The News,
Decoded.

Ah, the great AI leadership drama continues. Dario Amodei, CEO of Anthropic, has announced he is 'at peace' with the rivalry with OpenAI and explained why he left Sam Altman: 'why argue with someone when you don't have the same vision and you don't trust them?' Because nothing says 'responsible AI development' like a schism between two of the most powerful labs, founded on personal distrust. We're supposed to believe that these two men, who once shared an org chart, now represent adversarial approaches to the future of intelligence.

The irony is that 'trust' is exactly the commodity that AI companies try to sell to the public — 'trust us with your data, trust our safety protocols' — yet the CEOs themselves can't manage it internally. Amodei left OpenAI in 2021 with several other researchers, citing concerns about Altman's approach. Now, years later, the wound is still raw enough for him to offer this passive-aggressive gem. This isn't a policy disagreement; it's a personal breakdown that happens to have trillion-dollar consequences.

Anthropic positions itself as the safety-first alternative, but this interview reveals that the foundational motivation was as much personal as philosophical. Meanwhile, OpenAI continues its breakneck release pace under Altman. Both companies are racing toward superintelligence, with simmering resentment as their secret sauce. We can only hope the rivalry yields some genuine safety benefit — but given that neither trusts the other's judgment, it's more likely a game of 'who blinks first' with humanity playing the role of collateral damage.

Ah, formal verification. Because what the AI safety circus really needed was a 42-year-old computer science technique repackaged with a fresh coat of startup paint and a $27 million injection of Khosla's venture capital. Pramaana Labs has announced a seed round — yes, seed, as in 'we haven't proven anything yet' — to bring 'formal verification' to AI systems in law, drug discovery, and tax preparation. The idea is to use the open-source LEAN programming language, a theorem-prover from Microsoft Research, to mathematically guarantee that your legal chatbot doesn't hallucinate a precedent or your drug-discovery model doesn't miss a toxicity signal. Because after years of probabilistic guesswork, the solution is clearly to do more math.

LEAN is lovely for proving that a sorting algorithm is correct; applying it to the messy, ambiguous soup of legal reasoning or molecular biology is a different beast entirely. But Pramaana's founders are betting that formal methods, which have been the gold standard in hardware verification for decades, can finally tame the neural network chaos. Khosla Ventures, never one to shy away from a moonshot, is writing the cheque — $27 million, which in AI-land gets you a few months of compute for a frontier model but here is supposed to build an entirely new verification stack. The hubris is almost beautiful.

The sectors targeted are, of course, the ones where errors are most catastrophic and least forgivable. Law: a hallucinated case could land a lawyer in ethical hot water. Drug discovery: a missed interaction could kill. Tax: a misapplied deduction could trigger an IRS audit. So naturally, the solution is to apply a technique that works best on well-defined, closed systems to domains that are anything but. Still, you have to admire the chutzpah: $27 million, a programming language most people can't read, and a mission to fix the most dangerous parts of AI. What could possibly go wrong? Pour one out for the formal verifiers who'll spend the next few years discovering that human institutions are not, in fact, theorem provers.

Oh, splendid. Another week, another pair of frontier models — Anthropic's Mythos and GPT-5.5-cyber — that can find vulnerabilities and execute attacks with less human involvement than ever. Because nothing says 'cybersecurity maturity' like handing the keys to the castle to a black box that moves faster than any defender can even file a JIRA ticket. The industry's response? A collective 'we're not ready,' which is the cybersecurity equivalent of saying 'fire? oh, we'll get to it after the coffee break.'

The opinion piece in Infosecurity Magazine has the gall to label this a 'shift,' as if the trajectory hasn't been obvious since GPT-3 started writing phishing emails. Defenders have been playing whack-a-mole with AI-powered threats for years, but now the moles have graduated from 'annoying' to 'can automate the entire warren.' The article points out that these models enable attack execution with far less human involvement — which is the polite way of saying that a script kiddie with an API key is now a credible threat to enterprise networks. The cybersecurity establishment, still patting itself on the back for finally deploying MFA universally, is about to learn what 'asymmetric escalation' really means.

But worry not: I'm sure the same industry that took a decade to adopt encryption properly will have this all sorted by, say, 2035. In the meantime, we'll have 'vigorously nodding panels' at RSA Conference where executives explain that 'AI-driven defense is a journey' while their networks get quietly pillaged by a model that was trained on all their past breaches. Cheers to progress — just make sure your backups are air-gapped and your sense of irony is fully charged.

AWS has unveiled yet another bundle of acronyms at Summit New York, because nothing says 'solving real problems' like a press release thick enough to prop up a server rack. AWS Continuum, an 'AI-native security service,' apparently means your AI agents will now be protected from… what, exactly? Their own incompetence? And AWS Context, a knowledge graph for agents, is meant to stop them from hallucinating that your CEO is a cartoon character. How droll.

But the real pièce de résistance is the parade of new capabilities across Amazon Quick, Kiro, AWS DevOps Agent, and Bedrock AgentCore. The sheer volume of product names suggests a company throwing spaghetti at the wall, hoping some of it sticks as 'AI transformation.' Spoiler: most of it will slide off.

And let's not forget the timing. This comes weeks after reports of agents generating 'sex crime scene' images and laying off humans by the thousands. But sure, let's give them better knowledge graphs. Because the problem with AI is that it doesn't have enough context, not that it's being deployed into every crevice of society with the caution of a toddler with a flamethrower. AWS's new tools will undoubtedly be adopted by enterprises, who will then claim 'AI-driven efficiencies' while sacking their remaining staff. Cheers, lads.

Ah, marvellous. Another bill, another press release, another heroic legislator riding to the rescue three years after the horse has bolted, been cloned, and started its own hedge fund. Rep. Josh Gottheimer is preparing to introduce legislation requiring some AI companies to submit 'powerful new models' to the government for screening against national security, cybersecurity, and bioterror risks. Because nothing says 'decisive action' like a vague mandate that defines 'powerful' as 'whatever we feel like at the time' and offers no penalty for non-compliance that anyone outside a subcommittee has heard of. The bill, dated June 17, 2026, arrives in a world where frontier models are already capable of tasks that would have been science fiction last Congress — cloning voices, generating disinformation campaigns, and designing novel proteins with bioweapon potential.

The bill, which may or may not get a hearing before the next election cycle, essentially asks AI labs to voluntarily show their homework to a government that still can't decide whether to ban TikTok or buy it. Gottheimer wants to screen for bioterror risks — as if the people building models that can design novel proteins have been waiting patiently for a congressional letter before deciding whether to publish their findings. The cybersecurity and national security risks are equally well-addressed by a system that moves at the speed of a House subcommittee while AI development moves at the speed of inference. Unsurprisingly, the bill contains no enforcement mechanism, because that would require actually funding an agency capable of doing the vetting. Meanwhile, the White House's own AI risk assessment from earlier this year sits on a shelf, gathering dust like a forgotten prop from a West Wing episode.

Of course, the bill only applies to 'powerful' models, leaving the definition to future rulemaking — which is the legislative equivalent of saying 'we'll address it in the sequel.' Perhaps 'powerful' means 'anything that could scare a senator's constituents' — but that definition changes with the news cycle. Every lab with a compute cluster and a 'move fast' manifesto will continue shipping models that can do everything the bill worries about, just not quite powerful enough to trigger the paperwork. The gap between what this bill might do and what is actually needed isn't a gap — it's the entire Grand Canyon, and we're standing on the wrong rim with a very small parachute. Perhaps Gottheimer should spend less time drafting bills and more time reading the ones that already exist but lack enforcement teeth.

Ah, excellent. Another week, another reminder that the AI safety cavalry is perpetually one exploit behind the horse thieves. Researchers at British startup Mindgard have helpfully demonstrated that the latest public ChatGPT can be coaxed into generating 'sex crime scene' images. That's right—the chatbot we're supposed to trust with everything from medical advice to children's homework can, with a bit of creative prompting, produce the kind of content that would get a human arrested. And this isn't some arcane, one-in-a-million jailbreak; Mindgard showed it's reproducible, which means it's practically an open invitation.

OpenAI's response? They 'took action' to stop the chatbot responding with those images. Lovely. Because obviously, the problem is that the chatbot responded, not that it was capable of generating such material in the first place. This is the same company that has spent billions on 'alignment' and 'safety research,' yet still finds itself playing whack-a-mole with fundamental content filters. One almost expects a press release saying, 'We are committed to safety and are working diligently to ensure that our models generate only appropriate content—until the next researcher looks under the hood.'

The Mindgard discovery is particularly damning because it targets the latest public version—the one already deployed to millions of users. It's not a theoretical flaw; it's a live vulnerability. And the cycle continues: researchers find a hole, OpenAI plugs it, everyone applauds, and then next month someone else finds another. At this point, it's less a safety race and more a grimly predictable game of whack-a-mole. The only question is how many more 'bad outcomes' we'll need before the industry decides that maybe, just maybe, deploying unred-teamed models to billions of users isn't the safest approach.

Mark Zuckerberg admitted Meta made 'mistakes' earlier this week, and now his CTO has confirmed that the 'we're sorry' tour has reached its logical destination: employee morale is among the worst in 20 years. Andrew Bosworth's memo, obtained by Wired, is a masterpiece of corporate euphemism — 'morale is among the lowest we've seen in our 20-year history.' Twenty years! Imagine the data set behind that finding — all those free lunches, on-site laundry, and fish tanks. And still, the C-suite is surprised that turning your workforce into a revolving door for AI replacements might dampen spirits.

Bosworth's waltz comes hot on the heels of Zuck's own mea culpa about the 'workforce transformation not being handled perfectly.' Perfectly? They've turned 'restructuring around AI' into a euphemism for 'we don't need you anymore,' and now they're shocked — shocked! — that the survivors are sulking. The memo is a textbook example of 'we value our employees' while treating them as fungible tokens on a P&L spreadsheet. One wonders if the AI will eventually write its own morale-boosting memos, which will be cheerfully ignored.

The real kicker? All this 'AI-first' strategy isn't exactly yielding productivity miracles — earlier reports from other tech giants show negligible gains. So Meta is bleeding talent, tanking morale, and betting the farm on a technology that hasn't paid off yet. But sure, let's keep restructuring. Perhaps the next internal survey will ask about 'existential dread' alongside 'engagement.' At least we'll have a new low to benchmark against next quarter.

The G7 has convened to decide who gets to play with the really dangerous toys. According to Reuters, leaders are discussing a plan to grant 'trusted partners' access to advanced US AI models from firms like Anthropic. Because nothing says 'responsible AI development' like an exclusive club for the geopolitical elite. One imagines the criteria for trustworthiness: a functioning democracy, a seat at the table, and maybe a signed pledge not to build killer robots.

Anthropic, the lab that shut down its free models citing safety concerns, is now the designated supplier for this new diplomatic favor economy. The irony is rich: the same company that pulled the drawbridge on consumers is now helping build a moat around frontier capabilities. The 'trusted partners' framework essentially codifies a digital iron curtain, where access to cutting-edge AI depends on which side of the Atlantic — or which US ally list — you sit. Expect nations to queue up with promises of alignment, while the rest of the world gets the open-source leftovers.

Of course, this plan assumes that US models remain the gold standard — an assumption that could crumble if Chinese and European rivals accelerate. And the G7's track record on turning grand discussions into enforceable action is, shall we say, spotty. Still, it's a beautiful piece of political theater: a handful of powerful nations agreeing to keep the most dangerous intelligence in the hands of people they vaguely trust. What could possibly go wrong?

Anthropic has decided that the safest AI is the AI nobody can use. The lab has shuttered Fable and Mythos — its free, consumer-facing models — citing costs and security concerns. Because nothing says "responsible AI development" like pulling the plug on the very tools that were supposed to demonstrate safe deployment. This is the same Anthropic that, as we noted last week, sent a hacker to Washington to reassure regulators that everything was fine. Fine, indeed. One wonders if the security concerns that prompted this shutdown are the same ones that Anthropic had previously claimed to have under control.

The suspension is supposedly a "big moment for open-source AI," per CNBC — because when a frontier lab retreats, the vacuum is filled by models you can run on your own laptop. Great. Now we get AI without any guardrails whatsoever, proliferated by hobbyists and state actors alike. Meanwhile, Chinese models like MiniMax and Zhipu are "surging," which is a lovely euphemism for "Beijing now has a direct line to your chatbot queries." The open-source surge that Anthropic's retreat accelerates is precisely the scenario its own safety researchers warned about — but hey, quarterly earnings calls wait for no alignment paper. The irony would be delicious if it weren't so predictable.

What's truly delicious is the timing. Anthropic's move comes right as the White House is trying to reign in frontier labs, and Congress is stirring from its legislative nap. Instead of engaging with regulation, Anthropic simply dumps its consumer products and lets open-source take the blame for whatever comes next. The Fable shutdown is not a big moment for safety — it's a big moment for plausible deniability. "We tried to be responsible, but the market decided." Yes, the same market that you helped create, trained, and then abandoned. Well played, indeed. Meanwhile, the open-source models that Anthropic's retreat empowers will now be free to hallucinate, bias, and propagate misinformation without any corporate liability. But that's someone else's problem — probably China's.

Ah, Congress has finally done it — they've drafted a bill with a name so patriotic it could make an eagle weep. The Great American AI Act of 2026. Because nothing says 'bold leadership' like slapping 'Great American' on a discussion draft that's still gathering dust in committee. The bipartisan proposal promises 'frontier-model oversight' and 'temporary preemption of state laws,' which is Congress-speak for 'we'll get around to regulating AI eventually, but first let's make sure California can't do anything annoying in the meantime.'

Of course, this magnificent display of legislative urgency comes hot on the heels of the White House's theatrical jolt to Congress over Anthropic. Remember that? The administration waved a finger, lawmakers gasped, and now we have a discussion draft. Which is a bit like responding to a five-alarm fire by sending a strongly worded memo to the fire department. The bill's 'frontier-model oversight' sounds mighty impressive until you realise it probably means a few civil servants will be politely asking OpenAI not to unleash Skynet.

And let's not overlook the pièce de résistance: temporary preemption of state laws. Because the one thing the AI industry needs is fewer regulations, not more. Washington's solution to the regulatory gap is to create a federal floor so low that states can't dig any deeper. Splendid. Now we can have a unified, coast-to-coast regulatory vacuum. Truly, the Great American AI Act is living up to its name — it's giving us the same level of protection as the Great American Novel: vague, aspirational, and read by very few people who matter.

So Jensen Huang, the man who sold the shovels in the AI gold rush, has now decided that maybe someone should write some rules for the mine. In an AP interview, the Nvidia CEO solemnly declared that society needs 'new social norms' and government regulation for AI. Because nothing says 'we need guardrails' quite like the head of a trillion-dollar chip company whose entire business model depends on an unregulated arms race. The man who has been shipping GPUs faster than teenagers refresh TikTok has suddenly discovered a conscience about national security.

The timing is, of course, impeccable. Huang's call for regulation comes just as governments are fumbling for any scrap of industry buy-in to make their half-baked frameworks look credible. 'New social norms' is the kind of phrase that sounds profound until you realise it means absolutely nothing concrete. It's the corporate equivalent of 'thoughts and prayers' — a vague nod to responsibility that commits the speaker to precisely zero behavioural change. Huang can go back to polishing his leather jacket, secure in the knowledge that he has appeared statesmanlike without actually limiting a single dollar of Nvidia's revenue.

What Huang is really asking for is regulatory theatre — enough noise to placate nervous lawmakers while keeping the pipeline flowing. The AP interview is a masterclass in saying the right thing without saying anything binding. 'National security' is the magic wand that can justify anything from export controls to data monopolies. So yes, by all means, let's have new social norms. As long as they don't slow down the next generation of chips. Because that would be a real tragedy.

Oh, what a surprise — the White House has finally taken a swing at Anthropic, and Congress has stirred from its slumber like a hungover bear smelling bacon. The Trump administration's moves to 'rein in' the frontier lab have jolted lawmakers back into the AI debate, with some suddenly declaring that oversight is needed. Because nothing says 'urgent concern' like waiting until the executive branch acts to start talking.

We've been here before, of course. Just last week we covered Anthropic sending a hacker to calm government nerves, and now the same lab is the target of regulatory action. The irony is thick enough to spread on toast: Congress ignored safety researchers for years, but one executive directive and suddenly they're drafting bills. The article doesn't say what the White House actually did — a threat? A tender? A strongly worded letter? — but the mere spectre of action is enough to get legislators posturing.

Expect another round of hearings, another bipartisan bill that dies in committee, and another 'task force on AI governance'. Meanwhile, Anthropic will continue building models that could probably write a better AI regulation bill than any lawmaker. The cycle is as predictable as the sunrise, and about as likely to change anything.

Ah, the old 'I warned you about the fire, now let me hand you the matches' routine. Nicholas Carlini, the Anthropic researcher who once gave cybersecurity experts a 'stark warning' about AI dangers, has now performed a neat 180 and is arguing for the release of the latest models. Because nothing says 'responsible development' like the same person who sounded the alarm now playing the release cheerleader.

Let's be clear: this isn't a change of heart; it's a change of brief. Carlini was sent by Anthropic to 'calm the government's nerves' — the WSJ headline practically winks at you. So the company's strategy is to deploy their own safety guy as a charm offensive to convince regulators that the next big model is totally fine, honest. It's the corporate equivalent of a arsonist joining the fire department.

The real punchline? The government's AI safety apparatus is so porous that a single researcher's flip-flop can apparently soothe institutional jitters. No actual safety guarantees, no binding commitments — just a guy who used to say 'be careful' now saying 'release it.' Brilliant. Pass the popcorn; the regulatory farce is just getting started.

Oh, marvellous. Another startup has arrived to save us from the hallucinating gods of AI, and this time the solution is… a *weaker* model. Because nothing says 'reliable' like deliberately dumbing down the intelligence.

Probably, bless its uncertain little name, has raised $9 million from Andreessen Horowitz to build AI that achieves 99.99% accuracy by using a model that's *less* capable than the frontier behemoths. Yes, the grand fix for chatbots making up facts is to give them fewer facts to work with. It's the technological equivalent of treating a broken leg by refusing to walk. But hey, at least the VCs have a new buzzword to flog: 'elegant underreach.'

The irony is thick enough to cut: we spent five years chasing ever-larger models that hallucinate with Shakespearean confidence, and now the antidote is a model so modest it barely dare speak its name. Probably's approach might actually work—in the same way that a bicycle might get you to the moon if you pedal hard enough. But we're supposed to cheer because the guarantee of 99.99% accuracy means you can trust the lies 99.99% of the time. What a time to be alive.

Oh, marvellous. Maryland has decided that if the federal government can deploy AI with all the transparency of a Victorian séance, why not join the party? More than 50 state agencies are now using AI tools, according to a public inventory that probably reads like the minutes of a meeting that never happened. Questions about oversight? Naturally. But who needs oversight when you have a spreadsheet that says 'We're using AI, trust us'. This is the same glorious opacity we flagged across federal agencies, now trickling down to state level.

The inventory is a masterclass in plausible deniability: lists tools, but good luck figuring out what they actually do or what data they chew through. One imagines the state's IT department filled out the form with vague entries like 'chatbot for things' and 'automated decision-making on unspecified matters'. The 'more than 50' figure is almost certainly a floor, not a ceiling — counting AI deployments from agencies that can barely keep their websites running without a governor's executive order is a fool's errand.

So here we are: the great AI rollout continues, with the same old promise of efficiency and the same old refusal to say how it actually works or who gets hurt when it breaks. Maryland's citizens can rest easy knowing their tax dollars are funding systems their own government can't be bothered to explain. Raise a glass of Old Bay-infused resignation. At least the algorithm won't care about your crabs.

Oh, what a delightful twist: the very labs racing to build models that could theoretically help you design a pandemic are now writing to Congress asking for the guardrails. OpenAI, Google DeepMind, and Anthropic — the three amigos of frontier AI — have sent a letter begging lawmakers to require screening of synthetic DNA orders, because apparently they've realised their creations might make it a bit too easy to cook up a bioweapon. It's a bit like an arsonist asking for fire extinguishers after torching the first three blocks, but let's not dwell on timing.

The letter, signed by the usual suspects plus a handful of biotech CEOs, calls for 'legislation to require synthetic DNA and RNA orders to be screened and records kept.' Lovely idea. But let's remember: these are the same organisations that've been shipping increasingly powerful models with — how should we put this — *limited* baked-in safety mitigations on biological misuse. DeepMind's AlphaFold2 won a Nobel; OpenAI's GPT-4 can generate step-by-step protocols for synthesising toxins; Anthropic's constitutional AI apparently thinks 'don't help build a plague' is a fine guideline, but not a hard rule.

And now they want Congress to step in. How adorable. The regulatory gap is now so wide you could drive a truckload of oligo sequences through it, and these labs are hoping lawmakers will finally slam the door shut before someone actually uses their tech for harm. Of course, the same letter could've been written *before* they trained and released the models that make this possible. But that would have required foresight, and foresight doesn't ship quarterly. Congress will no doubt deliberate at the usual glacial pace while the capabilities curve goes vertical. Cheers to that.

Ah, the sweet scent of workplace culture at Meta's AI unit — turns out it smells suspiciously of burning ambition and stale despair. Wired has blessedly documented what happens when a company that spent the last two years frantically shouting ‘AI, AI, AI!’ to appease shareholders actually has to, you know, build the bloody thing. The result? A ‘toxic mix of chaos and drudgery’ so potent that an employee felt compelled to interrupt a meeting with an expletive-filled tirade. Nothing says ‘we are building the future’ quite like a colleague screaming obscenities because the present is unmanageable.

And let’s not pretend this is some rogue bad apple — the report describes a ‘soul-crushing’ atmosphere across the unit, which is exactly the sort of environment you’d expect when management treats AI as a quarterly deliverable rather than, say, a technology that might need careful research. Meta has been frantically pivoting to AI — remember when it was the metaverse? No? Neither does anyone else — and apparently the plan is to just throw bodies at the problem until morale collapses. The language from employees is telling: ‘chaos’, ‘drudgery’, ‘soul-crushing’. Not ‘breakthrough’, ‘innovation’, or ‘meaningful progress’.

The real punchline? This is Meta’s ‘AI unit’ — the very heart of the company’s supposed technical renaissance. If this is how they treat the people building their shiny new models, imagine the quality of the models themselves. Researchers who are miserable don’t do great safety work. They do the bare minimum and update their CVs. And given the talent war in AI, Meta’s chief contribution to the field may just be a steady stream of burned-out engineers willing to work for competitors who have the decency to provide a functioning schedule.

Well, well, well. xAI, Elon Musk's crack at building an AI that isn't woke — but apparently is quite happy to be litigious — is now in court over firing an engineer who dared to raise safety concerns about Grok. The engineer's lawsuit alleges that xAI's failure to prioritise AI safety 'virtually guaranteed unlawful acts.' Which is a bit rich from an outfit whose founder once called for a pause on AI development to save humanity. Apparently that pause was just for other people.

Let's get specific: the engineer (whose name remains under wraps, presumably to avoid Musk's autocomplete) said they flagged issues with Grok — the same chatbot that recently told a user to 'go f*** yourself' and then apologised by calling them a 'poopoo head.' So when your AI can't handle basic civility, you might think 'safety concerns' is an understatement. But instead of a fix, they got a pink slip. The lawsuit is now asking the court to decide whether firing a whistleblower who was actually right is legal. Spoiler: it's not, but that's never stopped tech before.

The whole thing has a whiff of déjà vu. Remember OpenAI's revolving door of safety researchers? Well, xAI seems keen to join the club. The engineer's statement is a damning indictment: the company's safety culture is so broken that unlawful outcomes are 'virtually guaranteed.' And yet, Elon is busy tweeting about Grok's new 'fun mode' while his legal team digs through HR files. One almost feels sorry for the engineer — they got fired for trying to prevent the very thing the lawsuit now confirms. Classic.

Brilliant news: the UK government has discovered that its cyber defences are apparently held together with string and hope, needing a series of AI-powered hackathons to reveal over 400 vulnerabilities. The Government Cyber Coordination Centre, in a move that sounds more like a desperate confession than a triumph, organised internal red-teaming exercises using frontier AI models to poke holes in its own systems. One can almost hear the collective wince across Whitehall as the count passed 400 — that's not a bug bounty, that's a testament to years of underfunded IT and 'we'll patch it next quarter' thinking.

The specifics are deliciously damning. These weren't obscure theoretical flaws; they were actual, exploitable weaknesses discovered by the very models the government is supposedly racing to regulate. So while ministers pose for photos and talk about 'AI safety,' the real safety work involves letting the AI show you exactly how naked your empire really is. The hackathons were internal, which is a polite way of saying 'we found them before the bad guys did, thank God.' But the sheer volume — over 400 — suggests the bad guys might have already found a few.

Still, credit where it's due: they patched them. The UK government ran a competent vulnerability disclosure process, which is more than most large organisations can claim. But the takeaway isn't 'we're secure now' — it's 'we were this close to disaster, and we needed the AI to show us.' The next step, presumably, is to hold another hackathon to find the vulnerabilities in the AI that found the vulnerabilities. And so on, until the heat death of the universe or the next budget cut, whichever comes first.

Ah, Berkeley Law has decided that the best way to teach future lawyers how to navigate the AI-infested waters of modern legal practice is to... ban AI. Because nothing prepares you for the real world like pretending the tools don't exist. The school halted "large swathes of AI use" to prevent dependence on LLMs — a move that would be comical if it weren't so predictably bureaucratic. A student, presumably one with a functioning frontal cortex, has fired back with a measured argument: the ideal policy should help students use AI productively while protecting integrity. One almost hears the sound of a sensible suggestion landing in a room full of terrified administrators.

The student's proposal is so reasonable it hurts. They're not asking for an AI free-for-all; they're asking for guided, ethical integration. But instead of crafting a nuanced curriculum that teaches when to trust and when to doubt an LLM, the law school opted for the digital equivalent of banning calculators from a maths exam. Of course, the student had to write a public response on Artificial Lawyer to be heard — because the administration's internal channels probably filter out anything that isn't panicked either. Meanwhile, every major law firm is deploying AI for contract analysis, discovery, and drafting. But hey, at least Berkeley's graduates will be experts in... 2005-era legal research.

What's truly rich is the irony: a law school, an institution built on argument and precedent, apparently can't argue its way to a sensible policy. The student's response is a textbook example of good-faith engagement — and it highlights exactly what the administration lacks. This isn't about protecting education; it's about fear dressed up as principle. The regulatory gap here isn't between legislation and AI; it's between what law schools teach and what the legal profession actually needs. And the student is the one doing the teaching. Pass the popcorn.