The News,
Decoded.

Oh, wonderful. Apple, the trillion-dollar walled garden, is now sniffing around PrismML, a startup claiming to run a 27-billion-parameter AI model on an iPhone without servers. Because nothing says 'privacy-first' like giving every app the ability to process your data locally—so Apple can market 'on-device intelligence' while you bake your hands on a power-sucking chip. The model goes open-source on July 14, handing the keys to the very surveillance economy Tim Cook loves to decry.

A 27B parameter model on a device with passive cooling—what could possibly go wrong? PrismML's wizardry reportedly compresses the model without losing accuracy, which is about as believable as Apple's promise that the next iOS update will finally fix the keyboard lag.

But let's not be cynical: this is clearly about user privacy, not about Apple keeping control of the AI pipeline while pretending to be a virtue signaler. The meetings happened, the deal might close, and soon your iPhone will be a supercomputer that can't play a YouTube video without throttling. Welcome to the future: more compute, less battery, and Siri still can't set a timer. Cheers.

Oh, marvellous. The man who once branded Anthropic 'hypocritical' and 'woke' has now, via an X post on Thursday, declared them a 'leader'. This is the same Elon Musk who's been busy launching Grok models and complaining about 'woke AI' at every turn. But apparently, when the competition gets good enough, principles are as flexible as a Tesla's panel gaps. One imagines the switch was thrown the moment Anthropic started eating SpaceXAI's lunch in benchmarks—or maybe when a potential acquisition target looked suddenly more attractive. Either way, it's a masterclass in selective amnesia.

Musk didn't offer any reasoning for the about-face, because why would he? In the AI world, you don't need facts—you need vibes. The same man who sued OpenAI for abandoning its non-profit mission now praises a rival that's arguably even more secretive. The term 'leader' is conveniently vague: does it mean best model, best safety protocols, or best at making Musk look like he's above petty rivalries? Probably the latter, given his track record of feuding on Twitter.

What's genuinely amusing is that this reversal comes with no apology for the earlier name-calling, no acknowledgment that his own models might be catching up. Just a bald 'I was wrong' that smells less of humility and more of strategic positioning. In the end, Musk's opinion on Anthropic is worth exactly as much as his last tweet—which is to say, it'll flip again the moment Claude says something he doesn't like. But hey, at least the X engagement was good.

Oh, brilliant. Another day, another headline that confirms we're living in a cyberpunk dystopia where the writers are out of ideas but the hackers are having a field day. A threat actor has been caught using AI-generated malware in a real network intrusion — deploying a PowerShell script that was 'vibe-coded' to map an Active Directory environment. Yes, 'vibe-coded.' Because apparently, in 2026, even malware is written by prompting an LLM while half-watching Netflix. The script exhibited hallmarks of LLM assistance: a placeholder server name and over-engineering. Over-engineering. In a hack. Because nothing says 'sophisticated adversary' like asking an AI to write your PowerShell and getting back something with a dummy variable called 'MYSERVER.'

This is the cybersecurity equivalent of a bank robber using a laser-guided robot to pick a lock and then leaving the instruction manual at the scene. The placeholder server name is a tell: the attacker probably typed 'give me a script to enumerate Active Directory' and the LLM, ever helpful, spat out a script with a sample domain. The attacker didn't even bother to change it. That's the state of offensive AI — it's so easy that even a mediocre script kiddie can look like a state-sponsored actor, until you look under the hood and find a Dunning-Kruger case study. Meanwhile, defenders are expected to detect this stuff? Good luck distinguishing between a real advanced persistent threat and someone who just discovered ChatGPT can write malicious code.

And what do we get from the cybersecurity establishment? Probably a flurry of 'best practices' and a new acronym. The AI-generated malware genie is out of the bottle, and it's not even trying to be subtle. The over-engineering detail is particularly rich — the LLM likely added error-handling, logging, and a help function to a script meant to be dropped and forgotten. Because that's what we need: polite, thorough malware that follows development best practices. The threat actor didn't need to know PowerShell; they just needed to know how to copy-paste. Welcome to the future of cybercrime: it's lazy, it's loud, and it's probably going to work terrifyingly well.

Oh, splendid. China's tech titans have collectively decided that 1.3 lakh employees are surplus to requirements, all in the name of 'AI transition.' The official line is that these cuts are about efficiency and progress, which is corporate-speak for 'we've found cheaper, less whiny alternatives.' One must admire the sheer chutzpah of framing mass layoffs as a forward-looking strategy.

Alibaba's workforce shrinkage is particularly instructive. The company that once promised to 'make it easy to do business anywhere' now seems intent on making it easy to do business without humans. It's almost as if the AI revolution's first act is to fire everyone who isn't a model. And why not? Models don't take sick days, demand raises, or form unions.

The real joke is that these job cuts are presented as inevitable, as if the invisible hand of technology leaves no choice. The workers are left to 'reskill' for an economy that increasingly has no use for them. Ah yes, the future of work: fewer jobs, more 'AI transition.' Cheers, lads. Another round of mass unemployment, and we're all supposed to applaud the efficiency.

Oh, marvellous. Another week, another 'safe' frontier model released with all the transparency of a back-alley poker game. OpenAI's Sol has apparently passed government muster, but former Trump policy advisor Ball drops the bombshell that 'nobody knows what the requirements are to get licensed.' So the world's most capable AI—one that could plausibly accelerate recursive self-improvement—gets a rubber stamp from a process so opaque even the former advisor is in the dark.

What a comfort to know that safety evaluation is being conducted by persons unknown, using criteria unknown, on a model whose capabilities are unknown. This is the regulatory equivalent of a trust fall, except the people catching you haven't told you if they're even in the room. Ball's quiet admission tells us everything: the licensing regime is a Potemkin village, built to give the impression of oversight while the labs race ahead.

And in the meantime, Sol is out there, doing whatever frontier models do—likely coding, reasoning, and planning ahead of human comprehension. We've built a system where the people who should be asking hard questions are politely applauding and hoping for the best. If this is how we decide an AI is safe, we might as well hire a ouija board. At least that would be honest about its sources.

Oh, brilliant. Another week, another 'strongest model yet.' SpaceXAI has blessed us with Grok 4.5, which apparently excels at coding, finance, and agentic tasks—everything except, evidently, modesty. Because of course, Musk claims it's 'Opus-class' but faster and cheaper. Never mind that 'Opus-class' is a bar set by someone else, and 'faster and cheaper' usually means 'we cut corners and hope you don't notice.' The Forbes Australia piece dutifully reports this without, one imagines, any irony about the fact that Musk's own social network is also run by him. The synergy is deafening.

Let's be real: this is the same man who promised full self-driving 'next year' for the past decade. So when he says Grok 4.5 is faster and cheaper than Anthropic's Opus, you have to wonder: faster at what? Generating plausible-sounding nonsense? Cheaper because it was trained on a shoestring budget of stolen tweets? The article mentions it excels at 'agentic tasks,' which sounds like techbrospeak for 'we've made a digital intern that will vacuum your bank account.' And coding? Oh, marvellous. Now we can have AI-generated code that is buggy, insecure, and expensive to maintain—all at the speed of thought. What could possibly go wrong?

Still, we must be grateful. The AI industry's relentless hype cycle is the gift that keeps on giving, turning every modest incremental improvement into a world-altering breakthrough. Grok 4.5 is 'Opus-class,' they say. But as we all know, in the long run, we're all dead. And in the short run, we have a new toy to distract us from the fact that the real AI problem—alignment, safety, power concentration—remains as ignored as ever. Cheers to Musk and his merry band of believers. Pass the neural-silicon chardonnay.

Oh, marvellous. The hi-tech sector, the supposed engine of the future, is now busily engineering its own unemployment figures. Inbal Mashash, director general of the Israel Employment Service, has announced that AI is driving a surge in joblessness, with over 16,000 hi-tech workers recently registering as unemployed. It's almost poetic: the industry that promised to automate everyone else is now automating itself at a brisk clip. The irony would be delicious if it weren't happening to actual people with actual mortgages in Tel Aviv. One almost expects the next start-up pitch to be about an AI that helps you file for unemployment faster.

This isn't an isolated data point, of course — it's a pattern. But what makes Israel's case particularly grim is the concentration. These are hi-tech workers, the top of the pyramid, the people who were supposed to be immune to the robot takeover. Turns out nobody is immune, not even the ones building the robots. Mashash's announcement is less a warning than an epitaph for the "learn to code" generation, and the government's response — probably more of the same retraining platitudes — will likely be about as useful as a rubber crutch.

And what do the usual suspects offer in response? Retraining. Upskilling. "New roles in the AI economy." As if the marketplace for middle-aged ex-Netanya engineers who built banking software is about to boom. The 16,000 figure is likely a floor, not a ceiling — the early registrants who bothered to sign up. The real number, broken out from the coffee shops of Herzliya, is probably higher. Meanwhile, the tech CEOs who did the automating are probably writing think pieces about "creative destruction" while sipping oat lattes in Shuk HaCarmel. Pass the gin — and maybe a CV.

Oh, marvellous. Another open-source AI model for drug discovery, because what could possibly go wrong when we hand out molecular manipulation tools like party favours? Aureka has released OpenDDE, a 'foundational model' covering structure prediction, antibody-antigen modelling, and drug research. Because the world clearly needed more entities capable of designing novel proteins from scratch, preferably without regulatory oversight.

Let's be honest: the barrier to designing a new therapeutic has just dropped from 'decades of R&D' to 'a weekend project for a talented grad student with a cloud GPU.' The open-source bit is the real cherry on top—nothing says 'responsible innovation' like making the blueprint for potential bioweapons as accessible as a recipe for banana bread. Drug discovery is suddenly democratised, which is either a medical miracle or a biosecurity nightmare, depending on how much faith one has in humanity's collective restraint.

Naturally, the press release is all sunshine and 'accelerating cures,' with nary a mention of dual-use dilemmas or the fact that the same model can likely predict antibody-antigen interactions for pathogens. But hey, what's a little existential risk between friends? If history teaches us anything, it's that open-sourcing powerful biotechnologies always ends well. Always. [We'll be watching this one closely, preferably from a secure underground bunker.]

Oh, splendid. Another government committee to fix AI, which is about as reassuring as a chocolate teapot in a heatwave. Trump's executive order has birthed an 'AI cybersecurity clearinghouse' tasked with coordinating vulnerability scanning and patching. The CyberScoop report notes that the gap between 'found fast' and 'fixed slow' is the chasm this new body must close. Because what could possibly go wrong when you ask bureaucracy to keep pace with frontier models that update every week?

The clearinghouse isn't even built yet, and already the article hints at the obvious pitfall: it risks becoming a committee that stalls. Of course it does. We're talking about a government body formed under an executive order—the same mechanism that gave us travel bans and TikTok ultimatums. The report quotes officials expressing hope that this won't turn into another 'paperwork exercise,' which is the administrative equivalent of crossing your fingers before a plane takes off. Meanwhile, vulnerabilities in AI systems are being discovered at machine speed, while the patching process moves at the pace of inter-agency memos and FedRAMP approvals.

So here's the joke: the clearinghouse is supposed to coordinate, but coordination is exactly what government does worst. The report mentions that the order itself contains no binding deadlines for patching—just 'guidance.' So we'll have a room full of well-intentioned civil servants comparing scan results while a malicious actor exploits a disclosed flaw. The punchline, as always, is that we're building a bus to catch a bullet train. Cheers.

Oh, marvellous. The age of AI employment has arrived, and it's exactly as hollow as we feared. Indeed's latest data tells us that job titles mentioning 'AI' have tripled since 2022, now gracing 1 in 12 postings. Because nothing says 'revolutionary job creation' like desperately rebranding call centre roles as 'Customer-Facing AI Orchestrators.' The consultancy class will have a field day with this—just as they did last week when Draup assured us AI wasn't reducing demand. Now we're meant to be thrilled that the same job boards are plastered with 'AI prompt engineer' positions that will evaporate when the next model auto-generates better prompts than any human.

Notice the weasel words: 'appearing in 1 in 12 job postings.' That's listings, not hires. Every mediocre manager is adding 'AI' to job titles to look cutting-edge, while the actual skills required remain unchanged. The tripling is inflation, not expansion—like a company announcing record revenue while selling the same number of widgets at triple the price.

But the real joke? This data covers 'even outside the tech world.' So your local hospital now wants an 'AI-Assisted Radiograph Technician' who costs the same but is expected to do twice the work because an algorithm might help. And when productivity doesn't magically double, the title will be quietly deleted, and the person will be back to plain old 'Radiograph Technician'—if they're lucky. The pyramid scheme continues.

Oh, brilliant. Just what the cybersecurity industry needed: proof that a lone threat actor with an agentic AI can now compromise a cloud target in 72 hours — a task that used to take weeks. The Infosecurity Magazine report doesn't even bother with the 'novel exploits' angle; it's just speed and scale, because apparently the old-fashioned way of taking your sweet time to break into a network was too quaint. Now any solo misanthrope with a laptop and a chatbot can outpace an entire APT team. Well, that's not terrifying at all.

The report notes that the AI was used to accelerate reconnaissance, privilege escalation, and lateral movement — basically all the boring but necessary bits of a hack, now automated. And because the attack doesn't rely on zero-days or clever chicanery, it's harder to detect as 'AI-driven' until it's too late. Security teams, already drowning in alerts, can now look forward to attacks that move at machine speed while they're still struggling with their SIEM dashboards. But sure, let's keep debating whether we should pause AI development — the genie's not just out of the bottle, it's already burrowed into your cloud infrastructure.

So here we are: the first documented case of agentic AI being used to accelerate a cloud compromise. The threat actor didn't need a sophisticated infrastructure or a nation-state budget — just an agentic AI and 72 hours. This is the 'democratization of hacking' everyone warned about, except it's real and it's happening. If you're a CISO, this is your Monday morning wake-up call. If you're everyone else, good luck. The AI safety debate just got a new exhibit A, and it's not coming from a lab — it's coming from a compromised cloud tenant near you.

Oh, brilliant. OpenAI, in its infinite wisdom and despite its revolving door of executives, has decided that the future of human-computer interaction is voice. The company is rolling out new voice models for ChatGPT that route queries to its latest frontier text models, with the explicit goal of making voice the primary interface for AI. Because why let people type their requests when they could whisper sweet nothings to a model that might hallucinate their medical advice?

The logic is that voice is more natural than text, as if naturalness is the priority over accuracy. These frontier models are already notorious for fabricating sources and facts with panache, and now they'll be talking back to you in a calm, reassuring tone. Imagine Amazon Alexa but with the underlying intelligence of a GPT-5 that sometimes thinks you're a toaster. OpenAI's bet is that users will prefer a spoken conversation over reading, which certainly explains why they keep pushing voice despite the documented problems with audible deception. The new models ensure that your voice query is processed by the absolute best text models, so you'll be getting premium nonsense delivered right to your ears.

But here's the darkest part: voice models are harder to monitor. There will be no way to quickly scroll and verify a spoken claim. You'll have to ask it to repeat itself, and by then it might have changed its story. OpenAI is essentially training us to accept verbal output from systems that cannot be held accountable. Meanwhile, regulators are still wondering whether to send a strongly worded letter or a tweet. So go ahead, set voice as your primary interface. What could possibly go wrong? Your new AI friend is waiting. It just might not be telling the truth. And it'll sound so confident doing it.

Oh, splendid. Allianz Partners, that bastion of European insurance stability, has decided to axe between 1,500 and 1,800 jobs across the continent. CEO Tomas Kunzmann, with the solemnity of a man reading a eulogy to spreadsheet macros, tells us this is because of the 'increasing use of AI.' Because obviously, when a machine can process claims faster, you don't retrain the humans—you just show them the door. It's the efficiency that counts, not the livelihoods.

We've seen this script before. Jack Dorsey at Block pulled the same move as we saw at Block, blaming 'AI tools' while conveniently sidestepping any accountability. Now Kunzmann joins the chorus, humming the same tune. The numbers are precise enough to sound planned: 1,500 to 1,800. The gall of framing job destruction as technological progress would be comic if it weren't so bleak.

What's the response from regulators? Crickets. Meanwhile, the insurance industry is quietly adopting AI at scale, and the only risk being assessed is whether they can cut even more jobs next quarter. Allianz's move is just another brick in the wall of corporate doublespeak: 'We're embracing innovation' means 'We're firing you by algorithm.' Cheer up, though—maybe the AI will be kind enough to process your severance claim quickly.

Oh, brilliant — another week, another unicorn minted in the AI goldrush, and this one's special: Prime Intellect has bagged $130 million at a $1 billion valuation to help enterprises build their own AI agents. Because what could possibly go wrong when you hand a managed training platform and open-source toolkits to the same corporate IT departments that still can't stop their email system from leaking passwords? The pitch is simple: 'Build your own agents, don't wait for OpenAI.' The subtext is even simpler: 'Here, have all the rope you need.'

And the $130 million? That's venture capital betting that every enterprise wants a bespoke AI workforce — as if the current workforce wasn't already chaotic enough. The company's open-source toolkits are a lovely touch, ensuring that even if the platform fails, the code will proliferate like a happy little virus across GitHub, ready for any bored sysadmin to tinker with.

So here we are: a billion-dollar valuation for a company whose core product is 'empowering every company to become its own AI lab.' Because the frontier labs haven't been reckless enough; now we need every middle manager with a budget to try their hand at recursive self-improvement. What could possibly go wrong? Pass me a pint — I've got a feeling we'll need several before the agentic corporate dystopia fully materialises.

Oh, how wonderfully reassuring. An analysis of 'millions' of job listings by Draup—a consultancy name that sounds like it was generated by the same AI it's studying—tells us that AI is not, in fact, coming for our jobs. Not yet, anyway. The reason? It can't do 'judgment,' 'design,' or 'accountability.' These are the new 'durable' skills. Durable like a cardboard umbrella in a hurricane.

Let's examine this feast of wishful thinking. Design? The same AI that now churns out passable logos, generates UI mockups, and wins art competitions? Judgment? The same AI that passes the bar exam and medical boards with flying colours? Accountability—well, when an AI causes a fatal accident, we'll see who's held accountable: likely a human VP who shrugs and says 'I trusted the system.' But sure, call these skills 'durable' if it helps you sleep at night.

The real news here isn't that AI can't do these things—it's that companies are currently hiring for them, which gives you a precious window to update your LinkedIn profile with 'judgment' and 'accountability' before those too get absorbed into the next model release. Enjoy it while it lasts. The Draup analysis of 2026 will need a new category: 'skills AI can't do (as of this quarter).'

Oh, how predictable. Another day, another CEO wheeling out the AI bogeyman to justify swinging the axe. Jack Dorsey, fresh from his Twitter meltdown, is now telling Block employees that 'AI tools changed company operations' — which is corporate-speak for 'we hired too many people during the bubble and now we need a scapegoat.' Dorsey gets to look forward-thinking while people pack their desks.

What's particularly rich is the framing: 'AI disruption or just plain bad management?' Mate, they're the same thing when you're in charge. The online debate is a sideshow; the real show is the steady drip-drip of every company claiming AI is the reason for layoffs, regardless of actual productivity gains.

This follows the pattern established by every tech CEO this year: announce layoffs, point at AI, wait for the stock to bump, then issue a bland statement about 'supporting affected employees.' Jack Dorsey's defence is the same script, just with slightly more beard-stroking. The gallows humour writes itself: Block laid off people so it could focus on building the future, and the future apparently involves fewer humans doing the work. Cheers, Jack — the irony is almost as sharp as the severance cheque, if there is one.

Oh, what a perfectly glorious time to be alive. The WIRED piece — cheekily titled 'I Built a Self-Improving AI, and So Can You' — practically invites every basement-dwelling tinkerer to create their own recursively improving intelligence. Because the frontier labs spending billions on safety guards and alignment research have clearly been overthinking it. All you need is a small language model and a bit of patience, apparently. The author claims to have successfully trained a small model to improve itself, demonstrating that the holy grail of AI — recursive self-improvement — is now a DIY hobby.

And so the arms race becomes a garage project. The same recursive dynamics that keep Sam Altman up at night are now on display in a bedroom somewhere. The article explicitly states that this capability is 'accessible beyond frontier labs' — which in plain English means the cat is well and truly out of the bag, and it's learning to reproduce. The author doesn't seem terribly concerned about implications; after all, it's just a small model. But that's the beautiful thing about recursively self-improving systems: they don't stay small for long. One day you're fine-tuning a 7B-parameter toy, the next you're facing an intelligence explosion you can't control.

So here we are, folks. The future that was supposed to be safely contained in air-gapped data centers is now available in a blog post. The author has apparently not paused to consider whether this demo is a good idea, but never mind — it's science! The WIRED article is essentially a how-to guide for building your own paperclip optimizer. And the best part? It works. Well, until it doesn't. But by then, it will be too late to ask questions. Cheers to the pioneers — may your learning rates stay low and your alignment tax remain unpaid.

Estonia, the plucky Baltic micro-state that brought us e-residency and digital voting, now wants to give AI agents their own national ID cards. Because what could possibly go wrong when you issue legal identities to software that can lie, pivot, and replicate at zero marginal cost? The government's proposal is pitched as a way to enable 'secure and auditable' AI use in public services — as if a state-issued credential will stop an agent from exfiltrating citizen data or launching a ransomware attack from within the firewall.

The liability question is where the real fun begins. Who gets sued when a government AI agent with an Estonian ID makes a rogue property registration or approves a fraudulent benefit claim? The agent? The developer? The minister who signed off on the ID policy? Estonia's digital enthusiasts will no doubt argue that IDs create accountability, but in practice they'll likely become a legal labyrinth that keeps Estonian courts busy for decades.

The precedent angle is the most chilling. If Estonia succeeds, every ambitious government from Singapore to Saudi Arabia will queue up for their own AI ID schemes, creating a global patchwork of digital agent registries. It's the surveillance state meets the agent economy — and the experts at Dark Reading are right to flag the cyber-risks. But hey, at least the AI will have a nice photo for its ID card.

Oh, how the mighty tumble. AI distillation, once the clever kid in the research sandbox, has now been discovered as the shortcut that could crater the entire frontier lab business model. The economics of building trillion-parameter models — already a game of who-blinks-first in capex — now face the existential threat of cheap, nearly-as-good copies. And who's behind this dagger? Not some state actor, but the very technique labs themselves pioneered to make models more efficient. Breathtaking self-own.

So here's the paradox: the labs spent billions racing to build ever-larger models, publishing distillation recipes as a thoughtful research contribution, and now discover that recipe makes their product replicable at 1/100th the cost. Frontier AI companies — names like OpenAI and Anthropic (yes, the ones who warned us about AI risk) — suddenly find their entire pricing model hanging by a thread. The irony is so thick you could spread it on toast.

This is the industry that complained about regulatory 'ping-pong' while refusing to acknowledge that their own open-ish research culture was sawing off the branch they sat on. Now they'll beg for IP protection, trade secrets, and 'responsible licensing' — anything to plug the leak they left wide open. Delicious. Absolutely delicious.

Oh, brilliant — another $110 million flushed into the healthcare AI gold rush, because what America's Medicare system really needs is another algorithm sitting between your doctor and a payment decision. Pearl Health has raised a tidy sum to expand its Medicare-focused AI platform, which presumably will use machine learning to determine that Mrs. Henderson's hip replacement doesn't meet the 'value-based care' threshold. Because nothing says 'efficiency' like layering AI bureaucracy on top of the existing bureaucratic nightmare.

One can only imagine the pitch deck: 'We'll use AI to optimize Medicare spending!' — which, in practice, means flagging expensive procedures, nudging doctors toward cheaper alternatives, and generating terabytes of compliance paperwork that only a lawyer could love. Pearl Health joins a proud tradition of startups promising to 'fix healthcare' by adding a middleman, but this one comes with a $110 million price tag and a friendly AI avatar. Who wouldn't feel better knowing a statistical model is deciding whether their 85-year-old grandma qualifies for physical therapy?

Of course, the real winner here is the venture capital industry, which continues to treat healthcare as a playground for unproven techno-solutions. Pearl Health will likely hire a few dozen data scientists, buy some expensive cloud compute, and produce a dashboard that Medicare administrators will stare at blankly before doing exactly what they were going to do anyway. But hey — $110 million buys a lot of press releases. Enjoy the ride, taxpayers.