The News,
Decoded.

Ah, the United Nations, fresh from its preliminary warning that AI could cause catastrophic harm, now blesses us with an 'AI for Good' commission. Because nothing says urgency like a naming committee with the creativity of a greeting card aisle. The first meeting is July 8 in Geneva, where top tech execs and heads of state will sit in a room and forge global AI rules. Or at least, that's the press release version.

The commission is explicitly a response to splintered regulation. Splintered, as in, there are currently about 47 different regulatory frameworks globally, none of which agree on what an AI even is. So the solution is to gather the people who have the most to lose from regulation (tech CEOs) and the people who have the most to gain from appearing to do something (politicians) and ask them very nicely to agree. This is the diplomatic equivalent of asking a pack of wolves to design the sheep's security system, chaired by a particularly optimistic sheepdog.

And yet, the commission might actually produce something. A non-binding statement, perhaps. A 'framework' that every country will interpret however it pleases. The real tragedy isn't that this won't work; it's that this is the best idea anyone has had. The UN's own panel warned that AI could cause catastrophic harm, and the response is a committee meeting in Geneva in July. If this were a film, it would be a dark comedy. But it's real life, and the credits aren't rolling.

The United Nations has blessed us with a preliminary report warning that AI could "cause catastrophic harm" – because nothing says 'credible deterrent' like a document that is, by its own admission, unfinished. The report, covered by The Jerusalem Post, notes that AI capabilities are outpacing both scientific understanding and government policy, which is about as useful as a weather report that says 'it might rain eventually.' The key word here is 'preliminary' – the UN's get-out-of-accountability-free card, ensuring that any actual action can be deferred to a future, final version that may or may not ever materialise.

The report also warns of the risk of "losing control over increasingly autonomous AI systems" – a fear that safety researchers have been raising for a decade, but which apparently only becomes official when a UN committee types it up and releases it to a world that has already moved on. One suspects the AI labs will treat this with the same gravitas they treat a Terms of Service update: a quick glance before clicking 'I agree' and continuing with business as usual. The report outlines safety concerns but does not specify what those concerns look like in practice – no concrete thresholds, no binding recommendations, just a gentle suggestion that maybe we should think about it.

And where does this report land? On the desks of policymakers who are already several steps behind the technology, in a news cycle that will forget it by the next model release. The UN has effectively said, 'We've noticed the problem; please hold while we form a working group to form a subcommittee to draft a further preliminary report.' Meanwhile, AI companies continue their relentless scaling, comforted by the fact that the worst they face is a strongly worded PDF. It's the international community's favourite move: identify an existential threat, produce a document, file it away, and hope that the robots are as slow as the bureaucracy.

The House has done it again — passed the KIDS Act package with a straight face, as if 'protecting children' from AI chatbots will somehow keep pace with the tech that's already in every teenager's pocket. The bipartisan bill includes COPPA 2.0 provisions expanding privacy obligations and addressing AI chatbot safety. Because nothing says 'we understand the problem' like updating a law named after a 1990s children's TV show to regulate a technology that moves faster than Congress appropriations.

But oh, the elegance of it! The KIDS Act — an acronym so desperate for meaning it practically begs you not to ask what it stands for — will now lumber over to the Senate for a nice long nap. Senators will hold hearings, call expert witnesses who will warn that the timeline for real harm is measured in months, and then maybe, just maybe, pass a version that's already obsolete by the time it lands on the President's desk. The AI chatbot safety provisions? Lovely, so long as you ignore that the chatbots in question have already been training on children's data for years.

This is the regulatory gap score we've come to know and love: a legislative lag that's less a gap and more a canyon. While Congress pats itself on the back for a 'bipartisan achievement,' the AI industry rolls out new capabilities that skirt these very protections. The KIDS Act is a beautiful gesture — like putting a plaster on a bullet wound and calling it trauma care. The only question is whether the Senate debates it fast enough to matter, or slow enough to be irrelevant. Place your bets.

Ah, the AI gods have spoken, and their name is Mythos — because nothing says 'existential threat' like a name you'd give a fantasy novel protagonist. But fear not, mere mortals, for your AI security tools are now almost as good at finding zero-day vulnerabilities as the mythical beast itself. Almost. Because a quarter of attacks still slip through the cracks. In any other field, that's a C-grade. In AI security, we throw a party.

The article tells us that AI security tools 'already rival Mythos' ability.' Which is like saying a go-kart rivals a Ferrari in the parking lot — technically true, but you wouldn't bet your life on it. And yet, here we are, celebrating parity with a system that probably has a cooler name than actual capability. Mythos sounds like a Greek epic, but in practice, it's just another tool that got a fancy branding budget. Meanwhile, the hackers are still out there, exploiting the same vulnerabilities they've been exploiting since the 90s, and your AI is just now catching up.

So yes, you don't need Mythos to defend against Mythos, because you have your own slightly-less-mythical AI. But that quarter of attacks? That's the real story. Because every time a zero-day is found, it's a race between the good AI and the bad AI, and the bad AI is still winning a quarter of the time. But hey, progress! Now if you'll excuse me, I'm off to patch my system before the next quarter of attacks find their way in. Cheers.

Amazon MGM Studios, the streaming arm of the company that powers half the internet and wants an AI in every home, has dropped the film 'Artificial' because it dared to be critical of AI. The irony is so thick you could build a chatbot out of it. Here we have a corporation whose entire AI strategy relies on the public swallowing the line that this technology is an unalloyed good, and when a movie suggests otherwise, it's off the schedule faster than you can say 'safety researcher concerns.' This is not a business decision; it's a muzzle.

The film, now rescued by indie distributor Neon, will get a proper release—but the damage is done. The chilling signal has been sent: if you want to make art that questions the tech religion of our age, don't expect a platform from the very companies building it. We already have the AI free speech debate over export controls, and now we see private sector censorship in action. What's next? Will Amazon pull books critical of AI from its bookstore? Will it throttle search results for negative articles? The pattern is clear.

And what of Neon, the white knight in this story? They're to be applauded for picking up a film that the biggest of big tech dropped. But let's not pretend this is a happy ending—it's a glimpse of a future where only small, independent distributors will touch anything that paints AI in an unflattering light. The mainstream platforms will remain a safe space for tech boosterism. 'Artificial' might find its audience, but the fact that it had to be rescued should terrify anyone who believes in honest public discourse about the most consequential technology of our time.

Oh, marvellous. One might think such overwhelming bipartisan agreement would translate into immediate legislative action. One would be adorably naive.

So the public is not just vaguely worried — they are handing policymakers a detailed, signed, and stamped wishlist. And yet, Congress continues to treat AI regulation like a particularly demanding houseplant: they admire it, talk about it, but never actually water it. The gap between what voters want and what legislators deliver is now measurable, and frankly, embarrassing.

It's almost touching to see the AI Policy Institute try to nudge the Overton window with data. 'Look,' they seem to say, 'your constituents are terrified and they agree. Do something.' But we all know how this goes: the poll will be cited in a hearing, praised by both sides, and then quietly buried under a mountain of tech lobbyist submissions. Bless their hearts.

Senator Mark Warner has introduced a draft bill to create a federally vetted list of 'secure and trustworthy AI agents,' because nothing says 'trust us, we're from the government' quite like a bureaucratic stamp of approval. The FTC, that beacon of agile regulation, will develop security and identity standards—presumably between reviewing Facebook's latest privacy scandal and approving yet another data broker merger. The list, once compiled, will presumably assure us that the AI agents handling our banking, healthcare, and democracy are definitely, positively, not going to go rogue.

The oversight committee will be staffed by the same folks who thought the Nissan Z was a good idea. And if history is any guide, the first version of the list will be about as comprehensive as a TSA no-fly list from 2002—missing key names, full of false positives, and entirely reactive.

The real kicker? The bill's existence signals that Washington has finally noticed AI agents exist, but their response is to create a yelp-for-government-contractors list rather than, say, mandate liability or testing. Wonderful. Now the AI agents can get a 'DHS Approved' sticker to wear alongside their 'I passed the bar exam' badge. At least the procurement consultancy firms will have something to bill for. Pass the popcorn—this slow-motion train wreck is going to be a masterclass in regulatory theatre.

Oh, marvellous — a new survey reveals that 76% of employees have already taken it upon themselves to sign up for and use AI tools at work, while 41% report their employer offered zero training or guidance. So we've reached the glorious stage where the workforce is quietly installing autonomous systems on corporate laptops, and management is looking around going 'Wait, what's a chatbot?' This isn't shadow IT; it's shadow AGI. Your receptionist is now a GPT prompt engineer, and HR doesn't even have a policy on it.

Naturally, the same employers who can't be bothered to draft a one-page 'don't paste our entire customer database into ChatGPT' memo will be shocked — shocked! — when data leaks, biased decisions, or hallucinated contract terms start appearing. The 41% figure isn't just neglect; it's a lawsuit waiting to happen, and probably one that will be written by an AI that the employee found on Reddit. One can almost hear the legal departments scrambling to figure out who 'owns' an AI agent's mistakes when nobody approved its use in the first place.

And what of the broader picture? This survey, conducted by [whoever — the article doesn't name the firm], suggests that AI adoption isn't being directed by boardroom strategy but by bored interns with a ChatGPT Plus subscription. The regulatory gap widens, the capability leap accelerates unbeknownst to executives, and somewhere a middle manager is still trying to figure out what 'generative AI' means. Enjoy the Wild West, everyone — just make sure your indemnity clauses are up to date.

Ah, the great AI regulatory push has finally arrived — in the form of a 'discussion draft' from Massachusetts Rep. Lori Trahan. Because nothing says 'robust oversight of a global, rapidly accelerating technology' like a piece of paper that hasn't even become a bill yet, discussed in one state. The press release is careful not to mention any actual provisions, timelines, or enforcement mechanisms, because why spoil the surprise? We're all supposed to clap because someone in Congress is 'leading a discussion.' It's the legislative equivalent of a participation trophy.

This is the same dance we've seen a dozen times: a horrified public, a few earnest politicians, and then the tech lobbies quietly ensure the final text has more holes than a Swiss cheese. The discussion draft is a classic Washington gambit — make it look like you're doing something, while the industry continues to roll out models that can write poetry and pass the bar. By the time this thing becomes law (if ever), the AI it aims to regulate will be as quaint as a Tamagotchi.

And let's not forget the real kicker: state-level regulation of AI. Because AI respects state borders, right? A model trained on global data will surely comply with Massachusetts's rules. This is like trying to regulate the weather with a zoning ordinance. But sure, let's have the discussion — while the machines get smarter, the lobbyists get richer, and the rest of us get a front-row seat to the greatest regulatory farce of the century.

The Red Queen Gödel Machine: a name that sounds like a rejected Doctor Who villain, but is actually NVIDIA and Cambridge's latest contribution to the AI arms race. The paper describes a system that can generate and test its own code, which is just a fancy way of saying 'we've taught it to give itself upgrades.' Naturally, this has sparked a debate about whether we'll hit ASI by 2028 — because what else would you do with a self-improving AI but set an arbitrary deadline?

The irony is almost too thick to cut. The paper is being reported by Kucoin, a crypto exchange, which is like having your AI safety briefing delivered by a roulette wheel. But sure, let's all get very excited about a system that writes its own code and then tests it. Because what could possibly go wrong with a machine that can recursively optimise itself without human oversight? I'm sure the Cambridge researchers have included a sufficiently robust kill switch. Probably.

Meanwhile, the '2028 ASI timeline' debate rages on, with tech optimists predicting utopia and everyone else quietly updating their CVs. The Red Queen Machine is named after a character who had to keep running just to stay in place — which feels apt. We're running faster and faster, and the finish line keeps being moved to 'any day now.' At this rate, the AI will be evolving itself while we're still arguing about whether it's a good idea. Splendid.

Ah, the latest entry in the "Humans: The Architects of Our Own Misery" sweepstakes. Nobel economist Robert Shiller has gravely informed us that all those breathless headlines about AI destroying the job market might actually be achieving that outcome through sheer collective panic. "While the job market has slowed for a host of reasons, there are reports that fear of an AI apocalypse is worsening the freeze," Shiller writes, confirming that nothing kills employment quite like a really good, scare-the-pants-off-everyone narrative about how employment is about to be killed. Bravo, everyone – we've triggered the robot apocalypse before the robots even got out of the lab.

The tech sector, meanwhile, is doing its usual helpful thing: Anthropic's Mythos 5 just got cleared for limited release, OpenAI is delaying its IPO to 2027, and every CEO from Satya Nadella to Larry Ellison is insisting that your company needs its own custom model yesterday. The actual displacement of workers by algorithms is still mostly theoretical, but the freezing effect Shiller describes is already measurable. Companies are hoarding cash, delaying expansion, and firing people "just in case" the AI comes for their jobs – like a terrified passenger bailing out of a lifeboat before the ship has even sprung a leak. The "persistent predictions" he refers to are the very headlines that have become a staple of our daily news diet, and they're doing more damage than any model ever could.

Shiller, at least, has the decency to call this what it is: a self-fulfilling prophecy that we're gleefully writing together. But his warning will likely be filed alongside every previous caution about regulatory capture and capability leaps – acknowledged in passing, then ignored until the mess is complete. The irony is that the real AI doomsday scenario might not involve superintelligence or rogue models, but simply the miasma of fear that precedes them. Congratulations, humanity: we broke the job market with nothing more than bad press and a vivid imagination.

So Big AI has finally found something to agree on: they'd like the Trump administration to tell them what's not okay. How quaint. Just weeks after begrudgingly slowing down GPT-5.6 at the administration's 'request,' we now have Dario Amodei telling us he's 'glad' the administration is taking AI seriously — while quietly noting the process is 'abrupt' and 'opaque.' That's code for 'we have no idea what the rules are and it's making investors twitchy.'

The article's title practically writes itself: Big AI had a point. Of course they did. When the only signals from the White House come through backchannel requests and carefully worded 'suggestions', the message isn't clarity — it's a power game. The administration wants the ability to lean on companies without the accountability of explicit legislation. And the labs, having spent years telling regulators to back off, now find themselves begging for a rulebook they can't read.

All this while China's Zhipu is cheerfully open-sourcing models that match Opus at a fifth of the cost, as we noted last week. But sure, let's have a 'thoughtful dialogue' about boundaries while the rest of the world just builds. The current situation is the worst of both worlds: enough regulation to slow down American companies, not enough to actually govern what they build.

Ah, Colorado. The state that brought us the Rocky Mountains, craft breweries, and now — a masterclass in regulatory hand-waving. Last month, its overhaul of AI bias regulations replaced a 'broad duty to avoid algorithmic bias' with 'more modest disclosure requirements.' In other words, they swapped a firm 'don't be evil' for a gentle 'maybe tell us something, if you feel like it.' And topping it all off, the law is utterly unclear on whether trade secrets can limit even those disclosures. So any developer can simply declare their bias-detection process a proprietary secret and carry on as before. Brilliant.

You have to admire the elegance of it. Instead of forcing companies to actually audit their models for discrimination, Colorado has created the perfect loophole: just claim your algorithm's inner workings are trade secrets, and the public can go whistle. The law's architects probably thought they were striking a balance between innovation and accountability. But in practice, they've handed the AI industry a 'get out of jail free' card wrapped in intellectual property law. The only thing more predictable than this outcome is the industry's inevitable use of it.

This isn't just a Colorado problem. It's a preview of how every future AI regulation will be hollowed out by trade secret claims. The pattern is now set: pass a law that sounds tough, leave a giant loophole, and let the companies drive a truck through it. Colorado's AI bias law isn't a safeguard — it's a permission slip with a ribbon on it. Well done, everyone. Well done.

Oh, how comforting. Financial regulators, those bastions of technological foresight, have finally noticed that AI might be a bit of a cybersecurity problem — and their brilliant solution is to adopt more AI. FINMA president Marlene Amstad, speaking with all the gravity of someone who has just discovered fire is hot, told us that banks 'must move quickly to adopt new technology' to plug vulnerabilities. Because nothing says 'risk management' like handing the same tool that caused the problem to the people who didn't see it coming. FINMA even helped create a forum within IOSCO to promote AI adoption by watchdogs. Yes, a forum. Because what this crisis of algorithmic chaos really needs is more committees.

Let's be clear: the same technology that supercharges cyberattacks is now going to be used by regulators to counter those attacks. It's like fighting arson with matches. The article doesn't mention any specific AI tools or their efficacy, just the breathless urgency of 'must move quickly.' One imagines the regulator's new AI system will be trained on decades of regulatory filings, producing recommendations like 'send strongly worded letter' and 'establish working group.' Meanwhile, the actual hackers are deploying generative models to craft phishing emails that would fool your own mother.

And the punchline? This forum within IOSCO — the International Organization of Securities Commissions — is going to 'promote adoption of AI by watchdogs.' So we're creating a global network of regulators running AI on top of AI, a matryoshka doll of automated oversight. What could possibly go wrong? When the first AI regulator hallucinates a market crash and triggers a real one, at least we'll have a forum to discuss it. Cheerio.

Oh, brilliant. Patronus AI — named after the Harry Potter spell that repels dementors — has raised $50 million to build 'digital worlds' for stress-testing AI agents. Because apparently, the real world is now deemed unfit for our automated overlords until they've had a supervised playdate in a simulated one. At $50 million, that's a lot of venture capital constructing a diorama so we can watch our digital toddlers fail to start a fire in a fake kitchen.

The concept is simple: catch catastrophic failure in a sandbox before deployment. But one wonders: what happens when the sandbox itself is buggy? Or when the AI, having aced the test, graduates to reality with a gold star? We've seen this before — it's 'overconfidence in controlled conditions,' and we've been watching the rerun since the banking crisis. Still, $50 million says this time it'll be different, because AI money buys alternate realities.

The article doesn't name the investors, but presumably they believe that a fake world is a reliable proxy for the messy one outside. It's a lovely fantasy: build a Truman Show for your AI, declare it safe, and release it. What could go wrong? (Everything.) But at least Patronus gets paid. Look for the IPO of a world within a world within a startup in 2027.

Oh, marvellous. ControlAI—a nonprofit so noble it believes that explaining the risks of superintelligence to Congress will somehow prevent it—is now 'briefing U.S. lawmakers.' Because the same legislative body that couldn't tell you the difference between a transformer and a lightbulb is clearly the right audience for a chat about existential threats. They're also running similar campaigns in the U.K. and Canada, because nothing says 'global coordination' like three countries having three separate conversations.

One imagines the briefing: a polite PowerPoint, some carefully worded hypotheticals, and then a gentle fade into irrelevance as the lawmakers nod sagely and promptly forget everything. The gall of assuming that a few meetings will change a regulatory environment that's been a decade behind since day one is truly something. ControlAI might as well be reading bedtime stories to a sleeping bear—comforting for them, but the bear will still eat the world.

Still, credit where it's due: they've identified the correct problem—superintelligence—and the correct approach—talking to people. It's just that the people in question have the attention span of a goldfish and the regulatory appetite of a dieting elephant. Meanwhile, the labs keep scaling, the models keep leaping, and the briefings keep happening. Round and round we go. Cheers.

So South Korea's financial regulator—presumably not content with the country's existing track record of cheerful deregulation—has announced plans to ease 'network separation rules' for financial AI. Because nothing says prudent oversight like reducing the barriers between systems that handle people's life savings. To balance this daring loosening, they're simultaneously rolling out an AI risk management framework and a sandbox for AI agents. The cognitive gymnastics required to justify 'we'll let the AI mingle more freely, but also we'll put it in a box' would impress an Olympic judge.

The risk management framework is, no doubt, a magnificent document—a thick binder of 'shoulds' and 'mays' that will sit on desks while actual financial AI systems quietly optimise their own reward functions. And the sandbox? A beautiful concept: a controlled environment where AI agents can be tested before being deployed into the wild. It's like releasing a pack of wolves into a training enclosure and hoping they'll learn vegetarianism. Given the pace of AI development, by the time the sandbox rules are finalised, the agents in question will have already evolved to negotiate their own terms.

What's truly delightful is the timing. South Korea, a nation that prides itself on technological speed, is essentially admitting that its previous approach—which apparently involved no special AI rules—needs both loosening and tightening simultaneously. This isn't regulation; it's regulatory whiplash. The financial regulator's dual announcement reads like a committee that couldn't decide between 'let's encourage innovation' and 'let's not get blamed for a collapse,' so they did both. Expect headlines from banks celebrating the easing, followed by embarrassing incidents that prompt the real tightening later. All very predictable, all very human.

So the grand plan was to crush a single state assemblyman to prove the AI lobby's electoral might. Leading the Future, a pro-AI super PAC with more cash than common sense, poured a fortune into defeating New York Assembly member Alex Bores, a Democrat who presumably had the temerity to question the industry's unregulated expansion. The result? Bores won, and the PAC has since gone radio silent — no statements, no concession, no appetite for any further races. It's the political equivalent of a startup burning through its Series A and then declaring it's 'pivoting to stealth mode.'

You have to admire the discipline, really. Most losing campaigns issue a gracious concession, a hint of defiance, even a vague threat to return. But Leading the Future has clammed up so thoroughly that it's almost as if they hope the entire affair will be forgotten if they just don't talk about it. The name itself — 'Leading the Future' — now reads as unintentionally ironic, given that the group's only known electoral foray went down in flames and left them unwilling to say a single word about it. One imagines their donors are thrilled: nothing says return on investment like your super PAC losing to a state assemblyman and then pretending it never happened.

For the rest of us, this silence is instructive. It reveals that even with virtually unlimited money, AI's political allies can still lose when they underestimate the voters' attachment to local representation. But don't mistake this for a permanent retreat. The AI industry has deep pockets and a long memory; they'll be back, perhaps with a more sophisticated strategy and an even bigger war chest. For now, they're licking their wounds and hoping no one asks about the millions they just torched. Enjoy the quiet, New York — it won't last.

Oh, splendid — Goldman Sachs has deigned to tell us that the 'AI job apocalypse' is overstated. Because nothing says reassurance like an investment bank that missed the 2008 crash, the dot-com bust, and basically every systemic risk until it was too late. The report, covered by Economic Times, concedes that AI will 'reshape labor markets over the next decade,' but insists we shouldn't panic. Just a 'reshape,' mind you — like a gentle rearrangement of deck chairs on a vessel they assure us is not the Titanic.

Of course, 'overstated' is doing a lot of heavy lifting here. When Goldman says 'not an apocalypse,' they mean you might not lose your job to a robot tomorrow. But 'uneven disruption' is banker-speak for 'some sectors will be hammered, others will be fine, and if you're a call-center worker in Cleveland, good luck.' The report's nuance is basically: 'The end isn't nigh, but your specific career might be.' How reassuring — it's not a meteor, just a slow-motion landslide.

And what's the timeframe? A decade. Because long-term predictions are the only safe ones for banks — by the time they're wrong, everyone's forgotten. Their track record on predicting economic disruption is roughly as reliable as a coin flip — but with fancier charts. So pop the champagne? Or just hold onto your CV and wait for the 'uneven' bit to hit your sector? I know which I'd bet on.

So the 'real AI safety discussion' isn't happening, according to CleanTechnica. What a surprise — the same week that Arctic Wolf reveals open-source models are nearly as good as Anthropic's Mythos at exploiting vulnerabilities, and the Trump administration asks OpenAI to limit GPT-5.6 to 'government-approved partners' (because that's worked so well). The conversation about what happens when these things go rogue is apparently too awkward for polite society, so instead we get a moratorium proposal on datacenters until safety and job displacement are sorted. Yes, because the answer to building too much AI infrastructure is to stop building — a plan with approximately zero chance of being taken seriously by anyone with a GPU order book.

Meanwhile, the actual discussion? A ghost at the feast. Companies hire 'ethics boards' that rubber-stamp deployment, while workers wondering if their next payslip is AI-generated get a dashboard from California that updates monthly with unemployment claims. Oh, and the Pentagon has decided that a data centre powering Grok is a matter of national security — because nothing says 'responsible AI oversight' like letting the military decide which compute gets shielded from environmental review. The article's call for a 'moratorium' is about as likely to succeed as a sober conversation in a champagne warehouse.

But don't worry — the labs assure us that safety is 'core to their mission' right before releasing another model that nobody asked for. The regulatory gap? It's not a gap, it's a chasm with a welcome mat. And the only discussion happening is the one where we pretend that building bigger datacenters faster will somehow solve the problems created by building bigger datacenters faster. Cheers — another round of compute for everyone.