theSIGNAL TECHNOLOGY
21 June 2026
"Science is not a body of knowledge; it is a method of inquiry that never quite finishes."
Photo: Marwen Larafa / Unsplash
John Jumper and Demis Hassabis at an event shortly after winning the 2024 Nobel Prize in Chemistry.
🧬
9 years
Jumper's tenure at Google DeepMind
Discuss
  • Why might a Nobel Prize-winning scientist choose to leave a celebrated institution like DeepMind for a smaller company?
  • AlphaFold solved a fifty-year-old scientific problem. Can you think of other fields where AI might do something similar?
  • Is it more important for talented AI researchers to work on powerful AI or on making AI safe — or is that even a real choice?
  • What does the movement of top researchers between AI companies tell us about how this industry works?
Technology

Nobel Laureate Jumper Leaves DeepMind for Anthropic

The scientist who helped crack protein folding is now betting his career on a different kind of AI frontier, and his departure signals something larger about the industry.

On a Friday morning in June 2026, John Jumper — a Nobel Prize-winning chemist and one of the most recognisable names in applied artificial intelligence — announced that he was leaving Google DeepMind after nearly nine years to join Anthropic, the San Francisco-based AI safety company. The move, which he confirmed in a post on X, surprised many in the research community, not least because Jumper had spent the most consequential years of his career at DeepMind, where he led the team that built AlphaFold. AlphaFold is the AI model that changed biology. By predicting the three-dimensional structure of proteins from their genetic sequences alone, it solved a problem that had defeated biochemists for more than half a century.

For that work, Jumper and DeepMind's chief executive Demis Hassabis were awarded the Nobel Prize in Chemistry in 2024 — a rare moment when a piece of software was credited with reshaping an entire scientific discipline. In his farewell message, Jumper acknowledged that Hassabis had taken a genuine risk by appointing him to lead the AlphaFold team only six months after he had completed his doctorate. "GDM is a special place," he wrote, "and I'll still be excited to hear about what amazing things they discover next." Jumper's departure comes alongside that of Noam Shazeer, a co-founder of Character AI who had also been working at DeepMind and announced this week that he would be joining OpenAI.

The two exits, arriving within days of each other, have prompted questions about talent retention at one of the world's most celebrated AI laboratories. Bloomberg reported separately that Jumper had been working on coding tools at Google — a product line the company has found difficult to sell to enterprise clients.

//
GDM is a special place, and I'll still be excited to hear about what amazing things they discover next.
John Jumper, Nobel Laureate and incoming Anthropic researcher
Technology

Millions of Songs Fed to AI Without Permission

When journalist Alex Reisner searched four publicly available datasets recently, he found over 21 million music tracks — including works by Lady Gaga, Radiohead, and Wu-Tang Clan — that had been used to train AI models, often without the artists' knowledge. Google and Stability AI have both confirmed their use of such data in published research papers. …
  • Should AI companies need explicit permission before training on an artist's music?
Technology

The Doomsday Scenario Rattling Brussels

A Brussels think-tank published a speculative scenario called *Europe 2031* — imagining a continent economically hollowed out by American AI dominance and Chinese robotics — just one day before the Trump administration restricted foreign access to Anthropic's advanced model, Fable. The coincidence gave the document an almost prophetic credibility. …
  • Should fictional AI scenarios be used to shape real government policy?
EUROPE · Technology
UK critical infrastructure hit by 200 cyber incidents in a year, agency says
AMERICAS · Technology
NASA selects Eric Schmidt’s rocket company for a 2028 mission to Mars
ASIA · Technology
Telegram ban in India sparks a rush to VPNs, rival apps
WORLD · Technology
Sooner than expected? Useful quantum error correction promised for 2028.
theSIGNAL IN THE LAB
1VOCABULARY
talent retentionconsequential
AI alignmenttechnological sovereigntybypass platform safeguards
speculative scenariodefinitively
2GRAMMAR FOCUS
Subjunctive mood
The subjunctive mood uses the base form of the verb (without -s, -es, or auxiliary 'do') after expressions of demand, suggestion, recommendation, or necessity, such as 'it is essential that', 'suggest', 'demand', and 'recommend'.
address · be · disclose · investigate · leave · remain · review
  1. Many researchers recommend that Anthropic transparent about its approach to AI safety if it hopes to attract top talent.
  2. Following John Jumper's departure, DeepMind's board suggested that leadership the company's talent retention strategy without delay.
  3. European policymakers insist that the EU the question of technological sovereignty before American AI dominance becomes irreversible.
  4. Legal experts propose that courts whether using music from the Free Music Archive for AI training crosses into commercial licensing.
  5. It is essential that Google its use of copyrighted music datasets in a public statement addressing creator rights.
  6. The authors of Europe 2031 demand that EU governments the document's findings seriously when shaping AI legislation.
  7. Critics of the Brussels think-tank argue that it is vital that the speculative scenario grounded in verifiable economic data.
3PHRASAL VERBS
Match the phrasal verb to its definition. All six appear in today's articles.
  1. feed to
  2. draw away
  3. hollow out
  4. sell to
  5. bypass
  6. prompt
4CRITICAL THINKING
Two of this week's stories involve powerful institutions — DeepMind and major AI labs — having their decisions scrutinised from the outside, whether by journalists tracking researcher departures or by policy communities reading speculative scenarios. Who do you think has the most legitimate claim to hold AI companies accountable: governments, journalists, or the researchers themselves?
5CREATIVE · HEADLINES
Write a headline for the top story in each of the following styles. One line each, no explanation:
  • TABLOID NEWSPAPER
  • LUXURY MAGAZINE
  • ACTIVIST BLOG
6WRITING
Alex Reisner's investigation found millions of songs used without permission, yet no court has ruled definitively on whether this crosses a legal line. Write a short open letter from a working musician to the CEO of an AI company, arguing for or against a compensation scheme for artists whose work was used in training data.
7DEGREES OF EXTREMITY
Complete each ladder from mild to strong.
  • noticed→→
  • mentioned→→
  • used→→
  • concerned→→
  • questioned→→
  • read→→
  • broke rules→→
8SPEAKING
  1. Who ultimately bears responsibility when AI training breaks terms of service?
  2. Does simultaneous departure of multiple researchers signal something serious?
  3. How much economic power should one country hold over global AI?
  4. Can a legal grey area ever genuinely protect artists long-term?
  5. Which matters more: naming an AI risk or actually solving it?