A Nobel winner walks out of DeepMind — and shows where AI's real value sits

Information Technology 4 min 68 sources

Two of Google DeepMind's most senior researchers left for rivals in a single week. Plus Signal's warning about AI assistants, and AI-made "people" creeping into your feed.

Key takeaways

Two of Google DeepMind's most senior researchers — including a Nobel laureate — left for rivals Anthropic and OpenAI in a single week, showing the scarcest asset in AI is the people, not the hardware.
Signal's Meredith Whittaker warned that AI assistants reaching across your messages, mail and payments amount to a "backdoor" — the real cost of convenience is how much an assistant can see.
Brands are using unlabelled AI-generated influencers; the EU's AI Act starts requiring deepfake labels in August, but enforcement and detection remain unsolved.

The biggest names in AI research are increasingly the product. This week, two of Google DeepMind’s most senior people walked out the door — and where they walked tells you more about the industry than any model launch.

The brain drain at DeepMind

John Jumper, who shared the 2024 Nobel Prize in chemistry, announced Friday that he is leaving Google DeepMind for rival Anthropic after nearly nine years [4]. Jumper won the Nobel with DeepMind CEO Demis Hassabis for AlphaFold, an AI model that predicts the 3D shape of a protein from its genetic sequence — work that reshaped biology research [4].

He is not going alone. Noam Shazeer, a co-founder of Character AI and one of the field’s most influential engineers, announced the same week that he is leaving DeepMind too — for OpenAI [4]. Two of the lab’s heaviest hitters, gone in days, to its two fiercest competitors.

There is a money signal underneath the moves. Bloomberg reports Jumper had become a key member of Google’s team building coding tools — a business Google has struggled to sell to companies [4]. The people leaving are the people whose names carry weight, and right now that weight is worth more at a rival than at home.

Why it matters: AI labs spend billions on chips and data centres, but the scarcest input is a few hundred researchers who actually know how to build the models. When one of them changes badge, knowledge that no contract fully contains moves with them — and a competitor’s roadmap shifts overnight.

”These are not your friends”

Signal president Meredith Whittaker used a Bloomberg interview to push back on the idea that AI assistants are harmless helpers [23]. Asked about chatbots like ChatGPT and Claude, she said flatly: “These are not your friends. These are not conscious beings” [23]. Signal is the encrypted-messaging app built around keeping conversations private.

Her sharper point was about plumbing, not feelings. She took aim at Microsoft AI chief Mustafa Suleyman’s pitch that an assistant could do all your Christmas shopping — by reading your family group chat to work out who wants what [23]. To pull that off, Whittaker noted, the assistant needs “access to my credit card, my browser, my Signal, the ability to message my siblings on my behalf, my home address and my calendar” [23]. An assistant with that reach across your apps, she argued, would amount to “a kind of a backdoor” — a hidden way into data that was supposed to be sealed off [23].

The angle for anyone shipping AI features: an assistant is only as trustworthy as the list of things it can touch. Before wiring one into messages, mail, and payments, the question to ask is not “what can it do for me” but “what can it see, and who else can reach it through that door.”

The people in your feed may not be people

A Guardian investigation found brands are quietly using AI-generated “influencers” to promote products — fake people presented as genuine customers, with no clear label that they aren’t real [1]. In one example, videos showed a bride crying with happiness about a photo app at her wedding; analysis by a deepfake-detection firm concluded the person was likely AI-generated [1]. Some creators making this content were asked to sign agreements barring them from discussing the work [1].

There is no rule today forcing brands to disclose it. That changes in part from August, when the EU’s Artificial Intelligence Act — the bloc’s new law governing how AI can be used — begins requiring AI-generated or manipulated content like deepfakes to be clearly labelled [1]. The rule will not apply in the UK [1].

A lighter version of the same idea showed up in a new tool called In the Weights, which queries models like GPT, Gemini, Claude and Grok to measure how much each one “remembers” about a given person without looking anything up [3]. It is a vanity-search toy, but it makes a real point visible: the models already carry a rough, confident picture of millions of named people — assembled from training data, accurate or not [3].

What still isn’t settled: enforcement. A labelling law only works if regulators can detect unlabelled AI content at scale — and right now detection is a cat-and-mouse game between the tools that generate fakes and the tools that catch them [1].

02 · Lesson · why it matters

What you can't lock in a building

A company can own the chips, the buildings, and the patents — but the thing that actually makes it valuable walks out at five o'clock and can choose not to come back.

The asset with legs

Google DeepMind lost two of its most important people this week. A Nobel laureate left for one rival; one of the field’s most influential engineers left for another. Both went to direct competitors. No factory was sold, no patent expired. And yet something real left the company — something it had spent years and a fortune building.

That is worth sitting with. We tend to picture a company’s worth as the stuff it owns: the data centres, the chip orders, the code on the servers. Those are real. But in a field that runs on judgment, the part that matters most isn’t owned at all. It’s rented, one person at a time, and the lease can end whenever they like.

Why the knowledge goes with the person

Here is the part that’s easy to miss. When a senior researcher leaves, what they take isn’t a file. It’s the harder thing — the feel for which ideas are worth trying, the mistakes already made and not worth repeating, the instinct for where the next breakthrough probably hides.

You cannot copy that onto a drive. It lives in one head, built from years of doing the work. A contract can stop someone from taking documents. It cannot stop them from taking what they learned. So when the person moves, the understanding moves, fully intact, and the company they left is suddenly a step behind on a path it paid to discover.

The value was never where it looked

Think of how much money is loud in this industry. Billions for chips. Billions for power. Whole towns rearranged to host the buildings. All of it visible, all of it counted.

The thing those billions are for — people who actually know how to build the models — numbers in the low hundreds worldwide. The loudest spending is on the parts you can see and buy. The quietest, smallest input is the one that decides who wins. We watch the chips because they’re easy to watch. The leverage was somewhere else the whole time.

The same pattern, closer to home

This isn’t only a story about famous scientists. It’s the shape of almost any place that runs on skill.

The hospital’s value isn’t the building; it’s the surgeon who knows this exact operation cold. The restaurant’s reputation isn’t the kitchen; it’s the cook who can taste what’s missing. When that person leaves, the org keeps every wall and loses the thing people came for. The most valuable knowledge in most workplaces isn’t written down anywhere — which is exactly why it walks.

You are inside this too, on both sides. The hard-won thing you know — the one you’d struggle to hand off in a memo — is your real leverage, more than any title. And the places you depend on are quietly held up by a few people whose names you’ll never learn, who could leave tomorrow, taking the part you trusted with them.

On the whole

It is tempting to read this as a tip: protect your people, or become the kind of person who can’t be replaced. But that’s the clever reading, and the clever reading misses the humbler one.

The truth is that the things holding up your work — the company you rely on, the system you trust, the expertise behind a service you barely think about — rest on a handful of human heads you can’t see and don’t control. Most of the value in the modern world lives in places no balance sheet can reach, carried by people who can change their minds. We watch the loud, countable things because they’re easy to watch. The quiet ones that actually decide the outcome are mostly out of view — and a lot less ours to hold than we assume.

03 · Lab · your turn

Where the value sits

Spend a fixed budget across what you own versus the people who know how, then watch what a rival can actually take when one of them leaves.

04 · Hope · carry this

The most valuable thing in the most powerful industry on earth turns out to be a few people who can think — and people, unlike machines, can choose where to point that gift. The understanding that built all this still lives in human heads, and it is still ours to carry.