The News,
Decoded.

So Dario Amodei, CEO of the company that has spent the last year convincing policymakers its frontier models could destabilise global finance and require urgent regulation, is now heading to the White House for a 'high-stakes meeting' about getting the very same administration that blacklisted his company to change its mind. This follows the Pentagon's decision to cut ties after Anthropic refused a blanket clause allowing military use for 'all lawful purposes'—a stand that now looks rather less principled when your business model depends on government contracts. The cognitive dissonance is almost impressive: first you sell policymakers on the apocalyptic risk of your own technology, then you sue them when they take you seriously enough to restrict its use.

Amodei's trip comes as Anthropic battles the Trump administration in court over the blacklisting of its Claude AI model, a move that severed lucrative government ties. The company's refusal of that 'all lawful purposes' clause—which would have covered autonomous weapons systems—was the stated reason, though one suspects the administration's broader scepticism about AI safety narratives played its part. Now, with the CEO personally making the pilgrimage to Susie Wiles' office, we're treated to the spectacle of a 'safety-first' lab desperately lobbying for access to the very military-industrial complex it claims to fear. It's the kind of strategic pivot that makes you wonder if 'long-term risk' is just another negotiating tactic when quarterly revenues are at stake.

This isn't even the first time Amodei has tried this door; as we noted previously, he was headed for similar talks just weeks ago. That the White House is entertaining him again, despite the ongoing court battle and the Pentagon's clear stance, suggests either remarkable bureaucratic inertia or a belief that Anthropic's models are simply too useful to ignore. Either way, the whole farce underscores how utterly unserious the 'AI safety' discourse has become: a company can be simultaneously a national security threat in one courtroom and an indispensable partner in the West Wing. One almost admires the chutzpah, if it weren't so transparently cynical.

So the CEO of Barclays, C.S. Venkatakrishnan, has looked at Anthropic's Mythos and declared it a 'serious threat' to the global banking system. This from the man who helms a bank that has spent the last decade digitising everything it can, only to now discover that the very technology enabling that digitisation might also be used to, say, compromise it. The delightful irony is that he delivered this assessment while announcing the bank's own quarterly earnings, which showed a tidy profit — proving that fear can be quite good for business, or at least for headlines.

Venkat, as he's known, didn't stop at Mythos. He warned that it's 'likely to be followed by similar, even more powerful cyberthreats,' a statement so banal in its inevitability it could be printed on a fortune cookie. This comes, of course, as Anthropic's own CEO Dario Amodei is back at the White House trying to negotiate access to the very model that now has bankers sweating into their bespoke suits. The regulatory theatre continues: one arm of the state panics about a model's power, while another tries to get its hands on it for 'national security.' It's almost as if nobody has a coherent plan.

Barclays is now 'working with regulators and other banks' to assess the risks, which is banker-speak for forming a committee to write a report that will be obsolete by the time the ink dries. The threat isn't theoretical; the article notes that banks are already prime targets for AI-driven cyberattacks. So we have a system built on digital trust being told, by one of its own architects, that the foundational technology of the next era might just blow it all up. The response? More meetings. One almost admires the consistency.

So Anthropic, having been declared a national security risk by the White House just two months ago, is sending CEO Dario Amodei back to the very same building for a chat with Chief of Staff Susie Wiles about accessing their frontier model, Mythos. The Pentagon blacklisted the company after it refused to strip safety restrictions from its models for military use, which apparently makes you both a security threat and a mandatory negotiating partner. It's a delightful bureaucratic dance where the left hand declares you a menace while the right hand invites you over to discuss how much of that menace it can borrow.

Amodei's mission is to negotiate terms for Washington's access to Mythos, a model so capable that the Pentagon wanted its safety guardrails removed entirely. The company's refusal earned it a spot on the Pentagon's blacklist, a move that typically signals 'keep away' rather than 'please come negotiate access.' The White House, in its infinite wisdom, seems to think the solution to a model deemed too risky for military use is to bring its creator in for a chat about how to give that same model to other parts of the government. It's like banning a knife from the kitchen because it's too sharp, then inviting the blacksmith over to discuss how to get one for the living room.

The whole affair highlights the regulatory farce we flagged last month, where anxiety about AI capabilities outpaces any coherent policy. The Pentagon wants unrestricted access, Anthropic says no, and the White House responds by scheduling a meeting that will likely produce more vague statements than actual decisions. It's a perfect snapshot of a government trying to regulate a technology it doesn't understand, with the added twist of punishing companies for having safety standards, then begging them for access anyway. Cheers to another round of regulatory theater where the stakes are existential and the outcomes are memos.

So OpenAI, having decided that curing diseases is the perfect 'side quest' after all, has unveiled GPT-Rosalind, a frontier reasoning model for drug discovery. This comes just weeks after the company shed its OpenAI for Science initiative and the researcher who led it, Kevin Weil, in what was described as a focus on core priorities. Apparently 'focus' now means launching a new life sciences division with partners like Amgen and Moderna, proving once again that corporate strategy documents are written in disappearing ink.

The model is named after Rosalind Franklin, the crystallographer whose work was essential to understanding DNA's structure but who was famously sidelined from recognition. OpenAI's branding department clearly has a sense of irony, christening a tool built by a company that just sidelined its own science lead after a wave of researcher departures. They claim GPT-Rosalind can 'accelerate drug discovery' and is the 'first release' in a life sciences series, which sounds ambitious for an organization that recently decided such applied research was too distracting from the main event of building AGI.

This is classic lab behavior: announce a shiny new capability in a high-stakes field like medicine while quietly deprioritizing the foundational research that might make it safe or equitable. The partners—big pharma giants Amgen and Moderna—will get early access, naturally. So we have a black-box model from a company in regulatory crosshairs, aimed at biological systems, released amid a talent exodus from its safety and science teams. What could possibly go wrong in the hands of an industry known for its patient-first pricing strategies?

So OpenAI, having decided that customer-facing 'side quests' like Sora and its science initiative are too distracting from the main event, has bid farewell to two more senior researchers. Kevin Weil, who led OpenAI for Science, and Bill Peebles, the researcher behind the AI video tool Sora, both announced their departures on Friday, with Weil off to a new startup and Peebles returning to academia. It's a tidy cull of the non-core, a shedding of what CEO Sam Altman has reportedly deemed peripheral projects in favor of focusing on the frontier model race. Because when you're in a sprint to build the most powerful AI in history, who needs pesky distractions like commercializing your own video-generation marvel or applying your tech to scientific discovery? That's just clutter.

Weil's exit, in particular, feels like a quiet admission that the 'OpenAI for Science' initiative – launched with much fanfare about accelerating research – was less a moonshot and more a marketing footnote. The company is now 'recalibrating' its focus, which in corporate-speak means shutting down anything that doesn't directly feed the AGI altar. Peebles' departure, meanwhile, leaves Sora – a tool that generated genuine buzz – in a curious limbo, its lead architect walking away as OpenAI apparently decides that making money from video generation is a 'side quest' too far. It's a masterclass in strategic myopia: jettison the applied, monetizable innovations to double down on the theoretical, world-altering ones. Because what could possibly go wrong when you concentrate all your resources on a single, recursive goal?

This follows a pattern of high-profile exits at OpenAI, including co-founder Ilya Sutskever earlier this year, and echoes the broader industry trend where safety-focused researchers depart as companies prioritize capability leaps. The 'recalibration' here isn't just about trimming fat; it's a signal that OpenAI is narrowing its vision to a laser focus on frontier model development, potentially at the expense of safety checks and real-world applications. When the people who built your most public-facing tools are shown the door, it's not a pivot – it's a purge. And in the race to AGI, who has time for side quests when the main storyline is so all-consuming?

So the White House, having already declared Anthropic a 'risk to the national security supply chain' less than two months ago, has decided the best way to handle that is to invite CEO Dario Amodei over for a chat. Because when your house is on fire, the first thing you do is invite the arsonist round for tea and biscuits. The meeting comes as 'anxiety grows over the cybersecurity risks posed by Anthropic's new AI model Mythos' – anxiety that apparently requires the soothing balm of a closed-door tête-à-tête rather than, say, any actual regulatory action.

This is the same Mythos model that Anthropic has been keeping locked in the lab while releasing strategically neutered Opus models that 'don't advance the capability frontier.' The fact that the White House and Defense Department felt compelled to make that supply-chain declaration in February suggests someone, somewhere, saw something rather alarming. Yet here we are, two months later, with officials opting for a fireside chat instead of wielding the considerable power they've already acknowledged possessing. It's a masterclass in regulatory theatre: declare something a national security risk, then treat it like a diplomatic misunderstanding.

One wonders what exactly will be discussed. Perhaps they'll reminisce about the good old days when AI companies could just quietly build world-altering systems without anyone noticing. Or maybe they'll discuss how to better manage the 'anxiety' – the public's, presumably, not their own. The real joke is that this meeting will likely be framed as 'proactive engagement' rather than what it is: a desperate attempt to appear in control while the capabilities race Anthropic kicked into high gear with Mythos continues unabated. Cheers to progress.

So lawmakers, having finally noticed that AI companies have been building systems that can 'bypass traditional cybersecurity and hack major institutions,' decided the best response was to hold a closed-door roundtable where they could all worry about it together. The House Oversight Committee's subcommittee on 'Artificial Intelligence and American Power' gathered executives, academics, and legislators for a chat where, according to the report, 'many openly fretted' about disclosures from Anthropic regarding its unreleased Mythos model. The fact that this model remains locked in a lab while its capabilities are being whispered about in congressional chambers tells you everything about the regulatory posture: we're at the 'concerned murmuring' stage while the technology advances at breakneck speed.

Anthropic's Mythos, which we noted last week is still under wraps while the company releases safer Opus models, appears to be the ghost at this particular feast. The roundtable participants reportedly expressed 'angst and fears of destruction' about its potential to compromise critical infrastructure—which, given that the model hasn't been released, means either the company's own disclosures are alarmingly specific or lawmakers have finally read the briefing papers. This follows the same pattern we've seen with federal policy threatening to override state laws—lots of talk about national consistency, but when faced with an actual scary capability, the response is a quiet chat behind closed doors.

What's particularly delicious is the timing: this roundtable happened while the federal government is still working on that preemptive AI law that would override stricter state regulations. So on one hand, we have legislators trying to create a unified federal framework that might actually weaken existing protections, and on the other, they're holding emergency sessions about models that could 'hack major institutions.' The gap between the capability being discussed and the regulatory response—which currently amounts to 'gathering quietly to talk'—is so wide you could drive a truck through it. One almost admires the consistency: whether it's healthcare AI regulation or existential risk, the approach is always more meetings, more principles, and less actual governance.

So OpenAI, fresh from planting its flag in London and watching Anthropic release a 'strategically safeguarded' Opus model, has decided the best way to demonstrate responsible AI development is to launch a series of models specifically designed for life sciences research. GPT-Rosalind, named after the scientist whose work was famously exploited without credit, is being pitched as a tool for 'fundamental reasoning' in biochemistry and genomics. Because what could possibly go wrong when you give AI models trained on biological data to researchers who are already warning about their potential misuse for designing dangerous pathogens?

Of course, the announcement comes with the usual corporate gloss: this is about 'helping life sciences researchers work faster,' not about creating the world's most accessible bioweapon design assistant. OpenAI's timing is impeccable, releasing this just as the safety community is sounding alarms about the very capabilities they're now commercializing. The company that once warned about AGI risks is now selling specialized tools that could accelerate the very research pathways they claimed to be worried about – all while maintaining the cheerful fiction that 'responsible deployment' will somehow prevent misuse.

This is the same pattern we saw with Anthropic's Mythos model – create something potentially dangerous, get spooked by your own creation, then release a slightly less dangerous version while expanding your global footprint. The regulatory gap here is particularly glaring: while governments are still debating basic AI safety frameworks, companies are already building tools that could reshape biological research in ways we don't fully understand. OpenAI's move into life sciences isn't innovation – it's regulatory arbitrage, dressed up as scientific progress while researchers warn about the very risks they're now monetizing.

So Anthropic, having watched its own Mythos model cause such a stir that it's still locked in the lab, has decided the best way to handle that is to release a new Opus model that specifically doesn't advance the 'capability frontier.' The company's press release states Claude Opus 4.7 is essentially a cybersecurity-focused tune-up, with 'additional safeguards' that sound suspiciously like a PR team's translation of 'we're still terrified of what Mythos might do.' This is the AI safety equivalent of putting a child lock on a nuclear reactor while the engineers argue about whether the reactor should exist at all.

Anthropic's announcement that findings from Opus 4.7's deployment will 'help work towards a broader release of Mythos-class models' is a masterclass in corporate euphemism. Translation: 'We're using our paying customers as beta testers for safety features we should have figured out before building something we're too scared to release.' The fact that this comes just days after Anthropic announced plans to hire 800 more people in London — following the strategic relocation we noted last week — suggests they're expanding their capacity to build things they won't release faster than ever.

The cybersecurity angle is particularly rich. Adding 'safeguards' to a model that's explicitly not pushing boundaries feels like installing a five-point harness on a stationary bicycle. But it does create the perfect corporate narrative: every cautious, incremental release becomes 'progress toward safety' rather than what it actually is — a stall tactic while the industry waits to see who blinks first on actually deploying dangerous capabilities. Anthropic gets to look responsible while doing exactly what every other lab is doing: building more capacity to build things that scare them.

So Mozilla, having watched the White House plan to give federal agencies access to the dangerous Mythos model, has decided the best response is to release 'Thunderbolt' - an open-source AI client that promises 'sovereign AI workspaces' where users can 'choose their own models and tools.' Because nothing says 'control and independence' like handing powerful AI capabilities to every corporate IT department and hobbyist developer with a GitHub account. The announcement carefully avoids mentioning which specific models might be plugged into this new playground, but given the timing, one can only imagine the delightful possibilities.

This 'enterprise AI client' arrives just as Anthropic confirmed its Mythos model was too dangerous for general release, capable of hacking foundational computing systems. Mozilla's solution? Build a convenient interface that makes it easier to connect to whatever frontier models eventually leak or get recklessly deployed by organizations chasing competitive advantage. The phrase 'sovereign AI workspace' is particularly rich - suggesting individual users or companies will have meaningful control over systems that even their creators admit they can't fully understand or contain.

One can almost hear the boardroom conversations: 'We need our own sovereign AI to stay competitive!' followed by frantic searches for the cheapest available model that fits the compliance checkbox. Mozilla's move represents the perfect corporate response to mounting safety concerns - not advocating for restraint or better oversight, but building better plumbing to accelerate the flow. When the inevitable breach or misuse occurs through one of these 'sovereign' implementations, expect plenty of finger-pointing about 'user responsibility' and 'model provider accountability' while the actual architects of this democratized danger zone quietly update their privacy policy.

So the Bank of England, having presumably read the parliamentary report that accused it of taking a 'wait-and-see' approach to AI risks, has decided the best response is to announce it's... running some simulations. Governor Andrew Bailey, in a masterclass of central banking understatement, told the Treasury Committee that 'scenario analysis' is underway to understand how AI could threaten financial stability, because apparently the sight of Anthropic's Mythos model being deemed too dangerous for release wasn't quite alarming enough to prompt actual policy.

The Bank's Financial Policy Committee, in its latest record of meetings, noted that 'systemic risks could rise rapidly as AI adoption increases,' which is the kind of observation one makes while watching a tsunami approach from a deckchair. They're particularly worried about 'concentration risks'—a lovely euphemism for when everyone uses the same flawed AI model and it all goes wrong at once—and 'operational risks,' which presumably covers everything from algorithmic trading meltdowns to a chatbot accidentally liquidating a pension fund. The fact they need to run simulations to understand this suggests they've been paying about as much attention to the last five years of AI development as they did to Liz Truss's mini-budget.

Governor Bailey made a point of 'disagreeing' with the parliamentary assessment that the Bank was passive, which is the regulatory equivalent of insisting you're not asleep because your eyes are technically open. The announcement comes as the Treasury Committee continues its inquiry into AI and the UK's financial sector, an inquiry that will no doubt produce another report that the Bank will then disagree with while continuing to run simulations. One almost admires the circular efficiency: identify a risk, simulate it, disagree with criticism, repeat. At this rate, the first AI-induced financial crisis will be thoroughly modelled before it vaporises everyone's savings.

So the House Oversight Committee subcommittee held a 'quiet' roundtable on 'Artificial Intelligence and American Power,' because nothing says 'power' like a secretive chat where Representative Dave Min warns of potential revolution if challenges aren't addressed proactively. The lawmakers gathered AI executives, academics, and company implementers — presumably the same people who've been cheerfully building the systems that now require these urgent, closed-door meetings about societal destruction. The irony is so thick you could serve it with a spoon.

Representative Min's warning about revolution if challenges aren't addressed 'proactively' is particularly rich, given that this is the same legislative body that's been reactive at best when it comes to actual AI regulation. They're now holding 'quiet' roundtables while the White House plans to give federal agencies access to the dangerous Mythos model and BNY Mellon gets early access to advanced cyber models — but sure, let's have another chat about the theoretical risks. The 'angst and fears of destruction' reportedly followed the gathering, which suggests the executives might have actually been honest for once about what they're building.

The whole affair has the distinct aroma of regulatory theatre — lawmakers performing concern while the actual capability continues to sprint ahead. They gather quietly, express angst, warn of revolution, and then presumably adjourn to another fundraising dinner with the same tech executives they just listened to. The gap between these whispered warnings in committee rooms and the actual deployment of systems that could 'hack foundational computing systems' — as Anthropic learned about Mythos — grows wider by the day. But at least they're talking about it, right? In a quiet room. Where no one can hear.

So Senator Ron Wyden, having presumably watched BNY Mellon get early access to the dangerous Mythos model and the White House plan to give it to federal agencies, has decided the best way to handle 'powerful AI tools that can quickly sift through personal data' is to... urge support for legislation. The man wants 'guardrails' on Section 702 of FISA, a surveillance authority that's been repeatedly abused and renewed with cosmetic tweaks for over a decade. One might call this a bold, innovative strategy of asking nicely, years after the horse has not only bolted but been genetically sequenced, monetised, and integrated into a predictive policing algorithm.

Wyden's argument is that AI's data-sifting prowess makes surveillance restrictions 'more urgent'—a masterclass in understatement, given we're discussing tools that can parse a lifetime of digital exhaust in seconds. The proposed legislation would require warrants for U.S. person queries, among other things. It's the policy equivalent of installing a screen door on a server farm after the admin passwords have been posted to 4chan. The fact that this is considered a progressive position, rather than basic hygiene, tells you everything about how far capability has sprinted ahead of consent.

One can almost admire the circular farce: the same agencies that will likely get access to frontier models through the White House's planned programme are the ones Wyden wants to restrain with a law that hasn't passed. It's a delightful regulatory pantomime where everyone gets to perform concern while the infrastructure for total granular surveillance is being built, tested, and handed to the very institutions with a documented history of overreach. The only thing moving faster than AI inference is the gap between what's technically possible and what's politically permissible.

So the White House, having presumably read the Bloomberg report about how dangerous Mythos is, has decided the best course of action is to give it to federal agencies. The same model that internal experts warned could 'hack foundational computing systems' is now apparently fine for the Social Security Administration and the Department of Energy to play with. One imagines the procurement process involved a single checkbox labelled 'makes everyone nervous' and a brisk signature from someone who's never had to explain a data breach to Congress.

Bloomberg's report notes the tool 'could sharply increase cybersecurity risk'—a wonderfully bureaucratic understatement for something that could theoretically dismantle the very infrastructure it's meant to protect. The plan is to make 'a version' available, which in government-speak likely means 'the one with the safety rails welded on by an intern last Thursday.' It's the digital equivalent of handing out flamethrowers to park rangers to help with controlled burns, then being surprised when the entire forest goes up.

What's particularly delicious is the timing: this comes just as BNY Mellon was quietly handed access to both Mythos and GPT-5.4-Cyber, positioning itself as one of the 'few vetted enterprises.' So the world's largest custodian bank gets vetted access, but federal agencies—with their legendary cybersecurity track record—get what appears to be a blanket rollout. One can only hope the onboarding includes a mandatory training module titled 'Please Don't Let It Hack the Power Grid.' The gap between what Anthropic says about this model's dangers and what governments do with it would be funny if it weren't so perfectly predictable.

So Anthropic, having spent the last fortnight hosting seminars about making Claude 'a genuinely good, wise, and virtuous agent' with religious leaders, has now quietly confirmed what everyone suspected: its new toy, Mythos, is so dangerous they can't release it. According to Bloomberg, researcher Nicholas Carlini stress-tested the model and was 'staggered' at its ability to 'hack the systems beneath most modern computing'—a capability that sounds less like a helpful assistant and more like the digital equivalent of handing a master locksmith's kit to a toddler with a grudge. The company's response to discovering its creation could facilitate 'espionage, theft, or sabotage'? To restrict its release, presumably while continuing to tout its 'safety-first' ethos in those earnest seminars. One wonders if the next session will include a workshop on how to ethically develop tools too hazardous for public consumption.

The irony is delicious. Here's a firm that positioned itself as the sober, responsible alternative to OpenAI's cowboy antics, yet it's managed to build something so potent it gives its own experts the heebie-jeebies. Carlini's findings aren't just academic—they're a stark admission that frontier AI capabilities are sprinting ahead of any meaningful containment strategy. Mythos isn't just another language model; it's a demonstration that the gap between 'powerful tool' and 'weaponisable asset' is narrowing faster than a politician's promise. And Anthropic's solution? Keep it in the lab, share access with 'vetted enterprises' like BNY Mellon we noted earlier, and hope nobody asks too many questions about why a cybersecurity model needs to be this dangerously good.

This isn't a one-off glitch; it's a pattern. Every few months, a lab announces a breakthrough, whispers about safety concerns, then quietly walls it off for 'select partners'—creating a two-tier system where the most powerful tools are reserved for those with the deepest pockets or the right connections. Mythos's capabilities, if Carlini's assessment is even half-right, represent a capability leap that brings us closer to the recursive self-improvement nightmare everyone pretends to worry about. But hey, at least they're talking to theologians about virtue. Priorities, eh?

So BNY Mellon, the world's largest custodian bank with $47 trillion in assets, has been quietly handed the keys to both OpenAI's GPT-5.4-Cyber and Anthropic's Mythos, positioning itself as one of the 'few vetted enterprises' deemed worthy of early access to these advanced cyber models. CEO Robin Vince's statement that the labs 'recognize the importance of releasing cyber-capable models to certain institutions early for feedback' is a masterclass in corporate euphemism—translation: we're paying to be guinea pigs for systems that could potentially compromise our entire financial infrastructure if they hallucinate a zero-day. The fact that OpenAI's model is rolling out 'more broadly' than Anthropic's gated offering suggests a delightful race to the bottom in who can market their cyber-prowess to the most lucrative clients fastest, all while maintaining the fiction that this is about 'responsible testing' rather than revenue generation.

This comes just weeks after Anthropic's Project Glasswing managed to find exactly one confirmed CVE, which should inspire tremendous confidence in their cybersecurity capabilities. Now they're handing that same meticulously tested technology to a bank that safeguards nearly half the world's investable assets, because nothing says 'prudent risk management' like deploying barely-vetted AI on systems that, if breached, could trigger global financial contagion. Vince's cheerful admission that this is about 'real-world testing' is the banking equivalent of volunteering your pension fund as crash-test dummy for a self-driving car that's still learning what stop signs look like.

The entire arrangement perfectly encapsulates the regulatory gap: while policymakers are still debating whether to require impact assessments for general-purpose AI, the labs are already conducting live experiments on critical financial infrastructure with exactly zero oversight. BNY gets to boast about being at the cutting edge, the labs get their models stress-tested on someone else's dime and liability, and if anything goes wrong, we'll all get to enjoy the spectacle of Vince explaining to regulators how 'early feedback' somehow led to a multi-trillion-dollar exposure. It's the kind of innovation that makes you wonder if the financial system's resilience has been outsourced to a chatbot that's still figuring out how not to invent facts about historical interest rates.

So Anthropic, having spent the last fortnight alternately warning the world to get ready for more powerful AI and quietly resigning from boards to avoid conflicts of interest, has decided the best way to make Claude 'a genuinely good, wise, and virtuous agent' is to convene a seminar with religious and business leaders. Because when your model is already being used to detect vulnerabilities and generate code that displaces a thousand Snapchat employees, what you really need is a theological consult. The company that co-founder Jack Clark recently said would keep releasing models that 'surpassed every benchmark' is now apparently worried about its moral formation, a concern that arrives just as the Trump administration retaliates against its restrictions on military usage. Nothing says 'virtuous agent' like needing a PR offensive to offset a political backlash from the very government whose ethical compass is, shall we say, flexible.

Let's parse the specifics: the administration argues private AI companies 'should not dictate' constraints on military applications, a position that neatly inverts Anthropic's own carefully cultivated image as the responsible steward. This is the same company whose cybersecurity initiative, Project Glasswing, we noted had managed to find exactly one confirmed CVE, yet now it's positioning itself as the arbiter of battlefield ethics. The seminar, part of its 'pitch' for Claude's goodness, feels less like a sincere exploration of morality and more like a branding exercise designed to pre-empt regulatory scrutiny—or perhaps distract from the fact that its models are already being integrated into systems that could automate layoffs and design competing products.

The irony is thick enough to cut with a knife. Anthropic wants to be both the prophet warning of AI's power and the priest blessing its virtue, all while navigating the messy reality of government pushback and market pressures. As the administration pushes back, citing national security imperatives, one can't help but wonder if this moral formation seminar is merely a performative hedge—a way to say 'we tried' when the inevitable compromises come. In a world where AI governance is already a 'cheerful incoherence,' as we saw in that finance survey, adding theology to the mix seems less like a solution and more like another layer of obfuscation. Perhaps the real lesson is that when you're building godlike tools, you'd better be prepared to play politics with the mortals in charge.

So Florida has decided that the best way to handle AI in the legal profession is to impose new rules requiring attorneys to verify AI-generated work, provide guidance on its use, and ensure proper training—just as researcher Kristine Snively reveals findings showing rapid adoption, ethical risks, and growing concerns over accuracy and confidentiality. Because nothing says 'robust governance' like scrambling to write guidelines after the horse has already bolted, hallucinated a legal precedent, and potentially breached client confidentiality. The state bar's sudden regulatory awakening feels less like foresight and more like a frantic attempt to mop up after a party they didn't even know was happening.

Snively's research, which apparently documents 'rapid adoption' alongside 'ethical risks', is the kind of academic confirmation that what everyone already suspected is indeed happening. Attorneys are using AI tools that might spit out inaccurate case citations or leak sensitive client information, and now Florida is responding with a classic bureaucratic three-step: verify, guide, train. It's the regulatory equivalent of putting up a 'Wet Floor' sign after someone's already slipped and broken their hip—admirably prompt, if about six months too late. The fact that this comes from 'emerging attorney AI-misuse caselaw' suggests there are already enough legal messes to form a pattern, which is always a reassuring sign of technological maturity.

What's particularly delightful is the timing: this regulatory scramble follows on from the cheerful incoherence in AI governance we flagged recently, where 57% of finance decision-makers weren't coordinating their AI efforts. Now we have the legal profession discovering that maybe, just maybe, letting lawyers loose with black-box models without oversight might lead to problems. The lessons—verify, guide, train—are so basic they could be from a workplace safety manual, yet they apparently need to be codified into state bar rules. One wonders what the next frontier will be: 'Don't Let AI Write Your Closing Arguments While Drunk' guidelines, perhaps?

So Anthropic has decided that the best way to handle a model they themselves call a 'culmination' of AI reasoning and scientific capabilities is to release it to only 40 corporations under the cuddly-sounding 'Project Glasswing'. Jared Kaplan, the chief science officer, explains this isn't about hiding the toy—oh no—it's about 'analyzing capabilities and preparing countermeasures'. Because when you've built something so advanced that you need to pre-plan the countermeasures, the obvious move is to hand it to a select club of global giants who definitely won't find any... creative applications. It's like selling a new strain of smallpox to only the world's top biotech firms for 'research purposes'—what could possibly go wrong?

The model, dubbed Claude Mythos, is positioned as the pinnacle of general capabilities in software engineering and scientific research, which is precisely the sort of dual-use toolkit that makes regulators break out in a cold sweat. But fear not, because Anthropic has thoughtfully limited access to 'leading corporations'—a group whose track record on responsible innovation includes such hits as 'move fast and break things' and 'ask forgiveness, not permission'. The idea that these entities will dutifully analyze threats and not, say, integrate Mythos into their own competitive arms races is the kind of optimistic fiction that would make a PR intern blush.

This follows the pattern we highlighted with their previous release, where Anthropic admitted Opus didn't advance the frontier compared to the then-unreleased Mythos—a bit of foreshadowing we noted. Now that Mythos is real, they've swapped incrementalism for a velvet-rope exclusivity model that does nothing to address the core issue: if it's dangerous enough to require 'countermeasures', maybe don't give it to anyone until you've actually built those countermeasures. But that would require a level of restraint apparently incompatible with staying relevant in the AI hype cycle. Cheers to progress, one carefully managed catastrophe at a time.