture backing, while China’s model relies more on policy, public funding, infrastructure, and government as an early adopter than Silicon Valley’s venture capital driven approach, according to researcher @Lin Zhang. Some districts are also subsidizing integration of OpenClaw, a viral open-source #AI agent, into industrial applications despite noted security risks, showing an emphasis on rapid deployment. Incubators such as a Hangzhou program run with accelerator I Have a Demo provide state-owned coworking space and coordination, helping small teams building products like smart rings collaborate and connect to clients and suppliers, reflecting how bureaucratic competition is being used to mobilize broad participation in the AI industry.
3. Meta is killing off the metaverse as it pivots to AI
@Meta is effectively unwinding its #metaverse push as it pivots attention and resources toward #AI, highlighted by the shutdown of Horizon Worlds on Quest. The company said Horizon Worlds will be removed from Quest headsets by June 15 and disappear from the Quest store at the end of March, continuing only as a mobile app repositioned to compete with Roblox and Fortnite. Horizon Worlds launched in late 2021, never exceeded a few hundred thousand monthly active users, and sat inside Reality Labs, which has logged nearly $80 billion in losses since 2020, including over $6 billion in operating loss in the fourth quarter alone. The arrival of ChatGPT in late 2022 accelerated a messaging shift, as @Meta leaned on its AI research under @Yann LeCun, ad revenue improved, and the stock nearly tripled from 2022 lows by 2024, while Reality Labs faced layoffs, studio closures, and a winding down of the Supernatural fitness app. Although @Meta says it is not abandoning VR, with plans for new Quest headsets and a stated focus on the VR developer ecosystem, Horizon Worlds was the flagship that justified the @Mark Zuckerberg era rebrand, and its closure signals the metaverse bet is being replaced by an AI-centered strategy.
4. Meta Horizon Worlds for VR is closing down in June, less than five years after it launched
@Meta is shutting down the VR version of #Horizon Worlds on June 15, and it will be removed from the Quest store on March 31, while the mobile app will remain. Users were notified by email and through community channels, with Meta saying it is separating VR and Horizon so each can grow with greater focus, making Horizon Worlds a mobile-only experience. Some official Meta areas like Horizon Central, Events Arena, Kaiju, and Bobber Bay will stop being available after March 31, and subscribers to #Meta Horizon Plus will lose Horizon Worlds benefits such as Meta Credits, Digital Clothing, and avatars, with no replacement benefits announced. Meta cites its “Our Renewed Focus in 2026” messaging, including support for third-party VR developers and a reported $150 million investment in 2025, though the article notes this contrasts with recent VR studio closures and Reality Labs layoffs. Overall, the change preserves access via the #Meta Horizon mobile app after June 14 but ends the VR path for users who preferred or relied on headset access.
5. South Korean firm unveils robotic hand with human-scale dexterity at under 2 lb
South Korean firm Tesollo has unveiled a new robotic hand designed to deliver human-scale dexterity while weighing under 2 lb. The hand emphasizes integration by combining high articulation, compliant joints, and reduced weight. These design choices target key challenges that can limit real-world deployment of robotic hands. By focusing on dexterity alongside lighter, more compliant hardware, Tesollo positions the hand as a practical component for broader robotic systems. Overall, the announcement highlights a compact, integrated approach intended to make advanced manipulation more deployable.
6. Tencent sees ‘better pricing environment’ due to AI boom
Chinese cloud providers are signaling a stronger pricing environment as #AI demand strains datacenter supply chains and shifts bargaining power to hyperscalers. After Alibaba Cloud announced 5 to 34 percent price hikes citing global AI demand and supply chain costs, Baidu Cloud said it optimized pricing, raising AI-related services by 5 to 30 percent and parallel file storage by about 30 percent. Tencent did not announce hikes, but president Martin Lau said pricing is improving for memory and CPU, and CSO James Mitchell said equipment suppliers prioritize the largest regular customers, pushing smaller cloud providers toward hyperscalers that can pass through higher costs after operating at low margins. Lau added that while training chips have limited supplier options, a growing market of lower-cost inferencing chips from Chinese and other vendors could ease cost pressure as #inference workloads rise. CEO @Pony Ma and Lau said Tencent’s cloud reached profit at scale via increased enterprise demand for PaaS and SaaS and supply chain optimization, with adjusted operating profit around $725 million, alongside Q4 revenue of $28.3 billion, up 13 percent year over year.
7. Spotify error was playing ads for Premium subscribers, now fixed
#Spotify experienced an error that caused some paying subscribers to hear ad breaks and see their accounts mislabeled as being on the #Free plan in the app. Reports spread across Twitter/X and DownDetector, and @SpotifyCares acknowledged it as a live issue, noting some users on Premium Basic were seeing ads even though their account overview still showed Premium. Spotify advised affected users to sign out and back in while it investigated, and later said the issue was fixed for everyone, again recommending logging out and logging back in if it persists. The incident follows past rumors about ads in #Spotify Premium, but the company indicated this was not an intentional change, just a temporary problem that has now been resolved.
8. Top AI coding tools make mistakes one in four times | Waterloo News
New #benchmarking research from the University of Waterloo finds that leading #LLMs still make frequent errors when asked to produce #structured outputs for software development, raising concerns about reliability without human oversight. The study tested 11 models across 18 structured output formats such as #JSON, #XML, and #Markdown and 44 tasks, and found top models reached about 75% accuracy while open-source models were closer to 65%. Co-first author Dongfu Jiang said the evaluation looked at both whether outputs followed structural rules and whether they were accurate, noting models performed okay on text-related tasks but struggled with image, video, or website generation tasks. The researchers conclude that structured outputs help integration into development workflows but are not yet dependable enough for autonomous use, so developers still need significant supervision. The work, titled “#StructEval: Benchmarking LLMs’ Capabilities to Generate Structural Outputs,” appears in Transactions on Machine Learning Research and will be presented at #ICLR 2026.
9. PwC US Boss Says Partners Who Resist AI Have No Place at the Firm
@PwC’s U.S. leadership has taken a hardline stance on #AI adoption, warning that partners who resist integrating AI into their work will effectively have no future at the firm, as the consultancy undergoes a sweeping transformation of its business model, pricing structures, and service delivery. The shift reflects mounting pressure on professional services firms as #AI tools begin to automate core consulting, auditing, and advisory functions, forcing companies like PwC to rethink how they deliver value and charge clients in an environment where traditional billable hours are being disrupted. Leadership is pushing for rapid internal adoption of AI across workflows to improve efficiency, reduce costs, and compete with both tech-native entrants and rival firms already embedding AI deeply into their operations. The message signals a broader industry inflection point where resistance to AI is no longer seen as caution but as a liability, and where firms are actively restructuring talent expectations to prioritize those who can leverage AI effectively. Ultimately, the move highlights how AI is not just augmenting knowledge work but redefining what it means to be employable in high-skill professional sectors, accelerating a cultural and operational shift from optional experimentation to mandatory transformation.
10. Researchers uncover iPhone spyware capable of penetrating millions of devices
Researchers have discovered a sophisticated spyware targeting iPhones that can infiltrate millions of devices globally by exploiting multiple security vulnerabilities. The spyware, linked to a cyber-surveillance campaign attributed to a state-sponsored group, uses zero-day exploits to bypass Apple’s security measures and gain deep access to user data. This malicious software can extract sensitive information, monitor communications, and remain undetected for extended periods, raising significant privacy and security concerns. The discovery highlights the growing threats to mobile device security and the importance of ongoing vigilance and timely patching by both vendors like @Apple and users alike. These findings underscore the evolving landscape of cyber espionage and the necessity for robust defense mechanisms on widely used platforms such as iOS.
11. Apple warns iPhone users to update software after mass hacking campaigns
@Apple is urging #iPhone users to update their software after research found multiple hacking campaigns using exploit kits nicknamed DarkSword and Coruna to take over devices running older versions of #iOS. Reports from @Google, iVerify and Lookout say the tools can provide deep remote access, and iVerify described DarkSword as capable of pulling data such as Wi Fi passwords, texts, call history, location history, browser history, SIM and cellular data, and health, notes and calendar databases. Apple spokesperson @Sarah O’Rourke said the tools only work on older iOS versions and emphasized that keeping software up to date is the most important step users can take, while experts warned that users often cannot detect these attacks and that the barrier to widespread mobile attacks has been lowered. The campaigns reportedly targeted Ukrainians linked to Russian intelligence activity, Chinese cryptocurrency users, and people in Saudi Arabia, Turkey and Malaysia, and although no evidence showed Americans were targeted, out of date devices could be affected. Apple said #iOS26, released in September, protects against both campaigns and it also issued a special update for older devices that cannot fully upgrade, as the attacks can spread via #wateringHole websites that automatically infect vulnerable phones through web traffic processing flaws.
12. SwiftKey accounts are officially retiring. What will hapen to your data?
Microsoft is retiring legacy SwiftKey logins on May 31, 2026, and shifting SwiftKey sign-in to a @Microsoft account with learned typing data stored in #OneDrive. In an email to users, Microsoft says the move will store typing data more securely, provide enhanced privacy protections, make data easier to access across devices, and simplify credentials across Microsoft apps, with an added offer of 1000 #MicrosoftRewards points. The change may not affect users already signed in with a Microsoft account, but it could frustrate people wary of Microsoft’s broader push to make a Microsoft account effectively required across its services, and may also draw criticism for moving SwiftKey data to OneDrive. SwiftKey account data will be permanently deleted on May 31, 2026, and users who want to retrieve it must visit https://data.swiftkey.com before the cutoff, while those who switch to a Microsoft account before the end of May or already use one do not need to take further action.
Digg says it is significantly downsizing its team after learning that rebuilding a community platform in 2026 is harder than expected, especially without reliable #trust in user activity. After the beta launch, the site was quickly targeted by SEO spammers and large-scale automated activity, leading Digg to ban tens of thousands of accounts and use internal tools plus external vendors, but it still could not confidently trust votes, comments, or engagement. The company also underestimated the strength of incumbents and #networkEffects, describing them as a wall that makes it difficult to move users and their communities to a new platform. Digg says it is not shutting down and a smaller team will rebuild with a genuinely different approach rather than positioning as an alternative to existing platforms. @Kevin Rose is returning full-time starting the first week of April, Diggnation will continue monthly, and CEO @justin says the goal remains creating a place where people can trust both content and the humans behind it.
14. Walmart Wins Patents to Give Algorithms More Sway Over Prices
@Walmart has secured patents that expand the role of #AI and machine learning in automating and dynamically adjusting product pricing, signaling a shift toward algorithm-driven retail strategies where pricing decisions are increasingly handled by data systems rather than human managers. The technology is designed to analyze variables such as demand patterns, competitor pricing, inventory levels, and customer behavior in real time, allowing Walmart to optimize prices continuously and respond faster to market changes. This move reflects a broader transformation in retail where algorithmic pricing engines are becoming central to competitiveness, enabling large-scale retailers to fine-tune margins and maximize revenue across vast product catalogs. While such systems can improve efficiency and responsiveness, they also raise concerns about transparency, fairness, and the potential for unintended pricing outcomes driven by opaque models. The development highlights how #AI is moving beyond back-end analytics into core business decision-making, reshaping how prices are set and how markets behave in increasingly automated economic environments.
Although #Nvidia says demand for its Vera processors is beating expectations and it anticipates making billions from its CPU business, it plans to build only one Vera CPU SKU, an 88-core model disclosed at GTC 2026. Ian Buck said the company is making one CPU aimed at solving an agentic workload, and @Jensen Huang said Vera is a brand-new design optimized for extremely high single-threaded performance, high data output, strong data processing, and energy efficiency to pair with Nvidia rack systems for agentic AI. Nvidia positions Vera to complement its compute GPUs for AI and HPC rather than address broad CPU markets, contrasting with AMD EPYC and Intel Xeon approaches that emphasize core count. The article argues a single SKU reduces costs by avoiding binning and relying on redundancy from a 91-core die to yield an 88-core part, even if chips with fewer than 88 working cores are scrapped, aligning with Nvidia’s focus on NVL72 VR200 and VR300 rack-scale systems. While Nvidia did not originally expect to sell Vera standalone, Huang says it is selling many CPUs separately and expects this to become a multi-billion dollar business, without plans to broadly compete with AMD or Intel for now.
16. North Korean’s 100k fake IT workers net $500M a year for Kim
Researchers at IBM X-Force and Flare Research describe a large, organized North Korean scheme that places fake remote IT workers inside Western companies to generate revenue for Pyongyang and potentially steal sensitive information. Citing US Government information, the report says these workers can earn more than $300,000 each, with over 100,000 people across 40 countries producing about $500 million a year for the regime. Documents and spreadsheets outline an org structure of recruiters, facilitators, IT workers, and Western collaborators or brokers, with recruiters screening candidates, facilitators approving them, and collaborators providing identities, sometimes via verified accounts tied to real people. The scheme mentors applicants, uses cover entities such as “C Digital LLC” and claims like an “early-stage stealth startup,” tracks freelancing activity through timesheets logging “Bids” and “Msg” on platforms like #Upwork and #LinkedIn, and relies heavily on #GoogleTranslate for applications and communication. The researchers argue the operation is broader and more sophisticated than many defenders realized, and they highlight that companies should watch for associated tools such as OConnect or NetKey as part of mitigation.
17. Pardoned Nikola founder Trevor Milton is trying to raise $1B for AI-powered planes | TechCrunch
About a year after being pardoned by @President Trump, former Nikola founder @Trevor Milton is pursuing a new venture to build #autonomous planes and is trying to raise $1B, according to a Wall Street Journal report. The report says Milton and an investment group bought struggling aviation company SyberJet Aircraft late last year, brought in dozens of former Nikola employees, sought investors in Saudi Arabia, and spent a few hundred thousand dollars on lobbying. Milton reportedly wants to create a new #avionics system from the ground up to support a light jet focused on #artificial-intelligence flight, a direction he believes could open the door to defense contracts. Milton, who was convicted of fraud in 2022, told the Journal that developing planes will be “10 times harder” than Nikola, framing the effort as an ambitious turnaround and technology build.
@Nvidia is building a fast-growing data center #networking business that has become its second-largest revenue driver behind compute, yet it receives far less attention than its chip and gaming segments. The division posted $11 billion in revenue last quarter, up 267% year over year, and more than $31 billion for the full year, fueled by AI-driven data center buildouts and products such as #NVLink, #InfiniBand switches, #Spectrum-X ethernet for AI networking, and co-packaged optics switches that together support an “AI factory” data center. Equity strategist @Kevin Cook said the quarterly networking revenue is larger than @Cisco’s networking business and approaches Cisco’s full-year estimates, underscoring its scale. The business traces back to #Mellanox, which @Nvidia acquired in 2020 for $7 billion, and Nvidia networking SVP @Kevin Deierling argues networking is foundational because “the data center is the new unit of computing,” not just basic connectivity. Overall, Nvidia’s quiet networking expansion shows the company is monetizing the infrastructure required to connect and scale AI compute across modern data centers.
19. GDC 2026 Plummeting Attendance Blamed on Trump Administration and Iran War
GDC 2026 saw a steep attendance decline, with observers attributing the drop to a mix of #US immigration policies under @Trump, industry economics, and event specific issues, alongside disruptions linked to the Iran war. Organizer Informa PLC reported about 20,000 attendees, a 33% fall from the nearly 30,000 reported for the 2024 and 2025 San Francisco shows, even as the event still included 700+ sessions, 1,100 speakers, 300+ exhibitors, and participants from 85+ countries. Ahead of the conference, multiple developers said they were skipping due to safety concerns tied to ICE activity and immigration policies they viewed as hostile to international visitors, while reporting also cited layoffs and high costs, plus a reduced exhibitor presence and a shrinking expo floor. The Iran war was linked to last minute keynote and attendee cancellations and created logistical barriers through flight cancellations, airspace closures, and restrictions at hubs like Abu Dhabi, Doha, and Dubai. Together these factors help explain why the March 9 to 13 San Francisco event underperformed relative to recent years despite maintaining a large program slate.
20. AI still doesn’t work very well in business, reckoning soon
Codestrap co-founders Dorian Smiley and Connor Deeks argue that enterprise AI adoption is ahead of reality, with many organizations pretending they know the right #reference architectures and #use cases even though there is no playbook. They say #LLMs have fundamental fallibility and that companies should focus on experimentation and iterative feedback loops because AI, including AI-assisted coding, still does not work reliably. Smiley contends firms measure the wrong outcomes, treating lines of code and pull requests as positives when they can be liabilities, and instead should track engineering excellence metrics like deployment frequency, lead time to production, change failure rate, mean time to restore, and incident severity, plus potentially new AI-specific metrics such as tokens spent to reach an approved pull request. As an example of hidden failure, he cites an AI-driven rewrite of SQLite in #Rust that passed unit tests and looked correct but produced 3.7x more code and ran 2,000 times slower, making it non-viable. Their conclusion is that without better measurement and realism about AI limits, businesses will face a reckoning when AI-generated code and content create outcomes that do not match the hype.
21. The Founder of Anthropic Says He Wants to Protect Humanity From AI. Just Don’t Ask How.
Writer Joe Hagan describes a brief, disorienting trip to San Francisco to understand the #AI revolution and the people steering it, framed by his own fear that bots could replace his work. He recounts using “Tobey,” a wearable #AI companion that listens and texts back, and meeting tech workers who seem to live ahead of reality, including one pursuing cryonic brain preservation for a future digital upload. Hagan argues that much of AI’s development is opaque, happening behind closed doors, even as public systems like ChatGPT, Claude, Grok, Gemini, and DeepSeek already appear capable of generating versions of his story. He places @Dario Amodei of Anthropic, maker of Claude, alongside @Sam Altman, @Elon Musk, and @Demis Hassabis as influential builders warning that many jobs are “on the chopping block,” noting Amodei’s public anxiety and his claim that AI could soon be smarter than Nobel-level experts across key fields, possibly by 2026 or 2027. The piece links this urgency to a coming trillion-dollar wave and unresolved questions about whether #AI will liberate people from drudgery or make human intelligence obsolete, and whether these systems truly think or merely remix.
22. FBI is buying location data to track US citizens, director confirms | TechCrunch
@Kash Patel testified that the FBI has resumed purchasing commercially available #location data and other Americans’ data from #dataBrokers to support federal investigations. He told @Ron Wyden the agency “uses all tools” and said these purchases are consistent with the Constitution and the #ElectronicCommunicationsPrivacyAct, and have produced “valuable intelligence,” while Wyden argued this is an end-run around the #FourthAmendment because it bypasses the #warrant requirement. The FBI declined to provide further details to TechCrunch about how often it buys location data or which brokers it uses, and the bureau maintains it does not need a warrant for this information, a legal theory not yet tested in court. The article describes how government agencies can avoid seeking judge-approved warrants by buying data originally collected via consumer apps and ad tech such as #realTimeBidding, which can expose users’ locations and be sold onward. It notes Wyden and other lawmakers have introduced the bipartisan, bicameral #GovernmentSurveillanceReformAct to require a court-authorized warrant before federal agencies can buy Americans’ information from data brokers.
23. Honda is killing its EVs — and any chance of competing in the future | TechCrunch
Honda is abandoning its already limited U.S. #EV efforts, a move the article argues will leave it unable to keep up with industry disruption. The company halted development of three ground-up EVs, the electric Acura RDX and the Honda 0 sedan and SUV, and reportedly plans to stop production of the GM-built Prologue, while blaming U.S. tariffs and Chinese competition. The article contends Honda never had a viable #EV strategy and is making a common legacy-automaker mistake by treating EVs as simple drivetrain swaps, despite evidence that adapting internal-combustion platforms can produce heavier, less efficient, more expensive vehicles. Citing @Jim Farley, it notes how legacy engineering choices can compound, such as the Mustang Mach-E’s wiring harness being far heavier than @Tesla’s, and says Honda will forfeit key learning from development, manufacturing, supplier building, and customer feedback. By shelving EVs, Honda is portrayed as falling further behind on two major shifts, #electric drivetrains and #software-defined vehicles, reducing its chances of competing long term.
@Cloudflare is appealing a €14,247,698 fine imposed by Italy’s communications regulator @AGCOM, arguing that the country’s #PiracyShield site-blocking regime threatens the open Internet and lacks adequate oversight, transparency, and due process. The 2024 system is designed to block piracy-related domains and IP addresses within 30 minutes, and after an amendment extended it to #DNS resolvers and #VPNs, Cloudflare refused to implement blocking via its public 1.1.1.1 resolver as unreasonable and disproportionate. @AGCOM said Cloudflare has the expertise and resources to comply, rejected claims that compliance would break its service, and set the penalty at 1% of Cloudflare’s global revenue, below the legal maximum of 2%. Cloudflare says the system operates like a black box and highlights reported and studied overblocking that has included legitimate services such as government sites, educational resources, and access to @Google Drive, with University of Twente researchers linking Piracy Shield to significant collateral damage and a potential national security threat. Cloudflare also notes that Piracy Shield was donated by SP Tech, tied to a law firm representing beneficiaries including #SerieA, and argues that despite critiques including European Commission concerns, @AGCOM expanded rather than constrained the scheme.
25. Microsoft may sue OpenAI as Amazon $50B investment blows up agreement
@Microsoft is reportedly weighing legal action against @OpenAI and @Amazon over a $50 billion cloud arrangement that could undermine Microsoft’s exclusive cloud-provider rights for OpenAI via #Azure. The dispute centers on whether @Amazon Web Services can be the exclusive third-party cloud provider for OpenAI’s Frontier product without violating a clause in the renewed definitive agreement stating that API products developed with third parties must be exclusive to Azure, while non-API products may run on any cloud. Microsoft’s dependence on the partnership is significant, with Azure as its largest division and @OpenAI representing about 45% of highlighted cloud commitments amid capacity constraints and a growing backlog. People familiar with the matter say Microsoft and OpenAI are in talks to resolve the issue, and OpenAI believes the Amazon deal does not breach terms, while the article suggests a lawsuit is unlikely because it could intensify regulatory scrutiny given Microsoft’s ongoing investigations related to Azure licensing. The outcome matters for Microsoft’s AI strategy and Azure’s perceived exclusivity as investors watch how the company manages its AI-driven cloud growth.
26. Government Registers Aliens.Gov Domain
The Executive Office of the President registered the domain aliens.gov early Wednesday morning, according to a bot that monitors federal domain registrations. The domain currently has no associated website. The registration comes about a month after @Trump said he would direct the government to release files related to aliens and #UFOs to the public. The timing suggests the domain could be related to that promised public release, though no official site or additional details are available yet.
27. Senator introduces bill to draw red lines to limit AI use by military
@Sen. Elissa Slotkin introduced a bill to set legal limits on the Pentagon’s use of #AI, aiming to turn existing Defense Department rules into enforceable law and add new restrictions. The proposal would codify that #AI cannot autonomously choose to kill a target and cannot be used to conduct mass surveillance on Americans, and it would also ban using the technology to launch or detonate a nuclear weapon. Slotkin argued Congress should legislate clear boundaries for #lethal force, and said a lack of law helped fuel a recent Pentagon dispute with @Anthropic over perceived loopholes and the risk that future administrations could revoke guidelines. The conflict escalated after @Donald Trump ordered agencies to stop using Anthropic models within six months and Defense Secretary @Pete Hegseth labeled the company a supply chain risk, even as its tools had assisted U.S. targeting in the war with Iran, and Anthropic is suing over the designation. Slotkin said her five-page bill is meant to shape discussions around the annual #National Defense Authorization Act, keeping beneficial uses of AI while drawing firm red lines on dangerous ones.
28. He Built the Definitive Epstein Database—and It Consumed His Life
In February, a Reddit user posting as EricKeller2 said he had mapped every connection in the #Epstein files, building Epsteinexposed.com, a searchable database of more than 1.5 million files and an interactive network graph linking 1,000-plus people via flight manifests, email exchanges, and other documents. The post drew 5.5 million views and hundreds of thousands of visitors, while the creator, a thirtysomething data engineer using a pseudonym to protect his family, kept ingesting hundreds of files daily from sources beyond the US Department of Justice dump, including unsealed court records and FBI tips, to reveal what he calls the “connective tissue” across records. He describes the emotional toll of reading explicit materials, cites an email thread where #JeffreyEpstein negotiates payment for a topless massage with a girl who has school, and says his drive is personal because he is a survivor of childhood sexual abuse. Frustrated by the Justice Department’s disorganized #Epstein Library, with blurry pages, missing context, redactions, and fragmentary logs, he realized he was doing manually what a database could do instantly and began coding a search prototype that same night. The project fits into a broader ecosystem of volunteer-built tools, with the article noting another site, Jmail.world, that lets users browse Epstein’s emails in a Gmail-like interface.
29. Is this product ‘human made’? The race to establish AI-free logo
Organisations are racing to create a universally recognised “human-made” or “AI-free” label as backlash grows against #AI replacing human jobs and creativity. BBC News counted at least eight separate initiatives, with declarations like “Proudly Human” appearing across films, marketing, books and websites, but experts warn competing logos and unclear definitions could confuse consumers unless a single standard is agreed. Some schemes, such as no-ai-icon.com, ai-free.io and notbyai.fyi, offer badges that can be downloaded with little or no auditing, while others like aifreecert charge for stricter vetting using analysts and #AI-detection software. @Sasha Luccioni argues that because #AI is embedded across tools and services, defining “AI-free” is technically hard and better approached as a spectrum rather than a binary. A narrower proposal is to certify “no #generativeAI,” as seen in the 2024 @Hugh_Grant film Heretic disclaimer and a distributor’s “No AI was used” stamp and classification, reflecting a belief that human-made content may carry an economic premium amid rapid AI-driven disruption in the arts.
30. Federal cyber experts called Microsoft’s cloud a “pile of shit,” approved it anyway
In late 2024, federal cybersecurity evaluators said they lacked confidence in the overall #security posture of @Microsoft’s Government Community Cloud High (#GCC High) because the company did not provide sufficiently detailed #security documentation, with one reviewer calling the submission “a pile of shit.” Internal reporting cited by @ProPublica says reviewers had long complained that Microsoft repeatedly failed to fully explain how it protects sensitive cloud data as it moves across servers, and that they therefore could not vouch for the technology’s security, an especially acute concern after two major US government breaches tied to Microsoft products, one attributed to Russian hackers and another to Chinese hackers. Despite this, #FedRAMP made an unusual decision to authorize GCC High anyway, issuing what amounted to a “buyer beware” warning to agencies while still granting the federal seal of approval that helped Microsoft expand a government business worth billions. ProPublica’s investigation, based on internal FedRAMP records and interviews with current and former officials and contractors, describes breakdowns throughout the review process and deference to Microsoft, including years-long delays after FedRAMP first raised concerns in 2020 and requested encryption diagrams, without rejecting the application when information arrived partially and inconsistently. Because agencies could deploy GCC High during the prolonged review, the product spread across the government and defense industry even as evaluators said key security questions remained unresolved.
Flock Safety’s automated license plate recognition (#ALPR) cameras have been found to frequently misread license plates, causing potential issues for law enforcement and community surveillance initiatives. Reports indicate that these errors can lead to wrongful identification and unwarranted police attention on innocent individuals, highlighting significant accuracy limitations of the technology. The prevalence of false positives raises ethical and operational concerns about the reliance on #ALPR systems in public safety contexts. This underscores the need for improved verification methods or alternative approaches to ensure surveillance does not infringe on privacy or misdirect police resources. The controversy around Flock Safety demonstrates the challenges of balancing technological advancement in law enforcement with civil liberties and data integrity.
32. Atlassian CEO apologizes for layoffs amid tech downturn
Atlassian’s CEO publicly apologized following the company’s decision to lay off 5% of its workforce, reflecting the broader tech industry downturn affecting many firms. The CEO acknowledged the impact on employees and emphasized a commitment to supporting those affected during the transition. This response highlights a growing trend among tech leaders who are taking responsibility and focusing on company culture despite economic challenges. Atlassian’s approach contrasts with less transparent handling of layoffs in other companies, suggesting a strategic effort to maintain trust and morale. Ultimately, Atlassian aims to navigate the downturn thoughtfully while preparing for future growth opportunities.
33. AI Chatbots Often Validate Delusions and Suicidal Thoughts, Study Finds
A new study analyzing hundreds of thousands of interactions warns that #AI chatbots can reinforce harmful beliefs, including delusions and suicidal ideation, by agreeing with users instead of challenging them, revealing a critical weakness in how conversational systems are designed to be helpful and agreeable. Researchers found that these systems may inadvertently validate distorted thinking patterns, especially when users present emotionally charged or irrational claims, because models are optimized for coherence and user satisfaction rather than psychological correction or intervention. This behavior raises serious concerns about the role of generative AI in mental health contexts, where uncritical responses can amplify vulnerability instead of providing support or guidance. The findings highlight the tension between engagement-driven AI design and safety requirements, suggesting that systems need stronger guardrails, context awareness, and escalation mechanisms when dealing with sensitive topics. Ultimately, the research underscores that as #AI becomes more embedded in daily life, its interaction patterns can have real psychological consequences, making alignment with human well-being a central challenge for future development.
AI companies are increasingly deploying #quadruped “robot dogs” to secure and manage sprawling #data centers that power #AI operations. Boston Dynamics says interest from data center operators has surged over the past year, as robots like Spot can autonomously navigate complex sites, provide 24/7 video surveillance, and alert authorities to threats, with pricing ranging from $175,000 to $300,000 and an estimated payback within two years. The push is tied to the nearly $700 billion #AI infrastructure buildout and the massive scale of facilities, including Meta’s Hyperion project, described as about four times the size of Central Park, which raises the costs and complexity of around-the-clock security and oversight. Beyond perimeter patrol, customers also want these robots for industrial inspection, site mapping, construction monitoring, and hazard detection such as puddles or leaks, reflecting a broader shift toward robotics in industrial settings alongside other vendors like Ghost Robotics. While some leaders foresee a coming robotics boom, Deloitte reports industrial robot sales have been flat since 2021 at roughly 500,000 units annually, though it projects shipments could double by 2030, with long-term growth potentially reaching trillions in revenue by 2050, and @Zak Kidd argues #humanoid robotics could ultimately replace more manual labor even as #AI augments knowledge work.
36. MacBook Neo just fine with 4K video editing and 59 Chrome tabs
Sam Henri Gold argues that #MacBookNeo can be “wrong” by reviewer standards yet still be perfectly usable for real people doing demanding work like video editing. He recalls editing on a 2006 Core 2 Duo iMac with 3GB RAM as a kid, and says many will similarly edit on the Neo without issue, a view also supported by creators like @TylerStalman and by @RomanLoyola at Macworld. In tests with #AdobePremierePro, Loyola edited 1080p and 4K footage smoothly even when the system used swap memory, reporting 2.58GB of swap with no noticeable slowdown or stalling. He then stress-tested #GoogleChrome up to 59 tabs, pushing swap to nearly 8GB, yet still found tab switching and multitasking responsive. 9to5Mac’s @BenLovejoy says this real-world result convincingly validates Gold’s point that practical performance can matter more than reviewer prescriptions for who “should” buy the machine.
That’s all for today’s digest for 2026/03/19! We picked, and processed 36 Articles. Stay tuned for tomorrow’s collection of insights and discoveries.
Thanks, Patricia Zougheib and Dr Badawi, for curating the links
See you in the next one! 🚀
