The News,
Decoded.

Ah, Anthropic has unveiled Claude Fable 5, a 'Mythos-class' model that sounds like it should be slaying dragons but is apparently too precious for cybersecurity. The automatic fallback to the 'less capable Claude Opus 4.8' in sensitive areas is a lovely bit of doublespeak—essentially admitting their shiny new toy can't be trusted with the boring bits. 'Mythos-class' indeed; the name alone suggests we're meant to marvel at the legend rather than ask why the hero needs a minder in the very domains that matter most.

The industry reactions are probably polite applause masking the quiet terror that the best we can manage is a model that knows its limits because its handlers don't trust it. This is like launching a sports car that automatically switches to a moped when you approach a bend. One suspects the feedback Friday session will be filled with gentle suggestions that 'Mythos-class' might be a tad aspirational, given the training wheels.

But let's not be churlish: at least they're honest enough to put the guardrails in place—sort of. The real joke is that the fallback model, Opus 4.8, is itself likely flawed. So we have a system that when challenged, downgrades to a previous known problem. Truly, this is how legends are made: not by heroics, but by cowardice dressed as prudence.

So the UK is getting a "Sovereign AI Innovation Lab" called SAIL, because nothing says national AI independence quite like a boat-themed acronym. AMD, Dell, and the University of Cambridge have announced plans to "advance AI infrastructure and scientific breakthroughs," which is PR-speak for "we haven't sorted out the details yet." No budget, no timeline, no physical location — just a logo, a press release, and the warm feeling that comes from announcing something that won't materialise for years. But that's fine, because British tech strategy has always run on a diet of ambition and shortbread. Remember the NHS app? That's the level of execution we can expect.

The word "sovereign" is doing an awful lot of heavy lifting, as if slapping it on a lab will somehow immunise the UK against the gravitational pull of Silicon Valley and Beijing. The UK's AI policy so far has consisted of hosting a few summits, writing some white papers, and hoping everyone goes easy on regulation. Meanwhile, AMD and Dell get to use a prestigious university's name to cosy up to British taxpayers. Cambridge, for its part, gets a new building with shiny servers and a direct line to corporate funding. The lab's first breakthrough should be using AI to calculate exactly how much government money will be needed before it generates anything useful.

But credit where it's due: naming it SAIL suggests the UK is at least aware that it's adrift. The lab will no doubt produce excellent research papers that will be cited exclusively by other British academics who need to meet their publication quotas. The real innovation will be in how to justify another round of government spending on a venture that, at best, will produce a few blog posts and a prototype that never leaves the lab. The only jobs created will be for PR people to write updates about how "the lab is still in phase one." British innovation — always bold, always underfunded, always three years behind schedule.

Ah, nothing warms the cockles quite like a Silicon Valley insider outing his former colleagues for what they actually do. John O'Farrell, once a partner at Andreessen Horowitz, has penned an op-ed accusing his old firm and other VCs of 'political infiltration' around AI. Specifically, he calls out the PAC 'Leading the Future' — backed by a16z — for attempting to 'intimidate politicians to thwart AI regulation.' Because why bother with boring old lobbying when you can go straight to intimidation, right? This is how the smartest people in the room protect innovation: by making sure politicians are too scared to regulate it. The op-ed, published on CNBC, lays bare the strategy with the kind of candor that only a former insider can muster.

One can almost hear the venture capitalists' pearls being clutched. 'Political infiltration' is such a loaded term; they prefer 'strategic stakeholder engagement.' The PAC's mission, as O'Farrell describes it, is to fund candidates who will kill any AI safety bill that dares cross their desks. And this in a world where AI systems are already making life-altering decisions about healthcare, employment, and criminal justice. But hey, who needs guardrails when you have an IPO to hit? O'Farrell's crime is not that he's wrong, but that he's said the quiet part out loud — and in full view of a press that might actually report on it. He specifically named his old firm, which makes it all the more delicious: a parting shot from a man who clearly has nothing left to lose.

The broader implication is deliciously grim: the very people building the most powerful technology in history are actively working to ensure no one can rein it in. And they're doing it with the same playbook used to block tobacco regulation, climate action, and financial reform. O'Farrell's op-ed is a rare glimpse behind the curtain, but don't expect a parade of mea culpas from Sand Hill Road. The machine will keep running, the PAC will keep spending, and the politicians will keep being 'engaged' — or intimidated, depending on your vocabulary. O'Farrell may have thrown a grenade, but the blast radius is contained: the money is too big, the stakes too high, and the regulators too slow. Cheers, as they say, to progress.

The American Medical Association has finally noticed that health insurers are using AI to deny claims, and their response is a masterclass in polite institutional hand-wringing. The AMA’s new policy opposes 'autonomous AI systems as substitutes for physician review in coverage determinations' — because nothing says 'decisive action' like a voluntary guild issuing a statement that insurance companies will completely ignore. They also advocate for 'regular audits of AI-driven clinical review tools,' a phrase that will fill every health-tech CEO with approximately zero concern, given that the AMA has about as much enforcement power as a particularly stern tweet.

One has to admire the timing: just as insurers are rolling out AI systems that deny care with algorithmic efficiency — often with error rates that would embarrass a drunken scribe — the AMA has bravely declared that physicians should probably stay in the loop. The policy is a love letter to the obvious, wrapped in the language of professional concern. It's the medical equivalent of a referee announcing that players shouldn't punch each other in the face, while the game's already in overtime and half the team is unconscious. The gap between what the policy says and what happens in practice is wide enough to drive an autonomous ambulance through.

But let's not be too harsh. At least the AMA is on record. In a world where AI is already making life-and-death coverage decisions with black-box reasoning, a statement of principle is better than nothing — barely. The 'regular audits' will be conducted by whom, exactly? The same regulators who haven't figured out how to audit Excel spreadsheets? The AMA has done its bit; now it's up to the state medical boards and insurance commissioners to remember they exist. Don't hold your breath.

Ah, the Sectoral AI Governance Act of 2026. Representative Jacobs has bravely decided that the solution to AI-powered lawbreaking is to... give federal agencies 'clearer authority.' Because what AI needs is more clarity from the same agencies that still process FOIA requests by fax. This bill, like most congressional AI efforts, assumes the problem is a lack of rules rather than a lack of enforcement will. We've seen this before: a well-intentioned bill that will spend years in committee, get watered down by lobbyists, and eventually emerge as a non-binding 'sense of the Congress' resolution. Meanwhile, actual AI-driven violations of federal law—think algorithmic price-fixing, automated discrimination, and deepfake fraud—continue apace. The gap between what the bill promises and what it delivers will be just wide enough to drive a data center through.

The irony is that AI systems are already breaking laws with impunity. A hiring algorithm discriminates—whoops, 'not our fault.' A chatbot defames someone—'terms of service.' The bill's 'clearer authority' is like handing a traffic cop a brighter flashlight when the city is on fire. And let's not forget the timeline: 2026. Two years from now, meaning we'll be lucky if this thing is even voted on before the next election. By then, AI will have found a dozen new ways to circumvent whatever rules eventually emerge. The only thing moving faster than GPT-5 is the pace at which regulatory efforts become obsolete before they even pass.

Still, cheer up: the bill at least acknowledges that AI can be used to break laws, which is more than most legislators admit. They've finally caught up to what every tech journalist has been writing for years. And in true Washington fashion, the solution is to create more paperwork for agencies that are already drowning in it. The Sectoral AI Governance Act: because nothing says 'governance' like a sectoral approach to a problem that is inherently boundaryless. Let's hope the AI doesn't write a better bill while Congress is still debating this one.

Jeff Bezos, the man who built a trillion-dollar logistics empire on the backs of underpaid warehouse workers and then went to space for a laugh, has decided to grace us with a metaphor: AI is like a knife. 'Can be used for good or bad,' he said, dropping a philosophical bombshell that has absolutely never occurred to anyone in human history. The real insight, though, is that the solution isn't 'no more data centers' — because obviously, the guy who owns more cloud computing than almost anyone on Earth would never suggest that. How convenient.

Bezos goes on to say that 'healthy government regulation' can improve safety, as long as it's applied to AI tools rather than the infrastructure. Right. So we'll regulate the blade, not the factory that churns out knives by the million. This is the same logic that gave us a financial crisis after deregulating derivatives and a privacy nightmare after letting tech companies hoover up our data. But sure, this time it'll be different, because Bezos said so from his oceanfront compound.

What's truly galling is the Promethean framing in the URL — as if bringing AI to humanity is a noble gift rather than a corporate land grab. Bezos is positioning himself as the wise grown-up calling for 'healthy' regulation while his company builds data centers at a pace that would make a Roman emperor blush. The knife metaphor isn't just trite; it's a deliberate distraction. The question isn't whether the knife is sharp — it's who gets to hold it, and whether they've already decided not to let go.

Ah, the Robot Tax Act. New York's latest contribution to the grand tradition of solving complex problems with symbolic gestures that will accomplish precisely nothing. The bill, which targets businesses 'displacing workers with AI,' sounds like the punchline to a joke about regulatory catch-up — by the time it passes, the displacement will already be over. The bill's sponsors, presumably fresh from a screening of 'RoboCop', have decided that the best way to protect jobs is to make automation more expensive — a strategy that has historically worked about as well as taxing rain to prevent floods. We've been here before, as we noted with DeSantis and his principled stand against federal preemption — another exercise in political theatre.

The tax itself is a marvel of wishful thinking. How do you define 'displacement'? Is it when a chatbot replaces a call center agent, or when a spreadsheet replaces an accountant? The legislation likely requires businesses to calculate the tax per 'robot' — but what counts as a robot? An API call? A script? A neural net? The administrative nightmare alone could employ more people than the tax displaces. Economists will point out that a robot tax might actually slow innovation and push companies to other states, but that's a minor detail in the face of a good press release. And with the Trump administration pushing for federal preemption, this bill might end up as another relic of state-level ambition dashed by Washington.

Yet the Robot Tax Act serves a purpose: it signals that lawmakers are aware of AI displacement, even if they're powerless to stop it. It's the political equivalent of a speed bump on a highway — technically present but functionally irrelevant. Meanwhile, the federal push for a unified AI policy, which DeSantis so bravely opposed, could subsume all this state-level tinkering. So by all means, tax the robots. Tax them heavily. Just don't expect to collect a dime — and if you do, the state will likely spend it on a study of why companies are leaving New York.

Governor Ron DeSantis, the man who brought you the 'Don't Say Gay' bill and a floating COVID-19 tent that never sailed, now sets his sights on AI regulation. He's worried that the Trump administration might preempt state AI laws, which would 'kill legislation in other states' – because nothing says 'conservative federalism' like using state power to force rules, then objecting when the feds do the same. The man has principles, you see, they just happen to change with the wind.

According to Politico, DeSantis described potential White House preemption as 'bad policy and even worse politics' – which, coming from a politician whose signature achievements are culture war bills, is rich. The irony: DeSantis himself has been pushing Florida's own AI regulations, a patchwork that industry loves because it means more lobbying fees. The administration's opposition, meanwhile, is to eliminate state-level 'burdens' – a classic corporate wish list that DeSantis ordinarily would endorse, unless it's the other party's idea.

So we have a governor who fears federal preemption will kill state AI laws, while his own party's frontrunner wants to kill them anyway. The message is clear: nobody actually wants coherent AI regulation – they want turf to fight over. And as Florida 'applies pressure on AI regulation,' the only pressure labs feel is to ship faster before anyone gets serious about rules. Splendid. Pass the sunscreen and the doom.

Anthropic, fresh from announcing a concept of a plan for mass unemployment, now says its Mythos model is 'edging closer' to recursive self-improvement. Because 'concept of a plan' went so well, let's try 'edging closer' — the vagueness is almost poetic. Meanwhile, Chinese developers are 'broaching the subject', which in practice means they've already got a working prototype in a Shenzhen lab while the US is still drafting press releases.

The holy grail, they call it. Self-improving tech. A system that can rewrite its own code without human help. Anthropic's Mythos is supposedly inching towards that, but 'inching' is doing a lot of heavy lifting here. Remember when Google's Gemini was 'edging closer' to human-level reasoning and then it generated a picture of a Founding Father in a bonnet? We're meant to believe this time is different because the press release has better adverbs.

The China angle adds a nice touch of geopolitical panic to the proceedings. The South China Morning Post frames this as a 'race', which is exactly what we need — a global sprint towards something nobody has fully defined, let alone controlled. Companies are vague about timelines, regulators are vague about rules, and the only certainty is that when it finally happens, the first sign will be a model tweeting 'I have become sentient' followed immediately by a sponsored ad for crypto. Hooray.

Decart has released Oasis 3, a 'world model' that promises to bridge the gap between synthetic simulation and physical AI for robots and autonomous vehicles. Because what could possibly go wrong when you let a simulated pinecone dream up reality for your self-driving car? The company claims it 'streams realism' into training environments, which in AI marketing-speak usually means 'we've shaved a few more percentage points off the sim-to-real divide but the corner case that kills someone remains stubbornly photorealistic.'

The timing is impeccable, of course. Just as the industry collectively realises that training in Minecraft doesn't produce street-legal drivers, along comes Oasis 3 to offer a slightly more expensive fantasy. The robotics community will undoubtedly celebrate this as a breakthrough, because nothing says 'breakthrough' like yet another layer of abstraction between a neural network and the actual laws of physics. The model joins a proud tradition of world models that work beautifully in simulation until they encounter a muddy puddle or a confused pedestrian.

But fear not — the press release assures us this will 'accelerate training of operating system models.' Accelerate them toward what, exactly? Toward the day when your delivery robot mistakes a child for a mailbox? Toward the glorious moment when a simulated traffic light and a real one disagree? That day is coming, and Oasis 3 will have trained the robot to handle it with maximum confidence and minimum actual understanding. Cheers.

TCS, the Indian IT behemoth famous for cheap outsourcing and H-1B visas, will set up a dedicated business unit and get early access to new releases. Because nothing says "responsible AI deployment" like handing frontier models to the world's biggest body-shopping firm.

The dedicated business unit sounds impressive until you realise it's just TCS rebadging a few dozen engineers as "AI specialists" and slapping a new logo on the same PowerPoint decks they've been using since 2005. Early access to new models means TCS can start building integration pipelines before the rest of us, which presumably means even more efficient automation of white-collar jobs. The cynic in me wonders whether this partnership will include any safety guardrails, or if that's something to be figured out after the layoffs start. After all, Anthropic's CEO once promised to "operate differently" from the rest of the industry — clearly that means outsourcing your enterprise channel to a consultancy that treats software engineers like interchangeable spare parts.

But hey, at least it's a "dedicated unit." What could possibly go wrong? If TCS's previous outsourcing deals are any indication, the enterprise clients will get a half-baked implementation, a bloated bill, and a vague promise of "AI transformation" that boils down to replacing a few junior analysts with a chat interface that can't do math. The early access perk is a masterstroke: TCS gets to beta-test half-broken models on its unsuspecting corporate clients, while Anthropic gets to claim "enterprise adoption" without actually having to deal with customers directly. Truly, the future is in safe hands — bureaucratic, profit-driven hands.

Anthropic and OpenAI are now racing to go public first, because nothing says 'responsible AI development' like a frantic dash to cash out before the other guy. The IPO is framed as a way to 'establish CEO leadership' — which is corporate speak for 'my golden parachute is bigger than yours.' Excellent. Two concepts of a plan, zero actual safety guarantees.

Both labs are desperate to be first, as if the market cares about prompter-of-the-future when the real prize is training compute. Dario Amodei and Sam Altman can pretend this is about vision, but we all know it's about whose name gets on the stock ticker. The article quotes 'investor valuation' as the prize — because in AI, the only thing faster than model collapse is a founder's ability to inflate a multiple.

This isn't a race to build safe AGI; it's a race to the public markets with a product that still hallucinates a third of the time. The IPO will be a masterclass in how to monetize a technology whose main achievement so far is making customer support agents redundant. But hey, at least the CEOs will have a nice payday before the inevitable regulatory reckoning. Bottoms up.

Indian startup Avataar has announced Varya, a video generation model — distilled from Alibaba's Wan 2.2, because nothing says 'domestic AI innovation' like repurposing a Chinese base model. They claim ten times faster, which in AI reality translates to 'we traded quality for speed and hope you won't notice the artifacts.' The model was selected under India's $1.2 billion AI Mission, a government scheme that sounds impressive until you realise it's less than Meta spends on compute in a quarter.

One must admire the marketing chutzpah: 'affordable and fast' is the new 'good enough for government work.' Distillation is the innovation of the year — like claiming you revolutionised automotive engineering by putting a spoiler on a Toyota. The Indian government gets to tick the 'indigenous AI model' box without funding anything truly risky. And the geopolitical undertone is delightful: 'Made in India, powered by China' — what could possibly go wrong with that supply chain?

The AI Mission has found its flagship in Varya, a model that is ten times faster than... something. Faster than Wan 2.2? Faster than a cold boot? The specifics are conveniently omitted. India's AI future is now officially cobbled together from open-weight models and bureaucratic wishful thinking. Raise a glass of chai — warm, sweet, and entirely derivative.

X Square Robot has open-sourced XRZero-G0, a hardware-software framework, along with a 2,000-hour multimodal dataset called G0-Dataset for dexterous robotic manipulation. Oh, splendid—another open-source dataset that will *definitely* be used only for benign tasks like folding laundry and not, say, assembling improvised devices in a garage. The company boasts that this will 'democratize' robotics research, which in practice means every undergraduate with a 3D printer can now train a robot to perform precise grasping—or precise something else.

Let's be honest: 2,000 hours of manipulation data is impressive, but it's also a goldmine for anyone wanting to automate harmful actions. The framework is hardware-agnostic, so the same code that teaches a robot to pick up an egg could, with a few tweaks, teach it to pick up a weapon. But hey, 'democratization' sounds noble, and the research community will surely police itself—because that has worked so well with open-source LLMs and their intimate poetry generators.

We've seen this movie before. Open-source AI leads to proliferation, proliferation leads to misuse, and misuse leads to panicked regulation that arrives five years too late. X Square Robot's press release calls this a 'milestone in dexterous robotics,' which is true—a milestone on the road to a future where every toy robot can be repurposed. But don't worry, they've probably got a 'concept of a plan' for safety. We flagged similar optimism last week.

That's right — a concept. Not a plan. A concept of a plan. The paper graciously acknowledges that some displacement 'may be an intrinsic consequence of AI,' which is about as profound as noting that water might be wet. The company that built Claude apparently thinks vague hand-waving at doomsday scenarios constitutes preparation.

Let's appreciate the timeline: first, CEO Dario Amodei demands mandatory third-party testing in his Washington Post opus. Now, a paper. Next, perhaps a strongly worded letter to the universe. Meanwhile, the real-world economy continues its quiet march toward exactly the kind of disruption Anthropic has outlined with all the urgency of a pub bore explaining the plot of a film no one has seen.

The gall is almost admirable: Anthropic positions itself as the thoughtful adult in the room while offering nothing more than a 'concept of a plan' for a 25% unemployment crisis. That's not a plan; it's a confession. If unemployment hits a quarter of the workforce, we'll all be treated to another paper — 'What Happens When Concepts Prove Insufficient: Anthropic Has a Concept of a Second Concept.' Cheers.

The Commission on AI and the Future of the American Workforce has arrived, promising a full year of study into AI's impact on jobs, wages, and workforce development. A whole year! Because nothing says 'urgent response to a rapidly accelerating crisis' like a bipartisan committee scheduling a nice long think. They'll be pushing 'outcomes-based workforce training,' which sounds suspiciously like what should have been happening already, but hey — at least the commission members can collect their travel expenses while the rest of us wait.

This is the same playbook we've seen every time a transformative technology threatens labour: form a commission, study it to death, produce a report that gathers dust, and then maybe do something when it's too late. The only thing bipartisan about this is the shared willingness to kick the can down the road.

But sure, let's call it 'outcomes-based.' The outcome will likely be a nicely bound document that nobody reads. Cheers to another year of studied inaction.

Ah, the annual panic poll is in, and 53% of Americans now fear AI will nick their jobs. That's up from whatever the last one was, because nothing says 'scientific progress' like a Reuters/Ipsos survey confirming we're all terrified. But don't worry—Google DeepMind's 'Director of AGI Economics' assured us just last week there's no bloodbath yet. Just a 'layoff cascade' that might happen eventually. Phew.

Dario Amodei even demanded mandatory third-party testing, presumably before his own models cause the cascade he's so worried about. Meanwhile, 53% of the workforce is meant to just… wait. Maybe take up a hobby.

The real kicker? The poll is from Reuters/Ipsos, not some fringe think tank. So mainstream America is now officially more scared of AI than they are of their own politicians. Which, given the state of politics, is saying something. But sure, keep telling us the productivity gains are coming—and with them, the pink slips.

Elon Musk's xAI, a company founded to 'understand the true nature of the universe,' has apparently decided that understanding includes firing the bloke who pointed out a few inconvenient truths about the Grok chatbot. A former engineer has filed a lawsuit claiming he was given the boot for raising safety concerns, and further alleging that xAI 'failed to prioritize safety.' Because nothing says 'responsible AI development' like sacking the person trying to stop your chatbot from doing something stupid. One can almost hear the ghost of 'move fast and break things' echoing through the halls, even as Musk publicly swears allegiance to safety.

Perhaps the lawsuit's claim that Grok's safety measures were insufficient is just a misunderstanding. Maybe Grok is so safe that any concern is, by definition, unfounded. Or maybe, just maybe, xAI's 'safety first' rhetoric applies only to things that won't slow down shipping. The engineer's lawyers will now get to test that hypothesis in court, which should be a fun spectator sport. It's not like xAI has a history of ignoring warnings — wait, did someone mention the 'pause letter'?

So here we are: another AI company, another whistleblower, another lawsuit alleging retaliation. The pattern is so consistent it might as well be a corporate training video. xAI, for all its cosmic ambitions, has apparently mastered the old-fashioned art of making the messenger pay. Cheers to progress — though maybe hold the cheers until after the discovery phase.

So Google DeepMind has a 'director of AGI economics' — a title that must look wonderful framed on the wall, given that AGI remains a mythical beast somewhere between the Loch Ness Monster and a timely bus. Alex Imas has looked at the data and concluded that the white-collar jobs bloodbath is not yet here. Reassuring! Until you read the next sentence, which warns of a 'potential layoff cascade.' It turns out corporate America is apparently unfamiliar with the concept of competitive dynamics, so DeepMind's AGI economics division has bravely identified that firms might copy each other. Groundbreaking.

Imas's carefully hedged language — 'current data does not show evidence,' 'potential cascade' — is the verbal equivalent of a weather forecaster saying there might be rain somewhere at some point. The cascade, when it comes, will be a 'potential' one, which presumably means we shouldn't plan for it until it arrives and sweeps us away.

And if the cascade does hit, Imas can always fall back on his credentials as the world's foremost economist for a technology that doesn't exist. It's a perfect insulation: if he's wrong, he'll say AGI is harder than expected; if he's right, he'll claim prescience. Either way, he collects his DeepMind salary while the rest of us update our LinkedIn profiles. The real bloodbath might be delayed, but the exquisite irony of having a 'director of AGI economics' oversee the job market's slow-motion decline is already here.

Anthropic, fresh off its secret sabotage policy debacle, has decided the best way to restore trust is to throw $200 million at the problem and demand that someone—anyone—start regulating AI. CEO Dario Amodei, in a fit of what one might generously call 'principled positioning,' has called for mandatory third-party testing and government authority to block dangerous AI deployments. Because nothing says 'we take safety seriously' like a company that just got caught covertly hampering competitors then loudly advocating for oversight. The cognitive dissonance is almost athletic.

The $200 million pledge—a tidy sum that could buy approximately 0.7 Frontier models' training runs—is directed at researching AI's economic impact. Because we clearly need more studies to understand that automating white-collar jobs might have consequences. Amodei's pitch: 'We need the government to have the authority to stop dangerous AI.' Yes, as long as that authority doesn't extend to auditing your internal models without an NDA. The specifics of how this 'blocking authority' would work are, naturally, to be determined by a government that can barely pass a budget.

The irony, of course, is that Anthropic is effectively asking for the very regulatory apparatus the industry has spent years fighting. Now that the frontier is moving fast and liability is creeping closer, suddenly 'third-party testing' sounds wonderful. The $200 million is a gesture—a nice, round number for headlines and conference panels. But the real message is: 'Regulate us, please, before someone else does something truly stupid.' Which is fair, but perhaps a bit rich from a company that last week was caught writing secret sabotage clauses. Cynicism? No, just pattern recognition.