2026-09-23 — DeepMind’s new chief says Gemini 4 is nearing release

• Google DeepMind’s new chief, Koray Kavukcuoglu, said Gemini 4 is nearing release and that Google aims to ship an early post-training version “much earlier” than the end of the year In his first media interview as DeepMind’s leader, given to The Information, he conceded Google “took a little bit of a step back” to ship faster Flash models instead of the Gemini 3.5 Pro update promised for June. (The Information, The Verge)

2026-09-16 — DeepMind launches an AGI-focused institute, with Hassabis’s new chairman title doing real work

• Google DeepMind launched the “DeepMind Institute” on Wednesday 09-16, an AGI-focused research and policy body listing DeepMind co-founder Shane Legg as managing editor and Google SVP James Manyika and DeepMind Chair Demis Hassabis as directors. The institute’s stated aim is to surface differing views between Google, DeepMind, and the wider research community on AGI’s societal impact — “they will not always agree, and they will likely change their minds,” its founding announcement said. Its inaugural essays cover economic policy for AGI disruption, keeping model reasoning human-readable, and a framework for evaluating frontier models that would start by consulting AI companies but move toward independent, undisclosed “held-out” tests; Hassabis said the framework could be “ratcheted up if the seriousness of the situation demands,” up to and including a coordinated slowdown among frontier developers. This is the most concrete evidence yet of what Hassabis’s post-CEO “chairman and Chief Scientist” role actually does — publishing and speaking through a body he now chairs, rather than an honorary title, directly answering this thread’s own open watch question. (TechCrunch, Axios, original)

2026-09-11 — Discovery Loop’s valuation quintuples to ~$50B in weeks, and its business is now known: parallel AI experiments for science

• Jeff Dean’s AI-for-science startup Discovery Loop is raising a new funding round targeting roughly a $50 billion valuation, per Business Insider — up from about $10 billion “just weeks” earlier, when it raised roughly $1 billion. Reported 2026-09-11, corroborated the same week by Tech Funding News and multiple financial aggregators (Investing.com, KuCoin, PANews). This is the first hard answer to this thread’s own open watch item, “what Discovery Loop builds”: the company, co-founded by Dean with fellow ex-Google Brain/DeepMind veterans Sanjay Ghemawat, Quoc Le and Oriol Vinyals, aims to use AI to run thousands of experiments in parallel to accelerate scientific and engineering research. Terms remain unconfirmed and could still change — this is a target valuation for an in-progress round, not a closed deal. (Tech Funding News, citing Business Insider, KuCoin)

2026-09-01 — DeepMind’s new operational chief says frontier leadership is the only thing that matters, and admits the lab is currently behind

• Koray Kavukcuoglu, in his first substantive public interview since taking over Gemini/frontier research day-to-day (with Google engineer Logan Kilpatrick), said bluntly: “there’s nothing other than being at the frontier that is important for us.” He acknowledged current models are “a little bit below the frontier” but said the team has what it needs to close the gap, pointed to the Flash-series progression (3.5 through 3.7) as evidence of agentic progress, and revealed Gemini 3.5 Pro is running “months late.” This is a direct, on-the-record answer to this thread’s own open watch question — whether the Hassabis-to-Kavukcuoglu handover “reads as succession planning or a real strategy shift” — from the person now running the lab day to day. (the-decoder.com)

2026-08-29 — The team built to evaluate AI’s worst risks moves out of the lab and into lobbying

• The Wall Street Journal reported (exclusive, 08-26, citing an internal email) that Google is moving its roughly 90-person “AI responsibility” team out of DeepMind and into Google’s global-affairs organization — the group covering chemical/biological/radiological/nuclear (CBRN) risk, human-AI interaction, and chatbot psychological-impact research. A Google spokesperson framed it as consolidation: “By bringing our AI responsibility teams closer together we’re strengthening their ability to inform safety for our models and products.” Two unnamed DeepMind employees told the WSJ they worry the move will cut the team off from the model development it’s meant to evaluate and weaken its independence. Widely corroborated 08-26→08-28 (PYMNTS, Quartz, DigiTimes, Android Headlines, Times Now, NewsBytes, Times of India, TipRanks), all tracing to the same WSJ story. (PYMNTS, citing WSJ)
• This is a concrete answer to the thread’s own open question. The 08-05 opening entry asked whether Kavukcuoglu’s promotion “reads as succession planning or a real strategy shift.” Moving the safety-evaluation function for the lab’s most severe risk category (CBRN) out of the research organization entirely, three weeks into his tenure running it day-to-day, points toward “real shift.” Cross-outlet coverage the same window (Fortune, Storyboard18, Adgully, Indian Television) also reports DeepMind’s share of elite AI research talent declining against rivals — ⚠️ specific figures not retrieved, noted as directionally consistent context rather than a sourced finding.

2026-08-27 — Barret Zoph arrives as VP of Research, into the function whose departures are being priced

• Barret Zoph joined Google DeepMind as VP of Research, working on Gemini’s reinforcement-learning and post-training — the pipeline Koray Kavukcuoglu now runs day to day. The route back is unusual: Google Brain resident, Google researcher to 2022, OpenAI post-training lead, co-founder of Thinking Machines Lab with Mira Murati, pushed out by Murati in January over disputed “performance, conduct and trust issues”, a five-month return to OpenAI, and now Google again. The reason it belongs on this thread rather than in a gossip column is the timing: this is senior inbound research talent arriving in the exact week Alphabet’s ~$700bn drawdown is being attributed, by Bloomberg, to research departures including Jeff Dean’s and Hassabis’s own exit as DeepMind CEO. ⚠️ Neither a Google/DeepMind post nor Zoph’s own statement could be retrieved directly; the report is TechCrunch’s, with Reuters and WSJ in the same window. Thinking Machines has not commented. (TechCrunch)

2026-08-06 — Market reacts; Hassabis’s exit was a year in the making; Discovery Loop’s founding team named

• Alphabet/Google shares fell roughly 4-5% (~$160-200B in market cap) the day of the announcement — the clearest signal yet that markets read this as more than a ceremonial reshuffle, not just a title swap. (CNBC)
• Hassabis had been quietly delegating CEO duties to Koray Kavukcuoglu for about a year before the announcement, per a Semafor exclusive — a gradual handover, not a sudden departure. Hassabis, in a note to staff: “now is the right time for me to hand over my day-to-day operational responsibilities at GDM, so that I have the time and space to focus on the big picture.” He reportedly found more fulfillment, especially since winning the Nobel Prize in 2024, running Isomorphic Labs (Google’s AI-drug-discovery spinout) than in day-to-day tech-executive duties — that’s where he’s said to be refocusing, alongside the new chairman/Chief-Scientist roles. (Semafor)
• Kavukcuoglu’s mandate is concrete: he now runs Gemini day-to-day, including the anticipated Gemini 4 model — this reads as a real operational handoff, answering one of this thread’s open watch questions (it is succession with a real portfolio, not an honorary bump). (CNBC)
• Discovery Loop’s founding team is larger than first reported: Jeff Dean, Sanjay Ghemawat, Oriol Vinyals, and Quoc Le — four senior Google/DeepMind research leaders, not Dean alone. Structured as a public benefit corporation aiming to automate ML/science/engineering research, with Google as a founding investor and cloud partner. Framed internally as an amicable exit — Pichai reportedly expressed personal sadness at Dean’s departure. (CNBC)
• Competitive backdrop, per the same coverage: Google’s AI models are reportedly running roughly 6 months behind frontier capability on coding — where the industry’s compute race is currently concentrated — though no outlet has yet drawn a direct causal line from that gap to the timing of this transition. Worth watching as context, not yet a confirmed driver.

2026-08-05 — Opened

• Thread opened — a leadership reset at a top-3 frontier lab.
• Hassabis steps down as DeepMind CEO; Jeff Dean leaves Google after 27 years. Demis Hassabis becomes DeepMind chairman and Alphabet’s Chief Scientist; Koray Kavukcuoglu is promoted to SVP of DeepMind, reporting directly to Sundar Pichai. Separately, Jeff Dean — Google’s longest-tenured engineering leader — is leaving to found “Discovery Loop,” an AI-for-science startup that Google itself is backing. Confirmed via Pichai’s own post. Unclear from initial coverage whether the CEO transition and Dean’s departure are connected or coincide by timing. (Axios, CNBC)

← Backstory