Timeline
- Google releases Gemini 4 Argon: it solved all 12 end-to-end simulated cyberattack scenarios (the previous generation solved 5); because of elevated capabilities in the cyber and biological domains, under the Frontier Safety Framework it ships with new safeguards — three layers of real-time protection — and the version without cyber guardrails went first only to trusted defenders in the Fairwind ProgramCybersecurity · AI safety · blog.google ↗
- Science publishes DeepMind’s paper on the AlphaProof Nexus agent: it autonomously resolved 9 of 353 open Erdős problems and proved 44 of 492 conjectures from the On-Line Encyclopedia of Integer Sequences, with every proof checked in Lean; in the same issue mathematicians Jeremy Avigad and Matthew Ballard write the commentary “AI for research mathematics has arrived”Science · AI in mathematics · doi.org ↗
- FLI AI Safety Index: the best grade among 9 companies is C+ for Anthropic; in existential safety none scores above C−; leading labs have weakened their commitments to stop at “red lines”Laws & regulation · AI safety · futureoflife.org ↗
- Google unveils its eighth-generation TPU as two chips: TPU 8t for training models and TPU 8i for running themInternet & infrastructure · Business & markets · AI accelerators · blog.google ↗
- Google Research, Google DeepMind and MIT compare 260 agent configurations: versus a single agent, multiple agents range from +80.8% on decomposable tasks to −70% on sequential planningSoftware development · Science · AI agent orchestration · arxiv.org ↗
- Google DeepMind unveils CodeMender, an agent that finds and fixes vulnerabilities; 72 of its fixes have already been accepted into open-source projects, all reviewed by humansCybersecurity · Software development · AI vulnerability discovery · deepmind.google ↗
- DeepMind's Gemini Deep Think officially achieves gold at the International Mathematical Olympiad: 5 of 6 problems, 35 of 42 points, graded by IMO coordinatorsScience · Education · AI in mathematics · deepmind.google ↗
- Google: Big Sleep, drawing on threat intelligence, found the SQLite vulnerability CVE-2025-6965, “known only to threat actors”; according to the company, the first case of an AI agent directly foiling an exploitation attemptCybersecurity · AI vulnerability discovery · blog.google ↗
- More than 40 researchers from OpenAI, Google DeepMind, Anthropic, Meta and other organizations: monitoring models' chains of thought is a rare but fragile opportunity for AI safetyScience · AI safety · arxiv.org ↗
- DeepMind's AlphaEvolve multiplies 4×4 complex matrices with 48 scalar multiplications, improving on Strassen's 1969 algorithm; on 50+ open problems it improves the best known solutions in 20% of casesScience · Software development · AI in mathematics · deepmind.google ↗
- The Big Sleep agent (Google Project Zero and Google DeepMind) finds an exploitable memory-safety bug in SQLite; according to Google, the first public example of such a find by an AI agent in widely used softwareCybersecurity · Software development · AI vulnerability discovery · projectzero.google ↗
- The Nobel Prize in Physics goes to Hopfield and Hinton for neural networks; the next day the chemistry prize goes to Hassabis, Jumper (AlphaFold) and BakerScience · Deep learning · aihub.org ↗
- DeepMind's AlphaProof and AlphaGeometry 2 solve 4 of 6 International Mathematical Olympiad problems — 28 points, silver-medal standardScience · Education · AI in mathematics · deepmind.google ↗
- Stanford AI Index 2024, citing Epoch AI estimates: training compute for GPT-4 cost about $78 million and for Gemini Ultra about $191 million at cloud rental pricesBusiness & markets · Large language model · hai.stanford.edu ↗
- Google DeepMind (Nature): FunSearch, a language model paired with an automated evaluator, finds large cap set constructions beyond the best-known onesScience · AI in mathematics · nature.com ↗
- Google unveils Gemini (Ultra, Pro, Nano); Gemini Pro starts powering BardSoftware development · Business & markets · Large language model · bostonglobe.com ↗
- Center for AI Safety: mitigating the risk of extinction from AI should be a global priority alongside pandemics and nuclear war; signatories include Hinton, Bengio and the heads of OpenAI, Google DeepMind and AnthropicScience · Geopolitics & defense · AI safety · safe.ai ↗
- DeepMind presents Chinchilla: with the same compute budget, a 70-billion-parameter model trained on more text outperforms the 280-billion-parameter Gopher; model size and the number of training tokens should be scaled equallyScience · Large language model · arxiv.org ↗
- The AlphaFold database: hundreds of thousands of protein structures in open accessScience · Health & medicine · Protein folding · mail.ncmir.ucsd.edu ↗
- AlphaFold 2 solves the protein structure prediction problem at CASP14Science · Health & medicine · Protein folding · siliconangle.com ↗
- AlphaStar beats professionals at StarCraft IIScience · Sport · Reinforcement learning · engadget.com ↗
- Google BERTScience · Large language model · arxiv.org ↗
- AlphaZero masters chess from scratch in hoursScience · Sport · Reinforcement learning · arxiv.org ↗
- The paper “Attention Is All You Need”: the Transformer architectureScience · Transformer · arxiv.org ↗
- AlphaGo beats Lee Sedol 4–1Science · Culture & arts · Reinforcement learning · koreatimes.co.kr ↗
- Google buys DeepMindBusiness & markets · Deep learning · thedomains.com ↗
- DeepMind: a neural network learns to play Atari games from screen pixelsScience · Reinforcement learning · arxiv.org ↗
- A Google neural network teaches itself to recognize cats in YouTube videosScience · Deep learning · 9to5google.com ↗
- The Google Brain project startsScience · Deep learning · en.wikipedia.org ↗
- DeepMind is foundedScience · Business & markets · Deep learning · en.wikipedia.org ↗