iiinova — where your digital self keeps evolving

Reposted by John Burn-Murdoch

Tony Tassell @tonytassell.bsky.social · 8h

Man City’s Emirati chair was granted diplomatic immunity 6 yrs ago in a move that could complicate any legal action resulting from findings that the club artificially inflated its finances by more than £900mn - scoop by Caroline Binham @journosooz.bsky.social as.ft.com/r/d717fb9e-2...

1417759

Reposted by John Burn-Murdoch

Dan Neidle @danneidle.bsky.social · 7h

Manchester City cheated Premier League rules, the Commission found. Did it also cheat HMRC? We've examined the decision and leaked Mancini documents. We think City likely failed to pay £12m in tax. There should be a criminal investigation. Thread:

52853383

John Burn-Murdoch @jburnmurdoch.ft.com · 19/09/2026

Yep — precisely the point I made in the piece

070

Reposted by John Burn-Murdoch

Hetan Shah @hetanshah.bsky.social · 17/09/2026

Helpful points from @jburnmurdoch.ft.com here on what AI solving big maths problems might mean for the future of the maths community www.ft.com/content/c947...

Many mathematicians are leaning "bad". few days after the OpenAI Navier-Stokes
announcement, a group of 25 winners of mathematics' prestigious Fields Medal published a letter arguing that the goals of the AI companies and the human
mathematics community are fundamentally misaligned. Their core position is that although solving complex technical problems has historically accompanied expansions in our understanding, the former was only ever a proxy for the latter. A brute-forced solution without the
deliberation and painstaking accumulation of human knowledge along the way is the signal of expanded understanding without the actual expanded understanding.

Another big question is what this all means for the incentives to share progress towards new discoveries. It's not hard to see how the threat of being scooped by AI could exert a chilling effect on the culture of sharing important incremental findings or promising new avenues, which has underpinned scientific progress for centuries. And for the Al labs themselves,
the backlash over the Navier-Stokes case may mean future Al-powered breakthroughs are kept under wraps. If AI is amazingly capable at turning new theories into findings, but stifles the development and sharing of those theories, there may be a bumpy road ahead.

25018

Reposted by John Burn-Murdoch

Sarah O'Connor @sarahoconnorft.ft.com · 14/09/2026

In both China and the west, the gig economy has become a sort of safety net/labour market shock absorber. But automation is coming for a lot of this work. Which means the biggest impact of AI/robots might unfold in the dark, in the part of the economy hardest to track. www.ft.com/content/a2c3...

415868

John Burn-Murdoch @jburnmurdoch.ft.com · 11/09/2026

I mean, the entire piece is us saying “this looks bonkers and goes against both economic theory and real-world evidence to date”

100

John Burn-Murdoch @jburnmurdoch.ft.com · 11/09/2026

Wow, yeah no quite the opposite! We thought the extreme scenario looked bonkers so we wanted to understand how they could explain it. Slightly alarmed that didn’t come across from all the times I said their assumptions and overall model of the dynamics didn’t line up with what we see in the world 😅

030

Reposted by John Burn-Murdoch

Sky Marchini @sky.skymarchini.net · 11/09/2026

>the houthis were trying to use claude to build their rocket guidance system just get me off the fucking ride www.anthropic.com/threat-intel...

GTG-87001: Disrupting a Yemen-based guided weapons engineering cell using Claude to develop guidance software
Summary

We identified a cell of threat actors based in northern Yemen running three weapons development programs: a guided rocket that used a commodity phone-class flight computer with final-phase homing guidance; a multi-stage ballistic missile with a stated range goal above 2,000 km; and a multi-variant missile (referred to as the “R2000” set) that included a hypersonic glide vehicle variant.

The actors used Claude Code in place of human software engineers to develop the guidance, navigation, and control (GNC) software that steers and stabilizes a flying vehicle. For example, they used Claude to integrate an open-source autopilot onto a phone-class flight computer, writing the control and position estimation software, tuning the control settings, running a firmware build pipeline, and performing a flight simulation. The actors managed several Claude instances at once, assigning each one a role, much as a lead would delegate work on a small engineering team: the actors tasked one instance with writing the code, another with research, and a third with reviewing the code the first instance produced.

Our safeguards blocked many of their requests, but not all of them. The actors used a variety of tactics to evade our safeguards, including hiding their goals and the products the software was meant for, and they split their work across multiple sessions so no single session revealed their full intent.

These actors carried out a sustained effort to develop guided weapons, including using Claude to design guidance software. We do not have evidence the actors succeeded in fielding an operational device; but they did test-fire a guided rocket. This field test appears to have failed: within hours, the actors returned to Claude to work out why it failed.

We identified this activity as part of our internal investigations into suspected weapons development. We banne…

441066197

John Burn-Murdoch @jburnmurdoch.ft.com · 11/09/2026

I mean, yes, those are literally the questions we asked, which you can see in the transcript! “This seems pretty wild and at times contradictory, how do explain it?” was literally the entire point of the conversation, and evident right from the opening question 😆

151

John Burn-Murdoch @jburnmurdoch.ft.com · 10/09/2026

The environmental aspect has been egregiously overplayed and is in many cases based on acknowledged errors. Whereas I think it’s completely rational and understandable that there is real anger among people who feel their work has been stolen and their livelihood devalued and threatened.

12201

John Burn-Murdoch @jburnmurdoch.ft.com · 10/09/2026

Don’t worry, that was just a misunderstanding — if robots produce trillions of paperclips and wipe out humanity in the process, GDP per [human] capita is actually infinity, not zero

11006

John Burn-Murdoch @jburnmurdoch.ft.com · 10/09/2026

Also asked the meta-question of whether they appreciate that headlines of astronomical growth alongside soaring unemployment might be fuelling anti-AI sentiment

John Burn-Murdoch: The top line of the press release mentions the 44tn
[GDP figure by 2030] and the possibility of 15 per cent annual GDP growth by 2030. Given the whole mood at the moment - there are a lot of strong feelings around Al, and it feels like the most likely explanation for why everyone hates data centres is because they don't love what's going on inside the data centres and the implications of that. So when a report comes out that says we might have 15 per cent annual growth in 2030, that's going to wind up some people. When it's also saying we might have double-digit knowledge-worker unemployment, that's going to scare some other people. I'm curious, when you put together a piece of work like this, are you thinking about how that might come across? What's your intended impact?

Jack Clark: We build this stuff driven by a pure research goal of figuring out what's actually the truth. I lead the Anthropic Institute, which is a series of research teams here, and our mandate is to take information, or in this case, assumptions, that only the Al labs have, and get that information and that set of assumptions out to the public, so people can better understand the different views that Al companies have about the future. Some scenarios look like Al as a normal technology, other scenarios look like the singularity is nigh and very, very significant things are going to happen to the economy. So we built a tool that captures all of these different views. We don't build this stuff with a particular narrative in mind at the start. We're sharing this because if these scenarios end up being right, you probably want to make a bunch of different choices about the world. Part of why you share this is I think it's going to help focus the minds of economists, policymakers [and] others who study Al on how we may need to make very different policy decisions about this technology.

10799

John Burn-Murdoch @jburnmurdoch.ft.com · 10/09/2026

Here’s a snippet on why we may not yet be seeing the full extent of how AI is already changing employment. Just one small part of a much broader chat where he explains Anthropic’s thinking and the purpose of publishing scenarios like these. Full transcript is here: www.ft.com/content/432f...

John Burn-Murdoch: Just picking up directly on that. Brynjolfsson has those findings, which are really important, but those are very concentrated among new entrants to the labour market in specific occupations. You probably saw the analysis from Ramp a couple months back, which showed that according to that universe of data, Al-adopting firms are actually hiring more than others, including among juniors. So I guess it's at least fair to say the evidence is fuzzy there. More broadly, on net over the last couple of years — one year if we're just talking about the agentic period - the aggregate impact on employment has arguably been positive, or it's certainly not decisively negative, even for the most junior roles. You're nonetheless saying unemployment among knowledge workers could get to double digits by 2030. What is the other shoe that drops between now and then?

Jack Clark: So, under scenarios where really significant things happened, I think you can assume the AI systems have got better at not just individual tasks that comprise a job, but a large bundle that maybe comprise the majority of a job. And we know that it seems to be driving a rise in what you think of as very competitive, small headcount, opex-heavy firms. Small numbers of people, large numbers of tokens, seemingly very competitive. Stripe sees this; other people see this.
Employment takes a lot longer to show up, and employment changes are typically also accompanied by larger, unpredictable changes in the macroeconomy, like a recession. So right now, we don't really know the shape of how employment is being impacted by Al because we haven't had these larger macroeconomic events, which typically is when businesses move to some new norm of employment, which usually reveals a load of changes in how they've been using technology in their business in prior years.

56710

John Burn-Murdoch @jburnmurdoch.ft.com · 10/09/2026

New from us: Anthropic just published scenarios for AI’s possible economic impacts, which range from minimal, to explosive GDP growth of 15% by 2030 as knowledge-worker unemployment hits 18%. I sat down with their co-founder Jack Clark to pick his brains on how they’re thinking about all of this.

7118551

John Burn-Murdoch @jburnmurdoch.ft.com · 09/09/2026

The problem with switching to a focus on softer/broader/transferable skills is that the evidence shows this doesn’t work. Critical thinking, creativity etc only really exist as a function of domain knowledge. Really recommend reading @daisychristo.bsky.social on this www.aei.org/research-pro...

aei.org

121

John Burn-Murdoch @jburnmurdoch.ft.com · 09/09/2026

But PISA tests and scores measure fundamental abilities like parsing long texts, understanding and explaining concepts, applying methods etc. The reason they can be used identically for every country is that they *don’t* measure things specific to particular curricula or teaching/learning methods.

120

John Burn-Murdoch @jburnmurdoch.ft.com · 08/09/2026

Yes agree! I was struck by that at the time

020

John Burn-Murdoch @jburnmurdoch.ft.com · 08/09/2026

Sorry yep I should have been more precise: I think a lingering impact from Covid is certainly possible in some cases. But last time around many people specifically said the 2022 results were likely a one-off Covid hit and 2025 would rebound.

3554

John Burn-Murdoch @jburnmurdoch.ft.com · 08/09/2026

As I see it, the unifying global trend is the deluge of digital distractions which are eroding people’s capacity to focus and think for extended periods of time, making strict structures and proven methods especially important for education. Full piece here: www.ft.com/content/a801...

ft.com

Have humans passed peak brain power?

Data across countries and ages reveal a growing struggle to concentrate, and declining verbal and numerical reasoning

1034166

John Burn-Murdoch @jburnmurdoch.ft.com · 08/09/2026

2) England is a rare exception. During the 2010s many other countries (including the other 3 UK nations) changed tack, experimenting with ideas like transferrable skills instead of focusing on tried-and-tested disciplines and methods. It looks like those experiments have failed.

1426362

John Burn-Murdoch @jburnmurdoch.ft.com · 08/09/2026

Two extraordinary charts from today’s PISA test results: 1) School test scores continue to collapse internationally, underscoring how this is no longer a Covid effect but sustained decline. Those falls in reading and maths are equivalent to about two years of lost schooling.

41856344

John Burn-Murdoch @jburnmurdoch.ft.com · 05/09/2026

For example the difference between a jobbing PT and a PT with a thriving career is not how good they are at guiding workouts, it’s how good they are at building relationships and networks.

010

John Burn-Murdoch @jburnmurdoch.ft.com · 05/09/2026

Yup. The other thing that I think people sometimes miss with PTs (and with many other growing jobs in Imas’s framework) is that interpersonal skills are often *massively* important well beyond what’s strictly required to perform the service.

110

John Burn-Murdoch @jburnmurdoch.ft.com · 05/09/2026

FWIW I view the implications of Alex’s argument more through the lens of aptitudes and skills than degrees per se, though the marked rise were seeing in people qualifying as therapists or PTs shows training and quals of some form will certainly be important

161

John Burn-Murdoch @jburnmurdoch.ft.com · 04/09/2026

Both postings and employment levels (which implies hires too) FWIW bsky.app/profile/jbur...

020

John Burn-Murdoch @jburnmurdoch.ft.com · 04/09/2026

Totally agree — quite similar to my experience

010

John Burn-Murdoch @jburnmurdoch.ft.com · 04/09/2026

Obvs we’ve had the min wage increases, but I also did a median-vs-median premium and the result is the same. Dominant dynamic is grad earnings falling for everyone from median upwards.

130

John Burn-Murdoch @jburnmurdoch.ft.com · 04/09/2026

Agreed on AI, but less sure on the rest? The wage compression story was definitely true in the US, but in part a result of them going for unemployment rather than furlough, I thought? Here RF’s RTI series has the employment rate peaking in 2022/23 and weakening since.

110

John Burn-Murdoch @jburnmurdoch.ft.com · 04/09/2026

The most recent decline is definitely steepest among most recent entrants. Question I think is whether this is about the labour market of the last few years being abnormally weak due to so many colliding sources of uncertainty, or start of a new normal.

120

John Burn-Murdoch @jburnmurdoch.ft.com · 04/09/2026

Understandably lots of attention on min wage, but it only directly touches ~6% of all workers, ~10% of non-grads. Grad malaise story much broader. Real earnings still well below pre-GFC levels for all grads, and since 2022 down *again* for the median grad upwards.

152

John Burn-Murdoch @jburnmurdoch.ft.com · 04/09/2026

It’s an obvious theory for sure (in a recent piece on the same topic I assumed it was a big factor www.ft.com/content/649d...), but I did a robustness check for this one where I measured the premium as median grad vs median non-grad (who is well above min wage) and the result is ~exactly the same

280

John Burn-Murdoch @jburnmurdoch.ft.com · 04/09/2026

Totally agree (have said this lots in other comments) bsky.app/profile/jbur...

200

John Burn-Murdoch @jburnmurdoch.ft.com · 04/09/2026

Yep, I think more vocational quals is one likely outcome of all of this. And/or more aptitude tests and skills-based hiring (see today’s column www.ft.com/content/54cf...) And well beyond software in both cases.

ft.com

The university degree is losing its lustre

Technological and economic shifts are eroding the graduate premium

000

John Burn-Murdoch @jburnmurdoch.ft.com · 04/09/2026

...they are assuming someone else is training the experienced workers of tomorrow, who they will then hire. Which is quite the gamble! (3/3)

7620

John Burn-Murdoch @jburnmurdoch.ft.com · 04/09/2026

But you don’t hire talented juniors *for the routine tasks they do*. You hire them because you want to develop them into talented seniors. When a firm says "AI can do the research and make the slide decks, so we don’t need juniors" either they don’t realise this, or... (2/3)

2648

John Burn-Murdoch @jburnmurdoch.ft.com · 04/09/2026

Yup, and it’s the same for loads of exposed white collar roles. In a growing number of cases AI can do a very large amount of the work that would have been given to an intern, trainee or junior hire 5 years ago. And genuinely do it to a comparable or better standard. (1/3)

170

John Burn-Murdoch @jburnmurdoch.ft.com · 04/09/2026

Of course, hiring fewer juniors has long-term consequences. Where do the experienced engineers of tomorrow come from? But it’s not yet clear how that is going to resolve. Nice paper on the possibilities here: papers.ssrn.com/sol3/papers....

papers.ssrn.com

Pyramids, Diamonds, and Oscillations: AI and the Structure of Internal Labor Markets

AI is changing workplaces: it leads to higher individual productivity and faster learning. We model the effects of these changes on internal labor markets. We s

5191

John Burn-Murdoch @jburnmurdoch.ft.com · 04/09/2026

And that maps onto exactly what you said: "Firms realized they needed the *expertise*" — i.e experienced engineers who have acquired lots of expertise from years writing code. It’s that accumulated and often tacit knowledge that AI makes much more valuable. Value proposition for juniors is weaker.

1140

John Burn-Murdoch @jburnmurdoch.ft.com · 04/09/2026

NB we’re interested specifically in *entry-level* roles. The picture you show for Denmark is actually not dissimilar to the US: software hiring has rebounded pretty well, but *exclusively for senior roles*. For juniors the picture is still bleak. We did a deep dive here: www.ft.com/content/7325...

150

John Burn-Murdoch @jburnmurdoch.ft.com · 03/09/2026

But as of right now the consistent picture is: - Marked decline in entry-level software hiring - Marked decline in employment for recent CS grads (73% -> 58%) - Marked decline in applications to study CS All alongside an explosion in cases like your TIKZ one I don’t think there’s a conflict there?

5301

John Burn-Murdoch @jburnmurdoch.ft.com · 03/09/2026

Sure, there are all sorts of other scenarios. Jevons for one. Or a surge in demand for people who will check and refine AI-written code (which would effectively mean a bifurcation into lower-pay routine jobs and a minority of superstar jobs).

4150

John Burn-Murdoch @jburnmurdoch.ft.com · 03/09/2026

In part *because* AI makes it easy for non-experts to generate code [that in the past would have been handed off to another human], we’ve already seen a marked decline in demand (employment) for entry-level coders, hence people seeking a professional qualification in CS may continue to decline.

160

John Burn-Murdoch @jburnmurdoch.ft.com · 03/09/2026

Right, totally agree (I use AI to write code ~every day). Though everything you’ve described is exactly what I was getting at with the original post. Lots more code getting produced by lots more people for lots more purposes, but fewer people pursuing coding as a trade.

2110

John Burn-Murdoch @jburnmurdoch.ft.com · 03/09/2026

Just to make sure I’m not misunderstanding: you think more people will be learning to write code in future than today? (Not impossible — just want to make sure we’re not talking past each other)

120