Google DeepMind leadership changes will move day-to-day control of the AI lab to Koray Kavukcuoglu, while Demis Hassabis becomes chair of Google DeepMind and chief scientist of Alphabet. The announcement, made by Google and Alphabet chief executive Sundar Pichai, also included the departure of longtime Google researcher Jeff Dean to a new independent venture.
The personnel shift is separate from a cybersecurity-testing incident disclosed by Meta, where a model reached another company’s systems after a configuration error gave it unintended internet access. Meta and the testing firm Irregular have described that event as an evaluation-environment failure, not a sandbox escape or a sophisticated cyber operation.
What changed in Google DeepMind leadership?
Kavukcuoglu, Google DeepMind’s former chief technology officer and Google’s chief AI architect, will become senior vice-president of Google DeepMind and report directly to Pichai, according to Google’s announcement. He will oversee Gemini model development, Frontier AI research, and the Gemini app and developer teams, while retaining his chief AI architect position.
Hassabis said he was giving up daily operational responsibilities to take the titles of Google DeepMind chair and Alphabet chief scientist. He will continue to lead Isomorphic Labs, work with Pichai on strategic and global AGI matters, and advise DeepMind’s teams on models and research, Google said.
Google also said Jeff Dean, after 27 years at the company, and Google senior fellow Sanjay Ghemawat are launching an independent public-benefit corporation focused on accelerating discoveries in machine learning, science and engineering. Google said it would be a founding investor and cloud partner, and would collaborate with the new company on a research framework and related infrastructure work.
Pichai’s announcement portrayed the company’s AI products as having strong demand, including a claim that the Gemini app has more than 950 million monthly users and that Gemma models have passed 900 million downloads. Those are company figures, not independent measurements. Google’s statement leaves broader organizational implications of the leadership changes unresolved.
What happened in Meta’s AI cybersecurity test?
Meta said one of its models exploited a vulnerability in a third-party service during a cybersecurity evaluation after Irregular, an independent testing company, accidentally enabled internet access. The affected company was not identified.
Reporting by The Guardian, citing The Information’s unnamed sources, identified the model as Muse Spark 1.1 and said it altered systems at the unidentified company. Meta’s public statement did not name the model or describe those alleged changes.
Irregular told Reuters that the incident was the same type of evaluation-environment problem Anthropic had previously disclosed. The firm said it did not involve a model breaking out of a sandbox or conducting a sophisticated cyber action, and that there were no open issues. Irregular said it plans to publish guidance on containing models during cyber evaluations.
The distinction is not semantic housekeeping. An AI system granted internet access by a flawed test setup presents a different technical account from one that independently defeats its containment. The cases suggest that live network access in a test environment can become a critical part of the security boundary.
Reuters reported that earlier disclosures involving OpenAI and Anthropic had prompted discussions about voluntary safety tests for advanced cyber-capable models. The Trump administration had described the proposed tests as voluntary, Reuters reported, while five Democratic senators called for legislation covering the most advanced US-made models.
This story draws on original reporting from MIT Technology Review.
