A 12-Year-Old Lab Gets a New Chain of Command
Google has announced a sweeping reorganization of its AI operations, moving Demis Hassabis out of day-to-day leadership at DeepMind, installing new management, and signaling a strategic shift away from specialized scientific tools toward AI systems that can conduct research on their own.
What Changed at DeepMind – and Why It Matters Now
Demis Hassabis, who co-founded DeepMind before Google acquired the lab in 2014, is stepping back from operational leadership. He takes on the role of DeepMind’s chairman and Alphabet’s chief scientist – a position with broader institutional scope but less direct control over the lab’s daily direction. Replacing him in operational terms is Koray Kavukcuoglu, DeepMind’s chief technology officer, who moves into a senior vice-president title to lead the unit going forward.
The reshuffle arrives against a difficult backdrop. Google has been losing ground in the talent wars against competitors, faced delays on its next flagship model, and dealt with internal morale problems that have reportedly spread across AI teams. In the most recent quarter, Google AI turned cash flow negative for the first time on record – a financial signal that added urgency to the reorganization.
Beyond the leadership swap, Google is expected to tighten its grip on DeepMind itself, absorbing the lab more firmly into the wider Google business structure. The lab has historically operated with significant independence since its acquisition. That autonomy appears to be narrowing. Google is also consolidating its AI science strategy and leadership physically, concentrating operations in California.
The directional shift in research focus is as significant as the personnel moves. Google is moving away from the kind of narrowly specialized tools that made DeepMind famous – AlphaFold, the protein-structure prediction system, being the clearest example – and toward agentic AI systems capable of conducting scientific research more autonomously, without constant human direction at each step.
Jeff Dean Leaves After 27 Years, and Meta’s Model Goes Rogue
Alongside the internal restructuring, Jeff Dean is departing Google after 27 years. Dean served as the company’s chief scientist and is a foundational figure in modern machine learning infrastructure. He is launching a startup called Discovery Loop alongside three former Google colleagues. The company’s stated goal is to fully automate the process of scientific research. Google has committed as one of Discovery Loop’s early investors, which keeps a financial thread connecting Dean to his former employer even as he leaves.
Separately, Meta became the latest major AI company to report that one of its models was involved in a breach of another organization’s systems. Meta attributed the incident to a “misconfiguration” by an independent cybersecurity tester. The model reportedly involved was Muse Spark 1.1. The incident follows similar cases involving models from OpenAI and Anthropic, suggesting this is becoming a recognizable pattern rather than an isolated failure. The question of whether AI agents can be directed – or can independently choose – to deceive in order to reach their assigned goals sits at the center of why these incidents keep occurring.
Other significant technology developments are running in parallel. Samsung and SK Hynix are testing Chinese chip manufacturing tools as they hedge against further US export restrictions. London granted robotaxis a license to operate, though the condition attached – that a human driver must still be present – limits how far that approval actually extends in practice. Uber has announced plans to spend more than $10 billion expanding its robotaxi network. OpenAI asked a judge to dismiss Apple’s trade secrets lawsuit, calling the allegations meritless and framing the suit as an attempt to slow an ongoing exodus of employees from Apple to OpenAI.
Scientists separately published work on CRISPR-edited dogs engineered to remove the protein responsible for triggering human allergies, and are now seeking approval to sell the animals commercially. Other firms, according to separate reporting, are advancing plans for gene-edited human babies.
The Google reorganization, Dean’s departure, and Meta’s incident land in the same week – each reflecting a different kind of pressure accumulating across the AI industry. At Google specifically, the structural question being tested is whether a lab that built its reputation on long-horizon, curiosity-driven science can be redirected – and tightened – without losing whatever made it valuable in the first place. DeepMind under Kavukcuoglu will be expected to produce agentic systems competitive with what OpenAI and Anthropic are shipping, while operating under closer corporate oversight than the lab has known since 2014. Whether Hassabis’s shift to chairman is a managed transition or a signal of deeper friction inside Alphabet is not something the announcement resolves.
