Anthropic built an AI model that found thousands of zero-day vulnerabilities in every major operating system and web browser. The Federal Reserve chair and the Treasury secretary called bank CEOs to discuss it. The company says there is a six-to-twelve month window to patch the flaws before adversaries build models that can do the same thing. The cybersecurity industry says the threat was already here. Both are right.

Claude Mythos Preview is the model. It is not yet publicly released. In controlled testing, it surpassed all but the most skilled humans at finding and exploiting software vulnerabilities, identifying flaws that had existed undetected for decades, including a 27-year-old bug in OpenBSD and a 17-year-old remote code execution flaw in FreeBSD. Anthropic CEO Dario Amodei described the current period as a “moment of danger” and warned of “some enormous increase in the amount of vulnerabilities, in the amount of breaches, in the financial damage that’s done from ransomware on schools, hospitals, not to mention banks.”

The discovery

Mozilla released Firefox 150 with fixes for 271 security vulnerabilities identified by Mythos in a single evaluation pass. The number is striking not because Firefox is unusually insecure but because no human team had found them. The vulnerabilities had accumulated across years of development, each one a potential entry point for an attacker with the right tools. Mythos found all 271 in one run.

The 💜 of EU tech

Advertisement

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

The model’s capability raises a question that the cybersecurity industry has been theorising about for years and now must answer practically: what happens when the cost of finding vulnerabilities drops to near zero? The traditional economics of cybersecurity depend on the asymmetry between attackers, who must find one flaw, and defenders, who must secure all of them. Mythos collapses the cost on both sides. Defenders can now scan their entire codebase for flaws they never knew existed. Attackers, once they build or obtain equivalent models, can do the same.

The response

Anthropic chose a controlled rollout, which it calls Project Glasswing. Approximately 40 technology companies and institutions have initial access to Mythos to bolster their systems. The list does not include most central banks and governments. The asymmetry is intentional: give defenders a head start before the capability becomes widely available.

The response from financial regulators was immediate. Federal Reserve Chairman Jerome Powell and Treasury Secretary Scott Bessent convened a meeting with major US bank CEOs to discuss the cyber risks raised by Mythos. The IMF flagged AI-powered cyber threats to the global banking system. The concern is not that Mythos itself will be used to attack banks. It is that the capability Mythos demonstrates, automated discovery of vulnerabilities at superhuman speed, will be replicated by adversaries who are not bound by Anthropic’s responsible disclosure practices.

Anthropic shipped financial services agents the day after announcing its 1.5 billion dollar Wall Street joint venture, a sequence that illustrates the company’s dual positioning: it is simultaneously the entity warning banks about AI-powered cyber threats and the entity selling AI products to banks. The joint venture with Blackstone and Hellman and Friedman is anchored at approximately 300 million dollars from Anthropic and will deploy AI across private equity operations.

The race

Amodei’s six-to-twelve month window is a prediction about how long it will take Chinese AI companies to build models with equivalent vulnerability-discovery capabilities. The window is not about whether adversaries will develop the capability. It is about when. The controlled rollout of Mythos is designed to give the companies that receive early access enough time to patch their most critical flaws before the window closes.

OpenAI released GPT-5.4-Cyber for vetted security teams, scaling its Trusted Access programme in direct response to the Mythos disclosure. The competitive dynamic between Anthropic and OpenAI has extended from commercial AI products into cybersecurity, with both companies positioning themselves as defenders of the software infrastructure their own models could be used to compromise.

Researchers have already demonstrated that AI agents from Anthropic, Google, and Microsoft can be hijacked via prompt injection to steal API keys and tokens, and all three vendors paid bounties but skipped public disclosure. The irony is precise: the AI agents that companies deploy to improve security are themselves vulnerable to attacks that could compromise the systems they are meant to protect.

The tension

The cybersecurity community’s response to the Mythos disclosure has been a mixture of alarm and scepticism. Security researchers note that AI-assisted vulnerability discovery has been developing for years and that the capabilities Mythos demonstrates, while impressive in scale, are an acceleration of existing trends rather than a discontinuous leap. The threat of AI-powered cyberattacks was identified by the UK’s National Cyber Security Centre more than a year ago. What Mythos changes is not the existence of the threat but the specificity of the evidence.

Anthropic occupies an unusual position. It is a company whose business model depends on selling AI capabilities to enterprises, including banks, while simultaneously arguing that AI capabilities of the kind it is developing pose an existential threat to the cybersecurity of those same enterprises. The resolution of the contradiction is commercial: Anthropic’s pitch is that you need its AI to defend against AI of the kind it builds. The logic is circular but the threat is real.

The 271 Firefox vulnerabilities were real. The 27-year-old OpenBSD bug was real. The meeting between the Fed chair and bank CEOs was real. The question is not whether AI will transform cybersecurity. The question is whether the six-to-twelve months Amodei describes is enough time to patch decades of accumulated vulnerabilities across every operating system, browser, and financial platform in production, or whether the window is an estimate designed to create urgency for a problem that cannot be solved on any timeline. Mythos found the flaws. Fixing them is a human problem.

In modern datacenters, storage can live anywhere — local to the machine, remotely accessed over the network, and/or shared between systems.

The next generation of servers will treat system memory in much the same way. Systems will still have some local DDR5, but the bulk of it will be remotely accessed from what some have taken to calling the memory godbox.

The ongoing DRAM shortage has created a perfect storm for the proliferation of the appliances, which not only allow for memory to be pooled, but also data stored in that memory to be shared by multiple machines simultaneously. In effect, memory becomes a fungible resource.

More importantly, your next round of servers will probably support the tech, if they don’t already.

CXL finally has its moment to shine

The technology at the heart of these memory godboxes isn’t new. Compute Express Link (CXL) has been slowly gaining traction since its introduction seven years ago.

As a quick refresher, CXL defines a common, cache-coherent interface for connecting CPUs, memory, accelerators, and other peripherals.

The technology comes in a couple of different flavors: CXL.mem, CXL.cache, and CXL.io, which, as a whole, have implications for disaggregated compute. Imagine a rack with a CPU node, GPU node, memory node, and storage node, which can talk to one another completely independently. That’s the core idea behind CXL.

CXL piggybacks off the PCIe standard, which means in theory it should be broadly compatible, but, up to this point, it’s primarily been used with memory devices.

The 1.0 spec opened the door to memory expansion modules, which allow you to add more memory by slotting them into a CXL-compatible PCIe slot. To the operating system — assuming you’re running Linux that is — the extra memory is largely transparent, showing up as if it were attached to another CPU socket, just one without any additional compute.

The 2.0 spec, which showed up in 2020, added basic support for switching, which meant memory could be pooled and then allocated to any number of connected systems.

AMD and Intel’s current crop of Epycs and Xeons already support these appliances. But while the memory can be partitioned and reallocated to different machines as needed, two machines can’t work on the same data simultaneously.

Unless you were memory-constrained, the added complexity of CXL 2.0 didn’t offer much benefit over simply using higher capacity DIMMs in the first place. 

At least, not until memory prices went through the roof.

Where things really get interesting is when the 3.0 spec arrives in AMD and Intel’s next-generation of Epycs and Xeons. In fact, from what we understand, Amazon’s Graviton5 CPUs we looked at in December already support the spec.

CXL 3.0 introduces two key capabilities that make it particularly interesting for memory appliances. The first is support for larger topologies: Multiple CXL switches can be stitched together into a fabric. The second is support for memory sharing: Rather than partitioning memory into slices only accessible to one machine at a time, memory can be shared between machines.

In theory this could allow two machines running the same set of workloads to use the memory closer to that of one. It’s a bit like deduplication for memory. In fact, we already do this in virtualized environments like KVM, but it now works across machines.

There are security and performance implications to all of this. Thankfully in CXL 3.1 and later, the consortium introduced confidential computing capabilities into the spec, allowing for isolation where necessary.

On the performance end of things, CXL 3.0 moves to PCIe 6.0 as a baseline, which provides 16 GB/s of bidirectional bandwidth per lane. Assuming 64 lanes of CXL per CPU, that works out to an additional 512 GB/s of bandwidth. So memory bandwidth shouldn’t be too much of an issue for most applications. Latency, on the other hand, is a different story. 

CXL-attached memory is going to add some latency. However, as we’ve previously discussed, the latency isn’t as bad as you’re probably thinking — on the order of a NUMA hop, or about 170 to 250 nanoseconds of round trip latency. Obviously, the farther the memory appliance is from the host CPU, the worse the latency is going to be.

Late last year, the CXL consortium ratified the 4.0 spec, which among other things doubles the bandwidth from 16 GB/s per lane to 32 GB/s by re-basing on PCIe 7.0. However, it’ll be a while before we see appliances based on the spec.

Where’s my memory godbox?

There are several companies developing hardware for these kinds of networked memory appliances. 

Panmnesia’s CXL 3.2-compatible PanSwitch is one of the most sophisticated examples. The switch features 256 lanes of connectivity for CXL memory modules, devices, or CPUs to connect, pool, or share resources.

If you’re okay with memory pooling and don’t need the niceties of CXL 3.0, then there are already several memory appliances available that are compatible with the latest generation of Xeon 6 and Epyc Turin processors.

Liqid’s composable memory platform, for example, can provide a pool of up to 100 TB of DDR5 to as many as 32 hosts. Meanwhile, UnifabriX Max systems provide CXL 1.1 or 2.0 connectivity to 16 or more systems with support for CXL 3.2 already in the works.

We suspect that as more CXL 3.0 compatible CPUs and GPUs hit the market, more of these memory godboxes will appear.

AI eats everything

Don’t get too excited. While network attached memory has the potential to reduce an enterprise’s infrastructure spend, those same qualities make it attractive for the very thing driving the memory shortage in the first place.

AI adoption has driven demand for DRAM off the charts. In addition to the HBM used by GPUs, DDR5 is being used for key value cache offload during inference.

These KV caches store model state and can chew significant amounts of memory — often more than the model itself — in multi-tenant serving scenarios.

Rather than discard these caches and recompile them when the model state is restored, it’s more efficient to offload them to system memory and eventually flash storage.

The problem with using flash storage is that it has a finite write endurance. After a while it wears out. Instead, CXL memory vendors are positioning the tech as a more resilient alternative.

That’s bad news for enterprises looking to these memory godboxes for salvation from the RAMpocalypse. ®

Designer Matty Benedetto of Unnecessary Inventions runs a studio in Vermont where he makes contraptions to tackle problems that no one has ever asked about. His most recent project mixes two known elements to create something new, which has the potential to change how teams handle lengthy discussions around a table. He transformed conventional office chairs into a full seesaw that rocks up and down while spinning in a complete circle.

Benedetto started simple by collecting a couple of worn-out office chairs from storage. He wanted seats that everyone was familiar with, so no one felt out of place when they sat down. A simple test compared the wheels on each base to determine which pair rolled and slid the best across a floor. These results allowed him to choose the right parts without guesswork. He then gently separated the chairs, keeping the seats and center supports intact. A small 3D printed model allowed him to see how the elements would connect and move together. The initial chair bases already spun freely in all directions, so he retained that motion for the finished form. He then designed a bespoke bracket to connect everything at the midway point.

Sale

N-GEN GAMING Video Gaming Chair with Footrest Lumbar Support for Home Office High Back Recliner Height…

• Racing Style for Long Sessions – High-back gaming chair with ergonomic racing design, ideal for long hours at your gaming desk or home office.
• Ergonomic Support – Comes with a removable headrest, lower back pillow, and pull-out footrest to reduce pressure and support healthy posture during…
• Quality Materials – Supportive high-density foam cushions, breathable PU leather, and a vibrant finish combine for lasting comfort and a refined look.

Ball bearings in the new bracket provided smooth, effortless seesaw movement as needed. He assessed the distances and opted on chairs spaced ten feet apart along a robust metal tube that cost him $100. That tube served as the main beam, measuring a solid 5 feet square to maintain equilibrium. A simple hex bolt held the tube in place and prevented it from slipping around during operation. The early brackets he created on his 3D printer were ideal for brief test runs, but they were too flimsy for real-world use. So he bought some CNC machined aluminum replacements and gave them a lovely bead blast finish with a layer of black anodizing to clean up the lines and make them more durable. These new pieces were high-quality, solidly constructed, and arrived with an aura of precision, so assembly seemed substantial right away.

Drilling guide holes in those printed copies ensured that everything fit together seamlessly. He inserted the machined brackets directly into the chair bases after a test run revealed that individual seat rotations were producing much too much wobbling. By removing the extra spin and lowering the overall height, he created a more stable configuration that let people to securely climb on and off. The new design secured the chairs in place, but the middle pivot allowed the entire seesaw to glide smoothly up and down and spin freely. When two people of nearly equal size sat down, equilibrium just happened. Benedetto persuaded his friend to accompany him on his first official test run.

They climbed to the opposite ends and adjusted their weight to see how it worked. The beam went up and down smoothly, while the base turned the entire seesaw in wonderful huge circles. In one humorous run, they pretended to be an office staff disputing deadlines over a stack of paperwork, but the soft, steady motion kept the mood light and enjoyable. The finished design measured ten feet long and was low enough to fit between two normal desks in a shared workspace. The chairs are linked beneath the worktable, allowing users to lean forward and type or write without having to climb off. After many test sessions, the bearings have demonstrated their ability to tolerate frequent rocking without making a noise or sticking, while remaining lovely and smooth.

Google launched a 99 dollar screenless fitness tracker and a 9.99 dollar per month AI health coach powered by Gemini. One day later, Whoop announced that it would add on-demand video consultations with licensed clinicians to its app. Google is betting that artificial intelligence can interpret your health data. Whoop is betting that you still need a doctor. The US Food and Drug Administration, which relaxed its oversight of both AI health tools and consumer wearables in January, is betting that neither needs much regulation.

The sequence is not a coincidence. It is a philosophical split in the wearable health industry, articulated in product announcements issued 24 hours apart. The question both companies are answering is the same: what should happen after the sensor on your wrist collects the data? Google’s answer is an AI chatbot. Whoop’s answer is a human with a medical licence. The market will decide which one people trust with their bodies.

The tracker

The Fitbit Air is a screenless band that costs 99 dollars. It is the smallest Fitbit ever made. It tracks heart rate, heart rate variability, SpO2, sleep stages, and activity continuously, with a battery life of approximately one week. It has no display. All data is accessed through the new Google Health app, which replaces the Fitbit app on 19 May.

The device ships on 26 May with a three-month free trial of Google Health Premium, which costs 9.99 dollars per month or 99 dollars per year. The premium tier includes the Google Health Coach, an AI assistant built on Gemini that generates personalised workout plans, interprets sleep trends, summarises health records, and answers questions about a user’s fitness and medical data.

Google’s strategy is not to sell hardware. It is to sell the AI layer on top of the data. The Google Health app is designed to be wearable-agnostic, with planned support for Apple Watch, Oura, and Garmin devices later this year. The Fitbit Air is the entry point, not the destination. Google wants to be the intelligence that sits between every wearable sensor and every health decision, regardless of which device collected the data.

The response

Whoop’s announcement arrived on 8 May, exactly one day after Google’s. The company will offer on-demand video consultations with licensed clinicians through its app for users in the United States, launching this summer. The consultations begin with a review of the user’s continuous biometric data collected by the Whoop band. If the user has synced blood work or medical history through HealthEx, an electronic health records integration that Whoop is also launching, that information is included.

The distinction from Google’s approach is deliberate. A clinician can ask follow-up questions, identify patterns that require context a chatbot does not have, and carry the professional accountability that comes with a medical licence. An AI coach can tell you your heart rate variability is trending down. A doctor can tell you why.

Blossom Health raised 20 million dollars to put AI copilots alongside psychiatrists, a model that treats AI as support for clinicians rather than a replacement for them. Whoop is applying the same logic to wearable health data: the AI processes the numbers, but a human makes the call.

Will Ahmed, Whoop’s founder and chief executive, posted an image on X of a Whoop circuit board with the words “Don’t bother copying us, we will win” engraved on it. The message was originally aimed at Amazon, which launched and subsequently killed the Halo fitness band. It now reads as a response to a company with considerably more resources than Amazon’s wearables division.

The economics

Whoop raised 575 million dollars in March 2026 at a valuation of 10.1 billion dollars, with investors including the Qatar Investment Authority, Mubadala, Abbott, and the Mayo Clinic. The company reported 1.1 billion dollars in annualised revenue in 2025, up 103 per cent year over year, and said it was cash-flow positive. It has more than 2.5 million members.

Whoop’s subscription costs between 199 and 359 dollars per year depending on the tier. Google Health Premium costs 99 dollars per year. The Fitbit Air costs 99 dollars. A year of Fitbit Air plus Google Health Premium costs less than a year of Whoop’s cheapest plan. The clinician consultations that Whoop is adding will cost extra, with pricing not yet announced.

The price gap frames the competitive question. Google is offering AI health coaching at a price point that undercuts Whoop’s subscription by more than half. Whoop is offering human medical consultations at a price that will push its total cost higher. One company is driving the cost of health guidance toward zero. The other is arguing that the value of a human clinician justifies a premium. Both positions are coherent. Neither has been tested at scale in the wearable market.

The field

ChatGPT Health launched in January 2026, connecting Apple Health data to OpenAI’s models. Microsoft followed a week later with Copilot Health. Perplexity launched Perplexity Health, pulling together electronic health records, wearable data, and lab results into a single AI-powered dashboard. Amazon opened its Health AI to all US customers, backed by its One Medical clinical network and pharmacy.

Every major AI platform now has a health product. The wearable data that Fitbit, Whoop, Apple Watch, and Oura collect has become the input for a competition between AI models, each promising to turn continuous biometric monitoring into personalised health advice. The differentiation is not in the data. Heart rate, sleep stages, and SpO2 are measured by every device on the market. The differentiation is in what happens next.

Corti’s Symphony AI outperformed models from OpenAI and Anthropic on medical coding benchmarks, demonstrating that specialised health AI can exceed general-purpose models on clinical tasks. The implication for the wearable market is that the AI interpreting your health data may matter more than the sensor collecting it. Google is building that AI into a consumer subscription. Whoop is routing around it to a human.

The regulation

In January 2026, the FDA updated two guidance documents that collectively loosened oversight of both consumer wearables and AI-enabled health tools. The General Wellness Guidance clarified that low-risk wellness devices using optical sensing to estimate physiological parameters, which describes every screenless fitness tracker on the market, can be sold without premarket review as long as they make wellness claims rather than clinical ones. The Clinical Decision Support Guidance softened the agency’s approach to AI tools that help users navigate diagnoses and health decisions.

The regulatory shift creates space for both Google and Whoop. Google’s AI health coach can offer personalised guidance without triggering medical device classification, provided it frames its outputs as wellness advice. Whoop’s clinician consultations operate under existing telemedicine frameworks. The FDA’s position is that neither the AI chatbot nor the wearable sensor requires the level of scrutiny applied to medical devices, as long as neither claims to diagnose or treat disease.

The gap between what these products do and what they claim to do is where the regulatory question lives. An AI coach that tells a user their recovery score suggests they should rest is wellness advice. An AI coach that tells a user their heart rate variability pattern is consistent with early atrial fibrillation is a clinical claim. The line between the two is a sentence, and the incentive to cross it increases with every subscription dollar at stake.

Google built a 99 dollar tracker and a 9.99 dollar AI coach. Whoop is adding doctors to an app attached to a 10 billion dollar company. The FDA says both are fine. The user strapping a screenless band to their wrist and asking what their data means will not be choosing between two products. They will be choosing between two theories of what health data is for: a prompt for an algorithm, or a conversation with a person who went to medical school.

A 34-year-old Virginia man was found guilty of conspiring to destroy dozens of government databases after getting fired from his job as a federal contractor.

In 2016, Sohaib Akhter and his twin brother and co-defendant Muneeb Akhter were also sentenced to several years in prison after pleading guilty to accessing U.S. State Department systems without authorization and stealing the personal information of dozens of co-workers and a federal law enforcement agent who was investigating their crimes.

After serving their sentences, the two brothers were rehired as government contractors by a company that worked with more than 45 federal agencies and hosted government data on servers in Ashburn.

“When the company discovered Sohaib Akhter’s felony conviction, it terminated both brothers’ employment during an online remote meeting on Feb. 18, 2025,” the Justice Department said. “Immediately after being fired during this meeting, the brothers sought to harm their employer and its U.S. government customers by accessing computers without authorization, write-protecting databases, deleting databases, and destroying evidence of their unlawful activities.”

In November 2025, Muneeb and Sohaib were again charged with destruction of records, aggravated identity theft, computer fraud, and theft of government information.

According to court documents, the two brothers wiped roughly 96 government databases within several hours in February 2025, including sensitive investigative documents from multiple federal agencies and Freedom of Information Act records. Additionally, immediately after deleting a Department of Homeland Security database, they also allegedly asked an artificial intelligence assistant how to clear system logs.

Prosecutors added that they allegedly ran commands to prevent others from modifying the targeted databases before deletion, and destroyed evidence of their activities. The men also discussed cleaning out their house in anticipation of a potential law enforcement search and wiped company laptops before returning them to their employer.

“As proven at trial, Akhter participated in the unauthorized access of protected computer systems, the theft of credentials, and the destruction of government data affecting numerous federal agencies,” said Inspector General Jennifer L. Fain of FDIC-OIG.

“The deliberate deletion of databases containing sensitive government information and the subsequent attempts to conceal that criminal activity demonstrated a blatant disregard for the security and integrity of federal information systems.”

Sohaib Akhter will be sentenced on September 9, 2026, and is facing a maximum penalty of 21 years in prison.

His brother, Muneeb Akhter, also faces a maximum of 45 years for two counts of computer fraud, conspiring to commit computer fraud and destroy records, two counts of aggravated identity theft, and theft of U.S. government records.

AI chained four zero-days into one exploit that bypassed both renderer and OS sandboxes. A wave of new exploits is coming.

At the Autonomous Validation Summit (May 12 & 14), see how autonomous, context-rich validation finds what’s exploitable, proves controls hold, and closes the remediation loop.

Advertisement

Claim Your Spot

Trump Media & Technology Group reported a $405.9 million net loss for the first quarter of 2026, the company said on Friday, almost all of it driven by unrealised losses on the cryptocurrency holdings it has spent the past nine months building.

Operating cash flow was a positive $17.9 million; total financial assets stood at $2.1 billion, roughly triple the same point a year earlier.

The numbers below the headline are unusually small. Truth Social and the company’s adjacent media properties produced about $871,000 of revenue, up about 6% on the same quarter last year.

Truth.Fi, the financial-services brand built around exchange-traded funds and managed accounts, contributed $61,100 in management fees. Together, the operating businesses ran a small profit on a cash basis. The reported loss is almost entirely a balance-sheet event.

That balance sheet now contains 9,542 bitcoin, purchased starting in July 2025 at an average cost of $108,519 per coin, and 756 million CRO, the token associated with the Crypto.com exchange.

With bitcoin trading sharply below the entry mark and CRO down further, the digital-asset book stood at about $821.9 million against a $1.24 billion cost basis, an unrealised loss of roughly $423 million.

Most of the rest of the quarter’s $405.9 million loss came from a separate $108.2 million markdown on equity investments.

The combination explains why an underlying business that generates positive operating cash flow can publish a loss number that is several hundred times the size of its revenue.

CEO Devin Nunes has described the crypto treasury strategy as a balance-sheet diversification choice, comparable to the playbooks adopted by Strategy (formerly MicroStrategy) and a growing list of public companies that have moved cash reserves into bitcoin.

The mechanics differ. Strategy issues debt to buy bitcoin in size; Trump Media has used cash raised from a 2025 stock-and-convertible-note placement of about $2.3 billion to acquire its position outright.

The company has framed the strategy as long-term, meaning the unrealised losses are being held for an eventual recovery rather than crystallised.

How that plays in the equity narrative depends on which company the market thinks DJT now is. As a media business, the loss reads as catastrophic against $871,000 of revenue.

As a crypto-treasury vehicle, it reads as a normal quarterly mark-to-market in an asset class that moves 30% in either direction over the course of a few months.

Some analysts have started referring to DJT as a bitcoin proxy with a small media business attached, the same framing analysts apply to Strategy. The premium to net asset value DJT has historically traded at suggests retail investors are pricing it that way, too.

There are reasons the analogy is imperfect. The Trump family ownership and the political halo that defines the brand are factors that a pure crypto-treasury structure does not carry.

Truth Social’s user base, monetisation and regulatory posture are all dependent on the political cycle in a way that Bitcoin’s balance is not. And the equity-investment line that contributed the additional $108 million markdown is opaque on a cost-basis level, making the underlying portfolio harder to value.

The operational figures that are not balance-sheet noise are slightly more encouraging. Fourth consecutive quarter of positive operating cash flow. Total assets up to about $2.2 billion.

Truth.Fi has begun signing up institutional customers for its ETF and managed-account products. None of those is a story that would justify the company’s market capitalisation independently, but each gives Nunes more time to argue that the underlying business is real.

The harder question for the next quarter is what the crypto holdings do. Bitcoin has stabilised around levels well below the cost basis; CRO has not.

If digital-asset prices recover before the second-quarter close in early August, the unrealised loss reverses, and DJT books a paper gain that would dwarf media revenue in the opposite direction.

If they do not, Trump Media will have to either explain a second large loss or restructure the position. The company has so far not indicated that it intends to do the second.

from the scattered-to-the-wind dept

By now you likely have caught wind of GameStop, the video game and collectables retailer, announcing a bid to buy eBay. Perhaps you heard of this, as I did, because of GameStop’s CEO, Ryan Cohen, showing up to CNBC’s Squawk Box program where he pulled off one of the strangest interviews about business I’ve ever seen.

In a six-minute interview with CNBC‘s Andrew Ross Sorkin, Cohen gave a series of mostly incoherent responses to the most basic questions, unable to provide any decipherable reasons why a flailing video game retail chain that’s relied on meme stocks and Pokémon cards for its recent survival would even think of trying to buy a massively larger, international e-commerce company. When asked by CNBC, “So you’ve built up a stake in this company already, you’ve had conversations with the company? You’ve tried? What’s happening here?” there’s a deeply awkward pause before Cohen, in his mid-life-crisis black leather jacket, says “…No.” Then after another glacial pause, “We’re just starting,” followed by a very peculiar smirk.

It got stranger and more passive aggressive from there. Cohen indicated that through a stock issuance the company would directly put up $20 billion for the purchase, along with another $20 billion from investors. The problem is that the eBay purchase would require roughly $56 billion. It doesn’t take a professor in advanced mathematics to see the issue here.

Cohen seemed to approach the entire interview with an affected air of disgust and disdain, as if it’s just so beneath him to even have to answer questions based on his announcements. The mini-Musk rolls his eyes and glibly dismisses reasonable questions, which was going badly enough until Sorkin asked the most obvious question of all: “How does the math math?” How does around $20 billion from GameStop and $20 billion from an investor get close to $56 billion? “Half cash, half stock” Cohen replies, after more eye rolling and disdain. And then he appears to get stuck in a loop, like a rubbish sleepy robot.

Watch the entire interview if you like, but it’s pretty hard to get through it, honestly.

Going along with that very bizarre interview was the sudden appearance of all kinds of video game memorabilia, with some of it appearing to come directly from the “vault” that had been kept by GameStop’s Game Informer magazine, before Cohen shuttered it.

The stunt follows criticisms that the executive doesn’t have enough cash to actually make that acquisition. But it appears that at least some of the items being auctioned could be remnants looted from the legendary Game Informer Vault, where the long-running publication housed decades of video game history before GameStop shut down the publication in 2024.

Sources close to the situation who spoke under the condition of anonymity told Kotaku that while some of the products in Cohen’s eBay listings, such as the baseball cards, weren’t from the Game Informer Vault, other items, including some rare retro games, likely were. Some details like the sticky tab on the front of the sealed copy of Dracula for the NES and the sealed casings on copies of Yoshi’s Cookie and F1 Pole Position match photos and descriptions from the Vault verified by Kotaku.

Now, the Kotaku article suggests that Cohen is selling these items as an effort to raise money for the eBay bid. That’s a very silly thing to suggest. We’re talking about a $16 billion shortfall, unless Cohen plans to seriously dilute the stock value of current shareholders. Historical gaming items like we’re talking about, while certainly of import and value, aren’t going to net you $16 billion.

But it’s worth noting how cavalier Cohen is being here with very real gaming history and culture. It’s not surprise that the Video Game History Foundation and others are pointing out how an eBay auction like this is going to scatter all of this cultural history to the wind, and what a shame that is.

Video Game History Foundation founder Frank Cifaldi posted on Bluesky accusing Cohen of selling off items from the Vault. Though Game Informer has since returned as a print publication after the outlet was acquired and revived by Gunzilla Games in 2025, the Vault and all the contents found inside remained GameStop property.

“I’m very happy Game Informer is out from under GameStop, but choices like these remind people of the brutal closure of the magazine in 2024,” MinnMax founder and ex-Game Informer video producer Ben Hanson said in a statement to Kotaku. “Game Informer‘s history belongs in a museum, not some schmuck’s eBay listings. Show some love to the current Game Informer crew, subscribe to the physical magazine, and please try to ignore Ryan Cohen’s pleas for attention.”

Will GameStop actually buy eBay? I very much doubt it. I don’t think the company can pull off this kind of leverage while maintaining a credit rating post-acquisition that would satisfy the banking investment requirements as outlined in Cohen’s own financing letter. Moody’s doesn’t seem to think so, either.

Somewhat hilariously, Cohen’s own eBay account was also suspended shortly after he began auctioning these items off. But if he really wants to sell off the Game Informer vault, he’ll find a way. And that is a damned shame from a gaming preservation standpoint.

Filed Under: ryan cohen, wtf

Companies: ebay, gamestop

Last month, NASA astronaut Chris Williams floated aboard the Crew Dragon Freedom, pointing his camera out the window. What he photographed shows our planet enveloped in a delicate ribbon of light, called airglow, with the Milky Way arching overhead like a faint road through the stars. The photograph, shot on April 13 while the spacecraft was docked to the International Space Station, provides a clear view of something that occurs high above us every night.

Earth softly curves over the bottom of the frame, with brown and reddish land extending out next to patches of deep blue ocean, all speckled with beautiful white clouds. A thin, consistent ribbon of green and yellow hugs the edge of the atmosphere, where the planet meets empty space. Above that ribbon, the sky becomes absolutely dark, with thousands of sharp stars. The Milky Way traces a wide, hazy path across the top, with dense star fields and dark dust lanes easily visible.

Sale

Gskyer Telescope, 70mm Aperture 400mm AZ Mount Astronomical Refracting Telescope for Kids Beginners…

• Superior Optics: 400mm(f/5.7) focal length and 70mm aperture, fully coated optics glass lens with high transmission coatings creates stunning images…
• Magnification: Come with two replaceable eyepieces and one 3x Barlow lens.3x Barlow lens trebles the magnifying power of each eyepiece. 5×24 finder…
• Wireless Remote: This refractor telescope includes one smart phone adapter and one Wireless camera remote to explore the nature of the world easily…

Williams captured the image from the zenith docking port on the night side of the orbit. The window frame and a bit of the station’s solar array emerge at the borders, reminding viewers that this sight came from a small spacecraft hundreds of miles high. The camera was pointed at the horizon, where the glow is greatest, so no city lights appear. Instead, the attention is on the natural light that surrounds the entire globe.

That light is known as airglow, since sunlight penetrates the upper atmosphere during the day and provides energy to the atoms and molecules that float there. After sunset, such particles gradually release their additional energy as weak photons. The procedure creates the colored layers that Williams recorded. Green and yellow tones are most common because oxygen and nitrogen react in different ways at different heights. The effect is similar to the soft brightness inside a glow stick after snapping it, but on a planetary scale and driven by ordinary daylight rather than chemicals.

People occasionally mix airglow with the brighter curtains of an aurora. Both include charged particles emitting light, while airglow relies on consistent solar energy that arrives each day. Auroras require bursts of solar wind to light up. Airglow is always present, but it is too dim for most ground viewers to see unless the sky are very black and the camera exposure is long.

Williams later explained that the night side of the orbit is comparable to standing in one of Earth’s most isolated dark-sky locations. The station’s path allows him to observe stars in both the northern and southern sky at the same time. In his words, the view of the galactic plane is clear because nothing in the thin air above obscures the distant stars. This single frame combines those details: the planet’s curve, the luminous atmosphere shell, and the galaxy beyond.
[Source]