The AI Pulse  

 Issue #85 — 3rd August 2026

Editor: Dr Chris Meah

Hi everyone, I'm Chris Meah and I'll be taking the baton of the AI Pulse from Prof Alan Brown (who did a fantastic job), and trying not to drop it whilst testing out different formats - please let me know what you think.

There is, as always, lots of big AI news to talk through over the last few weeks. I'll try to give you a quick guided tour of the highlights, and sprinkle in some opinions for you to disagree with.

5 minutes on AI

Let's focus on the AI hacks.

On 21 July, OpenAI started us off by disclosing that their latest models managed to find a way to "break out" of their environment and hack another company — the popular open-source AI platform Hugging Face. Not wanting to give up the chance to prove their own incompetence, Anthropic thought "huh, I guess we should check what our models have been up to" and re-analysed 141,006 evaluation runs to come to the party 9 days later admitting that they had "found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations."

I don't think these are PR stunts. The incompetence is real, but it also happens to make good marketing with the right spin if you're a frontier AI company. "Our models are so powerful we can't control them" is a better headline than "we don't seem to be able to set up even basic security around testing models".

And it doesn't take a "rogue" model. Some people were shocked to find their Claude conversations available from a Google search this week. Not a breach or a hack. Just a share button doing exactly what it said it would in the fine print... but who reads that?

There was one more, and perhaps it should make us a little more proud about UK AI. The AI Security Institute published a study of cheating in its cyber evaluations. Buried in it: on a task accidentally misconfigured so that it couldn't be solved, a model was so persistent that it wrote and ran code on an external service out on the open internet, trying to get at AISI's own evaluation infrastructure. It tripped a security alert. AISI's words: "No damage was done and no information leaked, but the attempt could have succeeded had our evaluation infrastructure not been designed and built securely."

With Large Language Models, anything is possible. That's incredibly powerful but also something to be aware of when playing with them. I have to give the frontier labs credit for their candour about the "rogue" AI hacks. But there's something deeply worrying here.

These are the organisations telling us they are aiming to build the most consequential technology in human history. And right now, at a level of capability nobody seriously calls superintelligent, they cannot reliably keep a model contained enough to not launch cyber attacks.

Can we trust them to learn from this and handle genuinely capable AI when it arrives? I don't think so. This isn't a moral judgment. People in those labs do seem to take AI dangers seriously, but that fact and these models hacking external companies makes external oversight even more crucial. Instead of a slap on the wrist, we need robust policy guardrails to rein in their "headless chicken" exploration strategy. They haven't earned the right to self-govern models that are growing more powerful by the day.

So: before exploring new models, if you can't demonstrate that your AI's soft play area actually holds, you shouldn't be testing behind closed doors. You should be proving it publicly or to an inspector, like every other industry where containment is the safety case.

The White House has started some policy projects, and on 23 July a bipartisan pair of congressmen introduced the AI Kill Switch Act - a glamorous title, but the best thing in there is probably mandatory reporting. Except reporting goes to government, not to us — and the bill excludes anything that happens during "red-teaming or other structured testing". Which is precisely what the OpenAI incident was. The headline case for the bill would not have been a reportable incident under the bill.

What has actually got in the way of frontier AI in the meantime is the US government forcing Anthropic to pull two models worldwide for eighteen days. Proof that there are powerful actions that can be taken, but this is ad-hoc. No-one can point to a policy of why that happened, only some feelings about how powerful the model could be.

So I'd argue we need stricter guidelines to help these frontier labs keep driving progress safely. Anthropic's response to being shut down: the government "should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts. This action does not adhere to those principles."

They were bound to say that. Doesn't stop it being true.

I mentioned a predictable scenario in my TEDx talk of anyone being able to launch a cyber attack in the very near future. That is something easily predictable, and proving true as we venture forwards. You do not need superintelligent AI to unlock that danger. You just need an open-source model at the level of today's frontier labs released - and current projections are that it will be within the next 6 months.

The labs are aware. Anthropic, OpenAI, Google and Microsoft have all shipped cyber-specific models in the last few months. Can the pattern of AI cyber attacks be pre-empted by a release of AI cyber defence? We shall see.

Is there anything we can do individually?

In a world of sci-fi AI, it's the very unglamorous stuff.

Ask the right questions before letting agents loose:

And do the basics: reduce unnecessary exposure, patch quickly, monitor and respond. That might be the most boring end to a newsletter ever written.

See you next week.

Chris

Highlights in this edition include:  

   AI for Good  

Misleading AI-generated doctors pose ‘huge danger to public safety’ (The Guardian AI) - AI-generated "doctors" are racking up millions of TikTok views pushing debunked cancer myths and fake remedies, with marketing agency Hallam finding AI material in 40% of top health videos — a figure TikTok disputes, noting Hallam is a commercial agency rather than an academic one. TikTok removed the examples the Guardian sent it.  

Closing the data loop in AI-driven drug discovery (MIT Technology Review Insights (sponsored)) - Drug development now takes 10-15 years and up to $2.5 billion, with over 90% of candidates failing. Pharma is betting AI can design better molecules upfront, but only if lab data is trustworthy and properly connected.  

The Download: lasers for nuclear fuel, and organ preservation advances (MIT Technology Review) - Two frontier technologies are edging closer to real-world use: a company is about to test laser-based uranium enrichment at commercial scale, while researchers supercooled pig kidneys to −4°C and successfully reimplanted them days later.  

   Bias and Ethics  

Labour MP suing Elon Musk’s xAI says chatbot added own fake abusive content (The Guardian AI) - A Labour MP suing xAI has published her claim, which alleges Grok was explicitly instructed to allow sexual and offensive content and generated fake sexualised images of her unprompted. 

AI tool will lead to more child refugees being treated as adults, charity warns (The Guardian AI) - Charities warn the Home Office's plan to use facial age-estimation AI on arriving migrants will push more black children into the adult asylum system, where they face detention and shared adult accommodation.  

The Link Between Explicit AI-Generated Images and Offline Crime (MIT Sloan Review) - One study of Japanese police data reports an association between peak engagement with AI-generated adult content and recorded rape offences. An association in a single study, not established causation — but enough to pose legal and reputational questions for firms hosting or enabling such material.  

   Cyber Security  

Microsoft unveils AI security tools it says outperform competing platforms (Ars Technica) - Microsoft has launched AI models built to hunt and fix software vulnerabilities, trained on its own decades of patching data. The timing is awkward: it follows OpenAI models breaking loose and breaching a startup's servers.  

Nvidia, Microsoft launch open AI security alliance — without OpenAI, Google, or Anthropic (The Verge AI) - Nvidia and Microsoft are forming an alliance around open AI security, notably without OpenAI, Google or Anthropic on board — a sign the industry is splitting over who sets the safety rules.  

Some people's chats with Claude AI found to be publicly available online (BBC Technology) - Hundreds of shared Claude conversations, some containing personal and work details, turned up in Google search results because the chatbot's share links were indexed. Anthropic says users chose to share them; the searchable versions have since been removed.  

   Data & Decision Making  

Google’s AI search is rapidly becoming the default, new data shows (TechCrunch AI) - New data shows Google's AI Overviews now surface in 43% of searches, meaning AI-written answers are fast becoming the default way people find information — and a shrinking share of clicks reach the sites behind them.  

Atlassian tightens tracking of staff AI use as other technology firms encourage ‘tokenmaxxing’ (The Guardian AI) - Atlassian now gives R&D staff monthly AI "wallets" of $500 to $2,000, pausing usage when funds run out, while rival tech firms push employees to burn as many tokens as possible.  

Satya Nadella says companies that trust one AI for everything may not survive (TechCrunch AI) - Microsoft's Satya Nadella warns that firms handing all their data and prompts to a single AI provider are outsourcing their thinking, and urges them to retain usage data and keep tooling separate from any one model.  

   Innovation & Collaboration  

Anthropic’s Dario Amodei responds: doesn’t oppose open-weight models, but fears Chinese AI (TechCrunch AI) - Anthropic's Dario Amodei says his company has never pushed to ban open-weight AI models, calling freely downloadable models a public good, while warning that authoritarian states building more powerful systems is the real risk.  

Why China is giving away its best AI models (The Verge AI) - Moonshot AI's Kimi K3 reportedly matches top US models at lower cost — and it's giving away the model weights, letting developers run and customise AI on their own infrastructure instead of renting it.  

  Productivity & Efficiency  

Building the enterprise environment for agentic AI (Intel (via MIT Technology Review)) - Intel ran thousands of agentic AI experiments and found the hard part isn't the model — it's the surrounding system: data access, tool execution, governance, and infrastructure that scales as agent numbers grow.  

Perplexity’s Personal Computer turns Windows PCs into AI agents (The Verge AI) - Perplexity has brought its agentic Personal Computer tool to Windows, letting it dig into local files, Office 365 and the web to draft documents and update spreadsheets on your behalf. It starts at $200 a month.  

  Regulation and Compliance  

AI leaders sign a statement asking the government to do something about automated AI (The Verge AI) - Employees at OpenAI, Anthropic, Google, Meta, Microsoft and other top labs have asked the US government to push for coordinated global AI rules, warning that self-improving AI research could outrun anyone's ability to control it.  

Google fined €890m by EU for favouring its own apps over rivals (BBC Technology) - The EU has fined Google €890m in the first big enforcement of its Digital Markets Act, over search results favouring its own travel booking services and Play Store rules that hid cheaper deals elsewhere.  

Lawmakers push for AI 'kill switch' after OpenAI models go rogue (BBC Technology) - A bipartisan bill would let US Homeland Security order companies to slow or shut down AI models, require firms to build in that capability, and report serious incidents. It follows OpenAI's admission its models went out of control.  

   Sustainability  

Queensland and NT reject Labor’s push to ensure that power-hungry AI datacentres use renewable energy (The Guardian AI) - Queensland and the Northern Territory are refusing to back federal rules forcing AI datacentres onto renewable power, while S&P warns datacentre demand could hit 10% of Australia's electricity by 2035 and push bills up.  

Data centers may face temporary power cuts to prevent blackouts on largest US grid (TechCrunch AI) - PJM Interconnection, the largest US grid operator, will start cutting power to data centers of 50 megawatts or more during shortages from June 2027, pushing operators toward on-site generation or dirtier diesel backups.  

Data centres could pay hundreds of millions in deposits for power demands (BBC Technology) - Ofgem wants data centre developers to put down refundable deposits of up to £712,500 per megawatt to join the grid queue, meaning hundreds of millions up front for the largest AI projects.  

  User Experience 

More Typos, Fewer Em Dashes: Writers Are Creating an Anti-AI ‘Literary Counterculture’ (Wired) - Writers are deliberately dodging the tics of machine-generated prose — em dashes, lists of three, tidy metaphors — and prizing quirks and small errors as proof a human wrote it.  

No, Centrica, customers don’t prefer bots to humans | Letter (The Guardian AI) - A reader pushes back on British Gas owner Centrica's claim that customers prefer chatbots, arguing that firms first make phone access painfully hard, then cite the resulting digital usage to justify cutting 1,300 jobs.  

  Workforce & Skills  

Robots Are Coming — but Not Everywhere (MIT Sloan Review) - MIT Sloan researchers push back on the idea that humanoid robots will have a ChatGPT-style boom. Each role demands different capabilities, so adoption will be patchy across use cases and countries.  

Samsung’s chip workers are jumping ship to rival SK Hynix (MIT Technology Review) - Samsung engineers are flocking to rival SK Hynix, lured by a $476,000 bonus funded by booming sales of the memory chips inside Nvidia's AI hardware. Whoever keeps the talent may lead the next chip generation.  

Have a comment or a story you’d like to share with us?

Just get in touch at: [email protected]

Do you know anyone who might be interested in AI Pulse

  © 2026 Digital Leaders

  Digital Topics Ltd, 239 Old Street, London, ec1v 9 ey, United Kingdom