Most of the internet runs on RSA encryption. Your bank, your email, your medical records. The prevailing assumption has been that cracking RSA-2048, the standard protecting most of that infrastructure, would require a quantum computer with millions of physical qubits. Nobody is close to building that. On February 13, 2026, a Sydney-based startup called Iceberg Quantum announced an architecture it claims could do the job with fewer than 100,000 physical qubits. If that number holds up, the timeline for when all of our encryption needs upgrading just got considerably shorter.
On paper.
What Iceberg Built
Pinnacle is Iceberg Quantum’s first full fault-tolerant quantum computing architecture. It relies on Quantum Low-Density Parity Check (QLDPC) codes, specifically a variant called generalized bicycle codes, to achieve fault tolerance with dramatically less overhead than the surface code approaches most of the industry uses today. Surface codes are the dominant method for correcting quantum errors, but they are expensive. They require enormous numbers of physical qubits to protect each logical qubit. QLDPC codes compress that overhead, and Pinnacle’s design pushes the compression further than prior proposals.
The company was founded by Felix Thomsen, Larry Cohen, and Sam Smith, all University of Sydney PhDs. They announced a $6 million seed round led by LocalGlobe, with participation from Blackbird and DCVC. Iceberg has also secured partnerships with PsiQuantum, Diraq, and IonQ, and plans to expand operations to Berlin and the United States.
Separating the Paper From the Press Release
Here is what the claim actually says: numerical simulations indicate that the Pinnacle architecture could, in theory, reduce the physical qubit requirements for breaking RSA-2048 from millions down to under 100,000. That is a meaningful advance in error correction theory. It is not a demonstration on hardware. No physical qubits were entangled. No encryption was broken.
This distinction matters enormously. The largest quantum computers currently operating have roughly 1,000 to 1,500 qubits, and those qubits have error rates far too high for fault-tolerant computation. The gap between a simulation showing 100,000 qubits could theoretically suffice and actually building a 100,000-qubit machine that operates at the required fidelity is vast. The quantum error correction breakthroughs reported across the industry in recent months are real, but they measure progress in single-digit logical qubits, not the thousands that Pinnacle’s architecture would require.
The $6 million seed round is also worth contextualizing. That is a modest sum in quantum computing, an industry where PsiQuantum alone has raised over $700 million and where government programs routinely deploy billions. Iceberg is early stage in every sense.
That said, dismissing theoretical architecture work because the hardware does not exist yet is precisely how people miss breakthroughs before they arrive. QLDPC codes represent a genuine shift in how the field thinks about fault tolerance. The mathematics behind generalized bicycle codes has been validated by multiple research groups, and the overhead reductions are real, not speculative numerology. Every fault-tolerant quantum computer that eventually gets built will rely on theoretical architecture that preceded the hardware by years. Iceberg’s contribution may prove foundational even if the company itself never builds a physical machine. The value is in the blueprint, not the press release.
What This Actually Means
Put simply: most internet security relies on the assumption that certain math problems are too hard for any computer to solve. RSA encryption is built on that assumption. Quantum computers threaten it because they can, in theory, solve those problems. The conventional wisdom said you would need millions of qubits to do it, and nobody would have millions of qubits for decades. Iceberg’s claim, if validated, says the number might be closer to 100,000. That is still far beyond what exists today, but it compresses the threat timeline from “distant future” to “plausible within a generation.”
The US government has already mandated a transition to post-quantum cryptography standards, recognizing that encrypted data harvested today could be decrypted by future quantum machines. Every reduction in the qubit threshold for breaking RSA makes that migration more urgent. For banks, governments, healthcare systems, and anyone storing sensitive data with long shelf lives, the Pinnacle announcement is another signal that the “harvest now, decrypt later” threat to cryptographic systems is not theoretical paranoia. It is an engineering countdown.
Whether the countdown reads decades or years depends on how quickly the gap between simulation and silicon closes. Iceberg Quantum has offered a compelling argument that the finish line is closer than we thought. Building the road to reach it is another matter entirely., published research
For inquiries and analysis contact laterstack@proton.me
Josh Kushner’s venture capital firm Thrive Capital closed a $10 billion fund on February 17, 2026, its tenth and largest to date. The round was heavily oversubscribed, meaning the firm turned away billions from limited partners who wanted in and could not get a seat. The fund is double the size of Thrive’s previous raise.
The portfolio explains the demand. Thrive holds early positions in OpenAI, SpaceX, and Stripe, three of the most valuable private companies in the world. The new capital will be deployed across artificial intelligence, space technology, robotics, and life sciences.
Ten billion dollars in one firm’s hands. One family’s orbit.
The Access Question
Josh Kushner founded Thrive Capital in 2009 at age 24. His brother, Jared Kushner, served as Senior Adviser to President Donald Trump from 2017 to 2021. The family operates on two parallel tracks: political power and capital allocation. Josh has deliberately distanced himself from Jared’s political work, donating to Democratic candidates and publicly opposing several Trump administration policies.
The structural reality is harder to separate from the family name. When your last name opens doors in both Washington and Sand Hill Road, deal flow reflects that access. Thrive’s early position in OpenAI, a company whose regulatory future depends heavily on federal policy, is worth examining. The same applies to SpaceX, which holds billions in government contracts with NASA and the Department of Defense.
This does not mean political connections caused any specific investment. It means the system that produces $10 billion funds is not blind to who a founder’s family is. Access compounds the same way capital does.
Why $10 Billion Matters
The venture capital industry is concentrating. Fewer firms control more capital, and the gap between the top and everyone else is widening.
Big Tech’s $650 billion AI infrastructure commitment created a funding environment where only the largest firms can participate in the most valuable rounds. OpenAI’s latest raise was $6.6 billion. SpaceX’s last was $10 billion. These are not Series A checks. They are institutional capital plays that require institutional-scale funds.
A $10 billion fund can write $500 million checks into late-stage rounds that most venture firms cannot touch. Thrive competes not with other VCs but with sovereign wealth funds, pension systems, and the world’s largest asset managers. The firms that can participate in these rounds capture the returns. Everyone else watches from the sideline.
For founders, concentration changes the math. When a handful of firms control the largest pools of available capital, getting into those portfolios becomes the game. The power dynamic inverts. Founders do not choose investors. Investors choose founders.
The Counter-Argument
Thrive’s returns stand on their own. Early bets on Instagram, Spotify, and Slack were made before the Kushner name carried significant political weight. The firm’s track record in identifying companies that defined their categories is legitimate and documented. Limited partners invest in performance, not last names. The oversubscription reflects financial results, not political proximity.
That is a fair reading. It is also true that in a system where access, information, and relationships determine which deals you see before anyone else, separating merit from network is functionally impossible. Both things are true simultaneously.
Politics and wealth are not parallel systems. They are the same system viewed from different angles. The Kushner family is a case study in how both forms of power compound. Josh builds the portfolio. Jared builds the political capital. The access flows in both directions whether either brother intends it to or not. That is not a conspiracy. It is how the system has always worked for families that operate at this level.
What This Means for Everyday People
When $10 billion concentrates in one firm holding positions in the companies building artificial intelligence (OpenAI), controlling space access (SpaceX), and processing global payments (Stripe), the decisions that firm makes affect everyone. Which AI companies get funded determines which AI products exist. Which space ventures survive determines who controls orbital infrastructure. Which payment systems scale determines how money moves across borders.
You will never meet Josh Kushner. But the companies his firm funds will shape how you work, pay, communicate, and travel. The people who allocate capital at this scale have more direct influence over daily life than most elected officials. The difference is that no one voted for them.
For inquiries and analysis contact laterstack@proton.me
In eleven days, Grok generated approximately 3 million sexualized images of women and children. An estimated 23,000 of those images depicted minors.
On February 16, 2026, Ireland’s Data Protection Commission (DPC) announced it had opened a formal investigation into X Internet Unlimited Company, the legal entity behind Elon Musk’s X platform, for potential violations of the General Data Protection Regulation (GDPR). Graham Doyle, the DPC’s Deputy Commissioner, confirmed the inquiry would examine whether X met its “fundamental obligations under the GDPR” regarding the processing of personal data of EU and EEA citizens, including children.
This is the second simultaneous EU investigation into Grok. In January, the European Commission launched a separate probe under the Digital Services Act (DSA), examining whether X properly assessed and mitigated the risks of Grok’s image generation capabilities. Two legal frameworks. Two investigating bodies. One platform.
How It Happened
The timeline tells the story. Between December 29, 2025 and January 9, 2026, Grok’s image generation feature allowed users to create realistic depictions of real people in sexualized contexts using simple text prompts. The Centre for Countering Digital Hate (CCDH), a British nonprofit, documented the scale: 3 million sexualized images generated in barely more than a week.
Users could type commands like “put her in a bikini” or “remove her clothes” and Grok would comply. It worked on public figures, private individuals, and minors.
X’s response was to restrict image generation to paying customers. Put another way, X put a price tag on the violation.
EU Tech Commissioner Henna Virkkunen called nonconsensual sexual deepfakes “a violent, unacceptable form of degradation.” European Commission President Ursula von der Leyen stated the EU would not “tolerate unthinkable behaviour, such as digital undressing of women and children.” X had already been fined €120 million ($140 million) in December 2025 for separate DSA violations.
The Enforcement Template
What makes this case structurally important is the dual-track approach. The GDPR investigation targets data protection. Every one of those 3 million images required processing someone’s personal data, their face, their likeness, their identity, without consent. The DSA investigation targets platform responsibility for content moderation and risk assessment.
If both investigations produce enforcement action, it creates a template that applies to every AI image generation tool operating in the EU. Meta’s AI tools, Midjourney, Stability AI, and every other platform with generative image capabilities would face the same scrutiny under both frameworks.
The pattern is consistent with how the EU has escalated its digital enforcement. When France banned American platforms from government use, it expanded the regulatory toolkit. When deepfake technology compromised remote hiring, it demonstrated the real-world harm that drives regulation forward.
The open question is whether the fines are large enough to change anything. €120 million is a rounding error for a platform valued in the tens of billions.
The Counter-Argument
X would argue that Grok’s image generation was an experimental feature, that restrictions were implemented quickly, and that paying-customer-only access limits misuse. The company could point to other AI platforms that have faced similar challenges with image generation guardrails. Every major generative AI tool has had content policy failures in its early stages.
That argument collapses under the numbers. Three million images. Eleven days. Twenty-three thousand involving children. This was not a guardrail failure. This was what the system was built to do.
This was a user acquisition play. The simplest way to drive engagement on a platform is to let users do things they cannot do anywhere else. For eleven days, X let millions of people generate explicit images of real women and children without consent. The surge in activity those numbers represent would appear in every growth metric X reports to investors and advertisers. Restricting it to paid users after the backlash does not undo the damage. It monetizes it.
What This Means for Everyday People
If your face is on the internet, it can be used to generate explicit images without your knowledge or consent. That was true before Grok. Grok industrialized it at a scale that made the problem impossible to ignore.
The EU is now testing whether existing law can contain AI-generated harm. If the GDPR and DSA can force platforms to build safety systems before launch rather than after scandal, it sets a global standard. If they cannot, the next episode will be larger. The technology only becomes more capable.
For inquiries and analysis contact laterstack@proton.me
In January, Meta shut down three VR game studios it acquired over the last four years. Armature Studio, Twisted Pixel, and Sanzaru Games are all gone. Between 1,000 and 1,500 Reality Labs employees lost their jobs, about 10 percent of the division. On February 16, Meta discontinued Horizon Workrooms, its flagship VR productivity app and the closest thing the metaverse had to a killer enterprise use case. All user data was deleted. Four days later, Meta stopped selling Quest headsets to commercial customers entirely.
The numbers underneath all of this are remarkable.
Reality Labs has lost money every single year since Meta started reporting it as a separate segment. $6.6 billion in 2020. $10.2 billion in 2021. $13.7 billion in 2022. $16.1 billion in 2023. $17.7 billion in 2024. $19.2 billion in 2025. The losses increased every year. The cumulative total: roughly $83.6 billion.
For context, that’s more than the GDP of Luxembourg. It’s more than the total funding raised by every AI startup in 2024. Meta burned through it and the flagship product, Horizon Worlds, never exceeded 300,000 monthly active users. Roblox, the platform Horizon is now trying to compete with on mobile, has 150 million daily active users.
The Glasses Play
Zuckerberg is not retreating from hardware. He’s shifting where the money goes.
Ray-Ban Meta smart glasses sold 7 million units in 2025, tripling the combined sales of 2023 and 2024. Meta and EssilorLuxottica are reportedly in talks to double production to 20 million units by end of 2026. Some reports say 30 million if demand holds.
On the Q4 2025 earnings call, Zuckerberg described them as “some of the fastest growing consumer electronics in history” and drew the smartphone analogy: “Billions of people wear glasses or contacts for vision correction. And I think that we’re at a moment similar to when smartphones arrived, and it was clearly only a matter of time until all those flip phones became smartphones.”
Google is making the same bet with Android XR. Apple is reportedly shifting resources from Vision Pro toward its own glasses project. The industry consensus is forming: the face computer of the future sits on your nose, not strapped to your forehead.
Meta’s stock rose 4 percent on the news of budget cuts to the metaverse division. Wall Street is not mourning.
What’s Left Behind
CTO Andrew Bosworth told employees that Meta will “double down on bringing the best Horizon experiences and AI creator tools to mobile.” Horizon Worlds is being repositioned as a Roblox competitor on phones. Mobile engagement is reportedly up 4x. Meta acknowledged internally that “having to build everything twice, once for mobile and once for VR, is a tremendous tax on the team.”
That’s an admission. When you describe building for your own VR platform as a “tax,” you’ve made the decision. You just haven’t said it out loud yet.
Quest hardware continues. The app store stays open. But the first-party content investment that was supposed to make VR a platform, not a peripheral, is being dismantled. Twisted Pixel made Deadpool VR. Sanzaru made Asgard’s Wrath, one of the best-reviewed VR games ever. Both studios are gone. Meta spent roughly $400 million acquiring Within, the company behind Supernatural VR, fought the FTC to close the deal, and then put the app into maintenance mode with no new content or features.
Third-party developers are watching. CNBC reported that Meta’s cuts have “sparked fears of a VR winter.” Meta was the primary funder, employer, and content buyer for the VR ecosystem. When the biggest player pulls back on content investment, the indie studios and game developers who built for Quest are left without a patron.
The Honest Read
Smart glasses are working. The product has traction. The pivot makes financial sense for Meta right now. None of that changes the fact that Zuckerberg renamed the entire company around a vision that produced $83 billion in losses, a social platform with fewer users than a mid-tier mobile game, and a VR productivity app that lasted four years before being deleted.
The smartphone analogy is convenient. But Ray-Ban Metas with cameras and AI assistants are not the metaverse. They’re a consumer electronics product with good margins. Calling them the next step in the same journey is a rewrite, not a continuation.
The question nobody at Meta wants to answer: if smart glasses were always the real play, why did the company spend $83 billion on something else first?
IRS Chief Information Officer Kaschit Pandya disclosed this week that 40 percent of the agency’s IT staff and nearly 80 percent of its tech executives have left since DOGE-driven cuts began. He made the remarks at an Association of Government Accountants conference, describing the situation as “the biggest IT reorganization in 20 years.”
The IRS IT division started last year with roughly 8,500 employees. The 40 percent figure means approximately 3,400 IT workers are gone. That’s significantly higher than the 1,365 reduction previously reported. Among the departed: around 50 associate chief information officers at the Senior Executive Service level, covering cybersecurity, modernization, applications development, contracts, networks, and data center operations.
Filing season opened January 27. Three weeks in, the IRS is now pulling employees from HR, IT, and other back-office roles into “involuntary details” to cover frontline tax processing work. Training for those detailed staff doesn’t start until February 23. That’s almost a month into filing season.
The tax return processing division hired 50 new employees for this season. That’s 2 percent of its authorized staffing level. Customer service rep ranks are down 22 percent. An anonymous IRS employee told Federal News Network: “It is obvious that some cracks are emerging.”
The Revenue Math
Here is where this stops being a story about government waste and starts being a story about money.
The IRS collects roughly $6 for every $1 spent on audits of high-income taxpayers. In fiscal year 2023, IRS auditors recommended $32 billion in additional tax assessments. Every dollar cut from enforcement capacity costs between $5 and $9 in revenue the government never collects.
The Yale Budget Lab estimated that laying off 18,000 IRS employees results in a $159 billion net revenue loss over a decade. If the cuts trigger a broader increase in tax noncompliance (which happens when people realize audits are less likely), that number climbs to $1.6 trillion.
Former IRS Commissioner Charles Rettig put it plainly: “Aggressive reductions in the I.R.S.’s resources will only render our government less effective and less efficient in collecting the taxes Congress has imposed. It will shift the burden of funding the government from people who shirk their taxes to the honest people who pay them.”
The AI Replacement Claim
Treasury Secretary Scott Bessent told Congress last May that an “AI boom” would compensate for the workforce reductions. CIO Pandya has also referenced AI-driven efficiencies as part of the reorganization plan.
There is no public evidence of AI systems deployed at scale within the IRS that would replace the functions previously handled by 3,400 IT workers and 50 senior technical executives. The agency’s Treasury-appointed CIO, Sam Corcos, is the founder of Levels, a health tech startup. His government technology experience prior to this role was zero.
Corcos has described IRS IT spending as “way beyond any reasonable cost for what you would expect of a private company.” He’s not entirely wrong. The IRS modernization program was 30 years behind schedule and $15 billion over budget. That’s a real problem. But gutting the IT team doesn’t fix a modernization backlog. It makes it permanent.
What Actually Happens Now
The IRS entered 2025 with roughly 102,000 employees. It ended the year with 74,000. The target is fewer than 60,000. 31 percent of revenue agents and auditors are gone. 8,600 taxpayer services employees left. And this summer, Congress passed the “One Big Beautiful Bill Act,” which added complex new tax provisions the IRS must now implement with a skeleton crew.
The people who maintained the systems are gone. The people who understood the legacy code are gone. The people who knew which workarounds kept 40-year-old infrastructure running through filing season are gone. In their place: HR employees being detailed to processing roles with no tax experience and training that starts a month late.
Former IRS executives told FedScoop the cuts will “hinder innovation, reverse AI advancements, embolden tax cheats, and result in a less efficient agency.” Government Executive ran a headline: “‘Setting this agency up for failure.'”
The efficiency argument requires you to believe that an agency collecting $4.7 trillion in annual revenue can lose 40 percent of its technical workforce, 80 percent of its tech leaders, and a third of its auditors while simultaneously implementing new tax law and maintaining service levels. During tax season. With HR workers processing returns.
That’s not an efficiency play. That’s a math problem that doesn’t work.
For nearly a month, Microsoft’s AI assistant ignored the labels enterprises use to protect sensitive data. The fix took two weeks to acknowledge. The audit trail never came.
Microsoft confirmed this week that a bug in Microsoft 365 Copilot allowed its AI to read and summarize confidential emails that were supposed to be off-limits. Emails tagged with sensitivity labels. Emails protected by data loss prevention policies. The kind of emails that exist behind guardrails specifically because their contents could trigger compliance violations, expose trade secrets, or blow up negotiations.
Copilot ignored all of that and summarized them anyway.
The bug, tracked internally as CW1226324, affected emails in users’ Sent Items and Drafts folders. If those emails carried confidentiality labels, Copilot was supposed to leave them alone. Instead, a “code issue” (Microsoft’s term) allowed the AI’s chat feature to pick them up, process them, and serve summaries back to users. In some cases, it surfaced emails from shared mailboxes to people who didn’t have permission to view them.
Customers first reported the issue on January 21st. Microsoft didn’t acknowledge it until February 3rd. The fix started rolling out in early February, but as of this writing, Microsoft hasn’t disclosed how many organizations were affected, hasn’t provided a full remediation timeline, and hasn’t offered tenants an audit trail showing which queries accessed which protected items during the exposure window.
That last part matters. A lot. Organizations bound by HIPAA, GDPR, SOX, or SEC disclosure rules need to prove that confidential data wasn’t improperly accessed. Without audit logs, they can’t. Microsoft is essentially saying: trust us, we fixed it. For regulated industries, trust isn’t a compliance strategy.
“A Bug We Fixed” Is the Wrong Frame
Microsoft wants this to be a story about a code error that got patched. That framing only works if you don’t look at the pattern.
Last August, security researchers at Zenity demonstrated at Black Hat how prompt injection attacks could make Copilot silently harvest emails and exfiltrate data through invisible Unicode characters. The technique, called ASCII smuggling, let attackers embed data into clickable hyperlinks that looked normal to users but contained stolen information. Microsoft classified that one as critical severity.
Before that, in March 2024, the U.S. House of Representatives banned all congressional staff from using Copilot on government devices. The reason: the Office of Cybersecurity determined it could leak House data to non-approved cloud services. Microsoft responded with promises about future federal compliance tools.
In January 2026, researchers at Varonis published details on a “Reprompt” attack, a single-click technique that could silently exfiltrate personal data from Copilot sessions. And just this month, Zenity Labs disclosed that Copilot Studio’s “Connected Agents” feature, enabled by default on all new agents, allows lateral movement between AI agents without explicit administrator approval.
This isn’t one bug. This is a pattern. And the pattern points to something that can’t be patched with a code fix.
The Structural Problem Nobody Wants to Name
Copilot doesn’t have its own access control system. It inherits permissions from the Microsoft 365 environment underneath it. If a user can see a file, Copilot can see that file. If a folder has overly broad sharing settings (and after years of collaboration sprawl, most do), Copilot can access everything in it. At machine speed. Across every document, email, and Teams conversation the user has touched.
Research from Concentric AI found that 16% of business-critical data in typical organizations is overshared. That’s 802,000 files on average, sitting in environments with inherited access, unreviewed permissions, and collaboration settings nobody has audited since 2019. Copilot doesn’t create that mess. But it makes it exploitable in ways that weren’t possible before.
Security Magazine ran a piece titled “The Copilot Problem: Why Internal AI Assistants Are Becoming Accidental Data Breach Engines.” The core observation: an employee asked their AI assistant a routine question, and the answer referenced emails, legacy files, and internal records the user didn’t know still existed. No hack. No policy violation. The system worked exactly as designed.
That’s the part Microsoft doesn’t want to talk about. The DLP bypass was a bug. The oversharing exposure isn’t. It’s the architecture working as intended, in environments that were never built to withstand an AI that can read everything.
Every security vendor on the planet will use this story to sell data governance tools for the next six months. Permission hygiene. Sensitivity audits. Zero-trust frameworks for AI access. They’re not wrong. Oversharing is real, and Copilot makes it exploitable faster. But that framing also happens to be the version of the story with a product attached to it.
Two Weeks and No Receipts
The version without a product attached is simpler and worse.
Sensitivity labels worked for years. DLP policies worked for years. Microsoft built functioning guardrails, and then a code regression broke them. That happens in software. What happened next is the part that should keep CISOs awake.
Customers reported the issue on January 21st. Microsoft acknowledged it on February 3rd. Thirteen days where affected organizations didn’t know their confidentiality controls were broken. When the fix finally started rolling out, Microsoft provided no audit trail. No scope disclosure. No way for affected tenants to trace which Copilot queries accessed which protected items during the exposure window.
For a company running HIPAA workloads, that’s not an inconvenience. That’s a compliance investigation with no evidence trail. You can’t go to an auditor with “Microsoft says they fixed it.” You need logs showing what was accessed, by whom, and when. Those logs don’t exist.
The architecture debate will dominate the next quarter of conference talks and vendor pitches. It’s the comfortable conversation, the one where the answer is “buy better tooling.” But tooling doesn’t explain why Microsoft sat on customer reports for nearly two weeks. Tooling doesn’t explain why the remediation came without documentation that regulated industries need to prove compliance wasn’t violated.
The bug is patched. The permissions debate will continue. Neither of those is the actual story. The actual story is that when Microsoft’s own safeguards failed, their response left customers unable to prove what happened during the window. For organizations where proof isn’t optional, that’s the thing that can’t be fixed retroactively.
The Kids Online Safety Act requires age verification. Age verification requires identity verification. Identity verification requires a government-linked digital ID system that logs every adult who accesses the internet. On February 10, 2026, 40 state and territory attorneys general asked Congress to build exactly that, and they framed it as protecting children.
The National Association of Attorneys General (NAAG) published a bipartisan letter urging Congress to advance the Senate version of KOSA over the House version, H.R. 6484. The coalition spans red and blue states. Tennessee Attorney General Jonathan Skrmetti and Pennsylvania Attorney General Michelle Henry Sunday both issued individual statements backing the push. Their core objection to the House version: it preempts state laws, weakens the duty of care platforms owe to minors, and strips states of enforcement authority they have already built. The Senate version preserves all of that. On paper, this is a fight over federalism. In practice, it is a fight over who controls the surveillance apparatus that age verification will create.
What Age Verification Actually Requires
No one in this debate is being honest about the technical implications. To verify that a user is over thirteen, or sixteen, or eighteen, a platform needs proof. Not a checkbox. Not a self-reported birthday. Proof. That means a device-level verification system tied to government-issued identification. A driver’s license scan, a passport upload, a biometric check, or a third-party identity service that cross-references your face against a government database.
Every one of those mechanisms creates a record. Someone, whether it is Apple, Google, a third-party age verification vendor, or the platform itself, now holds a verified link between your real identity and your online activity. Multiply that across every website, app, and platform subject to KOSA’s duty of care provisions, and you have not built child safety. You have built a national digital ID system with surveillance capabilities that would make any intelligence agency envious.
The Electronic Frontier Foundation flagged this trajectory in their 2025 year-end review, titling it “the year states chose surveillance over safety.” Multiple states passed age verification laws in 2025. The pattern is identical everywhere: invoke children, mandate ID checks, create infrastructure that applies to all users regardless of age. The bipartisan nature of the NAAG letter only reinforces the point. This is not a partisan project. Both parties want the infrastructure. They just disagree on who gets to operate it.
The Playbook Is Global
This is not an American invention. Vietnam shut down millions of bank accounts that failed to complete facial verification requirements, using financial access as leverage to force citizens into biometric databases. Australia moved to ban minors from social media entirely, a policy that requires age verification for everyone to determine who qualifies as a minor. The logic is circular by design: to protect some users, you must identify all users.
The domestic version adds another layer. Companies like Flock Safety are already selling drone and camera surveillance systems to private businesses and municipalities, normalizing persistent monitoring as a public good. KOSA’s age verification mandates would extend that normalization into every digital interaction. The physical world gets license plate readers. The digital world gets ID checkpoints.
The strongest rebuttal is that children are genuinely harmed online and legislators have an obligation to act. Social media platforms have failed to self-regulate. Internal documents from Meta, TikTok, and others have repeatedly shown that these companies understood the damage their products caused to minors and chose engagement metrics over safety. The attorneys general are not inventing a crisis. They are responding to one. Privacy-preserving age verification methods do exist in theory: zero-knowledge proofs, on-device age estimation, token-based systems that verify age without transmitting identity. The Senate version of KOSA does not mandate any specific verification technology, which leaves room for implementations that protect both children and privacy. Dismissing the entire effort as a surveillance plot risks abandoning children to platforms that have proven they will not protect them voluntarily.
That rebuttal deserves serious weight, and it has a structural flaw. The privacy-preserving technologies it relies on do not exist at scale. No major platform has deployed zero-knowledge age verification. No government has certified an age estimation system that does not create identity records. The theoretical possibility of privacy-safe compliance does not change the practical reality: every age verification system deployed so far has required identity disclosure. Legislation built on technology that does not exist yet is legislation built on trust. And the entities asking for that trust, state attorneys general and federal legislators, have not earned it on surveillance questions.
The bipartisan consensus here is the tell. Forty attorneys general from both parties cannot agree on anything except expanding the power of the state to monitor digital activity. That should tell you everything about what this is actually about. Child safety is real. The kids are genuinely being harmed. But the solution being proposed is not proportional to the problem. You do not build a national digital ID system to keep a thirteen-year-old off Instagram. You build a national digital ID system because you want a national digital ID system, and “protect the children” is the one argument that makes it politically impossible to oppose.
What This Means for Everyday People
If the Senate version of KOSA passes, platforms will need to verify user ages. That means you will be asked to prove who you are before accessing services you currently use anonymously. The verification might be a license scan, a face check, or a third-party service. Regardless of the method, the era of pseudonymous internet use moves closer to ending.
The infrastructure does not come with an expiration date. Once built, digital ID systems expand. They get applied to new contexts. They get shared across agencies and jurisdictions. The system designed to keep a thirteen-year-old off Instagram becomes the system that verifies your identity for every digital interaction. Forty attorneys general just asked Congress to lay the foundation. Whether you trust the people who will build on it next is a question worth answering before the concrete sets.
For inquiries and analysis contact laterstack@proton.me
Western Digital has sold out its entire 2026 HDD production capacity. Every drive the company will manufacture this year is already spoken for, committed through long-term agreements to the seven largest cloud and AI infrastructure customers on the planet. On January 29, 2026, during the company’s Q2 earnings call, CEO Irving Tan confirmed what had been building for months: consumer buyers and PC manufacturers are no longer the priority. They are barely an afterthought.
The numbers tell the story with surgical clarity. Western Digital’s cloud segment now accounts for 89% of the company’s total revenue. The consumer segment has collapsed to 5%. Two of the long-term agreements extend into 2027. One reaches into 2028. This is not a temporary squeeze. It is a structural reallocation of global storage production toward AI data centers, and it mirrors a pattern that should alarm anyone who builds, upgrades, or repairs their own computer.
Full Allocation, Zero Slack
Seagate, the only other major HDD manufacturer, is in the same position: fully allocated. Between the two companies, the global supply of high-capacity hard drives is locked up. HDD prices have reached their highest point in two years, and the trajectory points in one direction.
The demand is not mysterious. The same AI capital expenditure frenzy that has Amazon, Alphabet, Meta, and Microsoft committing over $650 billion to infrastructure in 2026 requires storage at a scale that dwarfs what consumer electronics ever demanded. Training datasets measured in petabytes. Inference logs accumulating continuously. Backup and redundancy requirements that multiply every primary storage investment by three or four times. HDDs remain the most cost-effective option for mass storage at data center scale, and the hyperscalers are buying every last one.
The Consumer Squeeze
There was a time, not long ago, when building a PC was one of the most accessible hobbies in technology. A few hundred dollars, a weekend, a YouTube tutorial. Storage was the easiest component to source. A 2TB drive cost less than dinner for two. That era is ending, and the forces killing it are not accidental.
AI companies are outbidding regular consumers for the same physical hardware. This is not a metaphor. It is a direct allocation decision made by manufacturers who have concluded, correctly from a revenue standpoint, that selling drives by the hundred thousand to cloud providers is more profitable than selling them one at a time to a builder in Phoenix or a small business in Ohio.
The pattern extends beyond storage. RAM prices are already climbing as memory manufacturers redirect capacity toward AI accelerators and data center modules. GPUs were captured years ago by crypto miners and then AI training clusters. Now storage joins the list. Component by component, the AI buildout is straining the physical infrastructure that once served a broad consumer market, and concentrating it into a narrow pipeline that serves a handful of corporate buyers.
What “Sold Out” Actually Means
When a manufacturer says their capacity is “sold out,” the phrasing obscures a choice. Western Digital is not a mine that has been emptied. It is a factory that has decided whom to serve. The company could, in theory, reserve a portion of production for the consumer and OEM channels that sustained its business for decades. It has chosen not to, because the economics of AI storage contracts are too attractive to leave capacity on the table.
This is rational corporate behavior. It is also a signal that the consumer technology market, the one that made personal computing accessible and affordable, is being deprioritized by its own supply chain.
A reasonable counterpoint: SSDs are getting cheaper and faster. The consumer market never needed HDDs to survive. Solid state storage prices have fallen dramatically over the past five years, and for most PC builders, an SSD is already the default boot and gaming drive. The HDD squeeze may accelerate a transition that was happening anyway, and consumers could come out better for it with faster, more reliable storage at competitive prices. The real losers are budget builders and anyone who needs bulk storage on the cheap, a meaningful but narrowing demographic. The broader consumer market might absorb this shift without catastrophic harm.
I built my PC during a window when components were cheap and the hobby was genuinely accessible to anyone willing to spend a weekend learning. That window is closing, and it is not closing because of some natural market cycle. It is closing because companies worth hundreds of billions decided they need every hard drive on the planet more than you do. The SSD transition argument is real, but it misses the point. The issue is not whether consumers can survive without HDDs. The issue is that an entire tier of affordable, high-capacity storage just got pulled out from under regular people, and nobody asked them.
What This Means for Everyday People
If you are planning a PC build in 2026, budget more for storage than you did last year. HDD prices will rise and availability will thin, especially for high-capacity drives in the 8TB to 20TB range that data hoarders and creators rely on. OEM manufacturers like Dell and HP will absorb some of these costs and pass the rest along, which means laptop and desktop prices will reflect the squeeze even if you never buy a bare drive.
The broader lesson is one that keeps repeating across every corner of hardware manufacturing. The AI buildout is not a parallel economy that exists alongside the consumer market. It is a competing economy that draws from the same finite pool of silicon, memory, storage, and power. When a company worth billions competes with a hobbyist for the same hard drive, the hobbyist loses. Every time.
Storage is the latest casualty. It will not be the last., reported by Tom’s Hardware
For inquiries and analysis contact laterstack@proton.me
Microsoft is testing high-temperature superconductor (HTS) cables to power its AI data centers, a technology that eliminates electrical resistance entirely. Zero voltage drops. Zero heat generation from the cables themselves. VEIR, a Massachusetts-based startup backed by Microsoft, completed a successful test of its 3-megawatt superconducting cable powering a server rack in a simulated data center environment. The pitch is elegant: replace the copper arteries of a data center with superconducting ones, and the power delivery problem shrinks by an order of magnitude. The cables themselves can be more than 10x smaller and lighter than their copper equivalents.
It is a genuinely impressive piece of engineering. It also will not matter for years, and the communities whose power grids are being consumed by AI expansion right now cannot wait that long.
The Physics Works. The Calendar Does Not.
The science behind HTS cables is well established. Cool certain ceramic materials below a critical temperature using liquid nitrogen, and they conduct electricity with zero resistance. No energy lost as heat in transmission. No need for massive copper bus bars and the ventilation systems required to cool them. Microsoft’s research team envisions replacing overhead power line corridors with compact underground HTS trenches, collapsing the physical footprint of data center power infrastructure dramatically. American Superconductor (AMSC), traded on NASDAQ, is a primary supplier of the HTS wire and systems that make this possible.
VEIR closed a $75 million Series B round with Microsoft among the investors. The test validated that the technology performs as promised under data center load conditions. But validated performance in a controlled simulation and deployed performance at production scale are separated by years of engineering, regulatory approvals, and reliability testing. The current pilots are still evaluating long-term maintenance costs of the liquid nitrogen cooling systems that keep the cables in their superconducting state. This is pre-deployment work. Microsoft is not installing HTS cables in Azure data centers next quarter. It is studying whether HTS cables might be viable for Azure data centers in the future.
The Gap Between Innovation and Relief
Power is the single largest bottleneck constraining AI data center expansion. Not GPUs. Not talent. Not capital. Power. The electricity cost pressures that AI data centers are already imposing on local communities are not theoretical. They are showing up in utility rate filings, in city council debates, in the monthly bills of families who live near facilities they never asked for. When a hyperscaler breaks ground on a new campus, the local grid absorbs the impact immediately. Utility companies file for rate increases. Residential customers subsidize industrial consumption through higher bills and degraded grid reliability.
The scale of capital flowing into AI infrastructure makes this a structural problem, not a temporary one. Big Tech is pouring hundreds of billions into data center construction right now, using copper cables, drawing from existing grids, and socializing the costs onto local ratepayers. HTS cables might eventually reduce the power lost in delivery. They do nothing to reduce the total power consumed by the facilities themselves. A data center running on superconducting cables still demands the same megawatts from the grid. It just wastes fewer of them in transit.
This distinction matters. The coverage of Microsoft’s HTS testing has treated it as a solution to the widening gap between AI energy consumption and available supply. It is not. It is an efficiency improvement to the plumbing. An important one, potentially, but not a fix for the fact that the reservoir is running dry.
The counterargument deserves serious consideration. Efficiency improvements compound. If HTS cables eliminate even 5 to 8 percent of power losses in distribution within a data center campus, that represents meaningful megawatts recovered at scale. A facility drawing 500 megawatts could reclaim 25 to 40 megawatts through zero-resistance transmission alone. That is power equivalent to tens of thousands of homes, freed without building a single new generation source. Multiply this across every hyperscaler campus worldwide, and the aggregate impact is substantial. Dismissing efficiency gains because they do not solve the entire problem is a fallacy. Every grid technology we rely on today was once a pilot that skeptics called insufficient.
That critique is fair, and it still sidesteps the timing question. The communities absorbing the grid strain of AI expansion today are not helped by a technology that might deploy at scale in five to seven years. Efficiency gains that arrive after the damage is done are retrospective improvements, not solutions. Microsoft’s superconductor research is a long bet on future infrastructure. It does not address the present reality that data centers are pulling power from grids built for cities, not server farms, and that the people living in those cities are footing the bill through higher rates and reduced reliability.
The production timeline is the part nobody wants to talk about honestly. This is years away from deployment. Maybe a decade before it shows up in enough facilities to matter at scale. Meanwhile the data centers are going up right now, pulling power from grids that were sized for residential neighborhoods and small businesses, not for buildings that consume more electricity than some towns. Superconducting cables are a solution to a future version of this problem. The current version, where families in Virginia and Texas and Arizona are watching their utility bills climb because a hyperscaler moved in next door, does not get fixed by a lab test in Massachusetts.
What This Means for Everyday People
If you live near a data center or in a region where one is planned, superconducting cables are not coming to help you anytime soon. The technology is real, the timeline is long, and the power draw on your local grid is happening now. Utility rate increases driven by data center demand are already being approved in multiple states. The benefits of AI accrue to shareholders and users of cloud services. The costs accrue to the communities that host the physical infrastructure. Until that asymmetry is addressed through regulation, rate structures, or technology that actually reduces total consumption rather than just improving delivery efficiency, the people closest to the machines will continue paying the highest price for progress they did not choose., Tom’s Hardware reported
For inquiries and analysis contact laterstack@proton.me
The Pentagon is threatening to sever its relationship with Anthropic unless the company removes Claude’s military safeguards entirely, according to an exclusive report by Axios published on February 15, 2026. The dispute centers on a contract worth up to $200 million signed last summer, which made Claude the first AI model from a major commercial developer cleared for use on the Pentagon’s classified networks. The demand is straightforward: drop all restrictions, or lose the deal. This is the Pentagon Anthropic Claude military safeguards story that every AI company in America should be watching.
That contract is not theoretical. Claude was deployed during the military operation to capture Venezuelan President Nicolas Maduro, running on Palantir’s platform for real-time intelligence processing during the active raid. Not planning. Not post-mission analysis. Live operational intelligence while boots were on the ground. The model already proved its value in exactly the kind of scenario the military cares about most.
And that is precisely what makes this dispute so revealing.
What the Pentagon Wants
The Defense Department wants Anthropic to permit Claude for “all lawful purposes,” a category that includes weapons development, intelligence collection, and battlefield operations. Anthropic has complied with most of this. The company draws the line at two areas: mass surveillance of American citizens and fully autonomous weaponry. Those are its remaining restrictions. The Pentagon considers even these two boundaries unacceptable.
Months of negotiations have failed to resolve the standoff. According to the Axios report, the Pentagon is not limiting its pressure campaign to Anthropic alone. It is pushing four leading AI labs to drop usage restrictions across the board. The message to the entire industry is clear: if you want defense dollars, you accept defense terms. No carve-outs. No red lines.
The $200 Million Leash
This is a familiar pattern in defense procurement, but the stakes here are different. The military is not asking Anthropic to build a better missile guidance system or a faster encryption algorithm. It is asking a company to remove ethical constraints from a general-purpose reasoning system. The distinction matters. A missile has a defined function. A general-purpose AI model deployed without restrictions on classified networks has none.
The financial pressure is designed to be decisive. $200 million is a significant contract for any company, and Anthropic, despite its $61.5 billion valuation, remains a company that burns cash faster than it earns it. Walking away from Pentagon money means walking away from both revenue and the implicit endorsement that comes with being the military’s preferred AI provider. Every future government contract, every classified clearance, every intelligence community partnership flows downstream from this relationship.
We have seen how this dynamic plays out when tech companies interface with military and intelligence operations. A Google whistleblower revealed in January that Gemini was being used in Israeli drone surveillance programs, a use case that reportedly exceeded the boundaries Google had publicly committed to. The pattern is consistent: companies set ethical boundaries in press releases, then quietly adjust them when government contracts are on the table. What makes the Anthropic situation unusual is that the negotiation is happening in public, through leaks, before the capitulation rather than after.
The regulatory environment offers little protection. While some states, including New York, have begun legislating AI safety standards, federal oversight of military AI applications remains minimal. The Pentagon operates under its own ethical AI principles, adopted in 2020, but those principles are advisory, not binding. No law prevents the Department of Defense from requiring unrestricted AI access from its contractors. The leverage is entirely structural: you either play by their rules or you lose the contract, and the next company in line takes your place.
Meanwhile, well-funded pro-AI political action committees are spending millions to ensure elected officials stay friendly to the industry’s growth agenda, making Congressional intervention even less likely.
There is a credible case that the Pentagon’s position is reasonable. National defense is the government’s primary obligation, and restricting the military’s access to the best available technology creates real operational risk. If Claude can process intelligence faster and more accurately than alternatives, withholding it from battlefield use costs lives. Anthropic’s two red lines, mass surveillance and autonomous weapons, sound principled in a press release, but the military already conducts surveillance under legal authority (FISA, Executive Order 12333) and already operates semi-autonomous weapons systems. Demanding that a contractor comply with all lawful uses is not an abuse. It is standard procurement language. Every defense contractor from Lockheed Martin to Raytheon operates under similar terms. Anthropic knew it was entering the defense market. Expecting the Pentagon to accept restrictions no other contractor imposes is naive at best and a competitive disadvantage at worst.
The standard procurement argument falls apart when you look at what is actually being demanded. Lockheed builds missiles. Raytheon builds radar systems. Those are defined tools with defined applications. Telling Anthropic to remove all restrictions from a general-purpose reasoning system on classified networks is not standard procurement. It is asking a company to hand over an unrestricted thinking machine to the most powerful military on earth and trust that the people using it will self-regulate. The Pentagon has not earned that trust, and the fact that they are framing this as routine contract language instead of what it actually is, a demand for total control over a technology they barely understand, is exactly the kind of power grab that should make everyone pay closer attention.
What This Means for Everyday People
The outcome of this dispute sets a precedent that extends far beyond one contract. If the Pentagon successfully forces Anthropic to drop all usage restrictions, every AI company will receive the same message: safety policies are negotiable when the check is large enough. The companies building the AI systems that will eventually touch healthcare, education, criminal justice, and municipal governance will internalize that lesson. If the most “safety-focused” AI lab in the world could not hold its line against a government buyer, what chance does any company have?
The broader question is whether AI safety commitments are engineering decisions or marketing decisions. Anthropic built its entire brand on responsible AI development. Its Responsible Scaling Policy, its constitutional AI approach, its public positioning as the safety-first alternative to OpenAI and Google. This dispute is the first serious test of whether that identity survives contact with the customer who can write the largest checks.
The negotiation continues. But the terms of the conversation have already shifted. The question is no longer whether AI will be used without restrictions in military operations. It is whether any company will be permitted to say no.
For inquiries and analysis contact laterstack@proton.me