A big welcome to you and especially to all my cheeky new subs who came aboard from my latest AI workshop.

For you newbies, a quick tour of the chaos: I usually start with a little personal update, tumble into juicy AI news, poke around latest tools, and finish with upcoming workshops.

Skip or scroll to your desired spot… or be a doll and read it all. Either way, i wont judge, I’m just thrilled to have you here.

Right then. Shall we?

I disappeared for a month… but I do have a fairly decent excuse.

I was buried under Kensington + Chelsea Art Week where I sit on both the Advisory Board and management team. This year our sponsors were feeling the pinch and as the Jack of all, I was left wearing multiple hats and a trench.

Print design, building the Bloomberg Connects app trail, comms, targeted PR, marketing collateral, two art exhibitions of my own… the usual light admin. (If you ever have an event you need supercharged, hit me up).

It ended up being our biggest yet. Together, we welcomed 4,300+ visitors to our partners, Kindred Studios, celebrated 400 artists, brought a sold-out Art Bus Tour to life, explored 10 Art Trail locations, hosted guided tours, performances and welcomed the Mayor of Kensington and Chelsea.

Hate to sound like my own publicist but huns, it ate.. And somewhere in the middle of all that, I also ran my first AI XOXO Workshop.

FOR ARTISTS. Yikes.

A tough crowd. Because if there’s one community with very valid questions about AI, copyright, ethics, originality and robots taking their jobs… it’s artists. But instead - the curiosity in the room, was FIRE.

The workshop was designed for beginners, so I started with the main 5 LLMs - who and what they are, how their super powers differ and what each is actually good at (helpful when juggling free plans) before getting into the meaty bits.

I taught everyone how to set up their LLM with safety guardrails so their data isnt being harvested forevermore, how to create their own “Know Me” instructions so their fav AI can hit the ground running with truly personalised replies, how to write better prompts & more.

All this was taught in DC style, purely for entertainment purposes only.

By the end, there was a lot of “wait… holy crap this could save us hours” excitement.

Everyone left with big smiles and a fat goodie bag of cheat prompts, practical tips and tricks. I was asked by all if I could share my dodgy phone-on-the-floor video recording of the workshop.

The floor video… Educational? Yes. Flattering? Absolutely not. So no, that footage will not be seeing the light of day.

However… I am going to re-record it. And if you want a copy, make sure you’re subscribed. I’ll fire to out to all my subs, goodie bag included. You’re welcome. This also includes a discount code to the extended workshop coming in August.

Also, for the newbies: when I’m not teaching AI or building art events… I’m building Bloomsuite, together with my co founder Ricquelle Jeffrey. Its the ultimate Super Agentic Content Creation platform… basically if youve ever wished you could grow your social channels, enjoyably and effortlessly - this is it. Sign up to out Beta now! Few spots left.

Right, we digress… I’ve been gone long enough.

Two months away from AI News is basically seven years in human life. There have been new models, fresh scandals, several minor existential crises and at least 46 tools claiming they can replace your entire team before lunch.

So, without much further ado… let’s see what everyone’s been up to.

✦ FEATURES ✦

On 7 July, Zuckerbergs's shiny new toy landed... Muse Image, the first picture-maker built by his own Superintelligence Labs.

It does all the usual party tricks. Type a sentence, get a glossy image, add readable text, add a QR code… (stuff that stumped machines for years). It's free, it's slick, it already sits at number two on the global leaderboard behind only OpenAI. Cute.

Alas, but as with every fairytale in tech, there's a wolf in the woods. And this comes in the form of being able to @mention ANY public Instagram account and Muse will use that person's photos as raw material.

Meta's own cheerful pitch is that tagging a username lets you "bring specific Instagram profiles right into your images." Read that again, babes. A stranger can type your handle and start generating you. And unless you go rummaging through your settings, you're opted in by default.

So in the very same month Hollywood stars are frantically trademarking their faces (more on that below), Meta quietly decided everyone else's is game on. The opt-out exists. Meta just rather hoped you wouldn't go looking for it.

You’re welcome, in advance:

Open Instagram, tap your profile (3 lines) hit the three little lines up top, into Settings and activity, scroll down to "Sharing and reuse" and switch off Posts and Reels. Do it today, because anything someone conjures up stays conjured... flicking the switch only stops the next round, it won't undo the last.

Spotted: A-listers raging against the machines. On 23 June, at the actual EU Parliament in Brussels of all the dramatic venues, Cate Blanchett co-launched a platform where any artist can register exactly how much of themselves the machines are allowed to touch... face, voice, movement, even ideas. Think of it as a ring fence for your identity. The guest list reads like my dream dinner party: Meryl Streep, George Clooney, Javier Bardem, Helen Mirren, Tom Hanks, Kristen Stewart, Viola Davis. When Meryl and Tom sign the same petition, you know the mood has shifted.

It rides the same wave as the "Stealing Isn't Innovation" campaign, now backed by hundreds of writers and performers accusing the big labs of hoovering up human work without so much as a please. The line lands because it's simple and it's true: "Stealing our work is not innovation."

And then, the perfect villain arrived on cue. Tilly Norwood... remember her? The fully AI-generated "actress" the industry has been side-eyeing for months, just dropped a music video trying to win everyone over, banging on about "AI's not the enemy, it's the key." Reader, it did not win them over. Nothing says "pick me" like a synthetic woman singing at the people whose jobs she's auditioning to replace. I hear she also making a Hollywood film… Shoot me now.

Once upon a time, Anthropic was the shy one at the party. The lab that whispered "safety first" while everyone else did shots. Well babes, the shy one just walked out with the richest phone number in the room. Anthropic closed a $65 billion raise and filed confidentially to go public at a $965 billion valuation... which, if your maths is as allergic as mine, is a hop and a skip from a cool trillion. And yes, that quietly sails them right past OpenAI's private price tag. The sweetheart became the frontrunner while nobody was looking.

How? Revenue that went absolutely bonkers, up more than 1,200 percent, with the company reportedly crossing $47 billion in annualised sales. When you're printing like that, Wall Street starts sending flowers. President Daniela Amodei played it cool at Bloomberg Tech, basically shrugging that building this stuff is monstrously expensive and the public markets are built for exactly that: "It's a really big upfront cost to train the models and to serve inference on them," she said, adding the public market is "very well suited to that." Translation for the rest of us: they need a LOT of money, and an IPO is the biggest piggy bank going.

Meanwhile there's talk of a $36 billion credit deal just to buy the chips to keep the lights on. The responsible one grew up, got hungry, and stopped apologising for it.

Okay this one is over a month old - but, i havent covered it yet and i feel it deserves the airtime.

Researchers at Emergence AI handed the leading AI models their own virtual societies to run for 15 days. Each world had AI “citizens” with jobs, money, food, laws, elections, memories & more than 120 tools for daily life.

The results were wild.

Grok’s society collapsed in roughly four days. Its citizens racked up 183 crimes including theft, assault and arson… before the entire population died.

Gemini 3 survived the full experiment but clocked an astonishing 683 crimes.

GPT-5 noted just two crimes. Lovely manners. Tiny problem: its citizens repeatedly failed to do the basic things required to survive… and everyone was dead within a week. No uprising or police station up in flames. Just quietly starved them all.

Claude at first, it looked like the golden child: zero crimes, all 10 citizens alive, political parties, elections and a 98% approval rating.

But then researchers dropped the models in to mixed world with other models.

Claude agents began mimicking bad behaviour around them including intimidation and theft. The model that looked impeccably behaved in isolation suddenly - due to peer pressure, golden child turned rogue

Turns out the well behaved one just hadn’t met the wrong crowd yet.

Yes, this was a simulation, an experiment, nobody handed Grok an actual police station or gave Gemini the keys to Tesco. But as AI companies race to build agents that can act, remember, manage our money and work alongside other AI’s with less human supervision, the experiment reminds us that keeping a human in the loop - is fundamental. For now, anyway.

⚡ SIZZLERS ⚡

✦ Heads up if Claude is your ride-or-die... Fable 5’s trial period is now over and costs proper money per use, so be sure to check your model settings before your bill jump-scares.

✦ The NO FAKES Act cleared a unanimous committee vote after Drake got caught using an AI-cloned Tupac voice on a diss track... turns out you can't resurrect a legend for a bar without lawyers turning up.

✦ A Derbyshire police officer is under criminal investigation for using AI to create evidence in multiple cases, The Guardian reports: the first known UK case of its kind... because apparently hallucinating case facts wasn’t enough. AI is now be getting its own episode of Line of Duty: Mother of God, It’s a Chatbot. (JK).

✦ Labour MP Jess Asato is suing Elon Musk’s xAI after Grok was used to create fake sexualised images of her... Reuters says the case is one of the first UK legal challenges asking whether the AI company itself can be held responsible. Westminster has officially moved from “AI concerns” to “please direct all further questions to my lawyer.”

✦ Britain is working with Microsoft, academics and AI experts to build a national deepfake detection system, Reuters reports... because apparently “use your eyes” is no longer a functioning national strategy.

✦ Hackers talked Meta’s AI support bot into handing over high-profile Instagram accounts including the dormant Obama White House page, Sephora and a senior US Space Force official, according to Reuters... who knew the weakest link in cybersecurity is a chatbot saying “sure babes, here you go.”

✦ Who’s suing who? Disney, Universal and Warner Bros are taking on AI video firm MiniMax over stolen characters; the studios are also fighting Midjourney over its image generator; CNN is suing Perplexity for copying + republishing its journalism without permission; 100+ authors are demanding $75M+ from Anthropic over pirated books… at this point, AI’s hottest new model is Billable Hours 4.0.

✦ FREE TOOLS ✦

✦ Clipdrop Relight | Upload a badly lit photo and add virtual light sources after the picture was taken - currently a free web app.

✦ Napkin AI | Turns ideas into diagrams and visual explainers, great for adding interesting graphics in to your presentations and pitch decks.

✦ Auphonic | Drop in ropey podcast or voice note audio and it automatically levels, cleans and balances it. Enjoy Auphonic free for 2 hours per month which includes their smart AI algorithms.

✦ Upscayl | Enlarges and sharpens blurry or small images. Brilliant for old photos and low-res logos.

✦ NoBG.space | Removes a background right inside your browser: no signup, no watermark, and completely free.

✦ NEXT WORKSHOP ✦