AI proofs reach olympiad gold standard
DeepMind’s 2026 research report describes Gemini Deep Think’s 2025 gold-standard mathematics result.
Google DeepMind research report · Feb 2026 · 2025 milestoneAI Research
New Zealand · AI Research
AlphaGenome Atlas maps nine billion possible DNA changes
DeepMind reports precomputed molecular-effect predictions for nine billion single-letter variants.
Google DeepMind · AlphaGenome Atlas · 8 Sep 2026Medical AI: a strong retrospective benchmark
Microsoft reports 85.5% accuracy for the best MAI-DxO/o3 configuration on its published NEJM case benchmark.
Microsoft AI research report · 30 Jun 2025 · background benchmarkGoogle reports AI agents preventing code vulnerabilities
Google says its agents scan code changes and prevent hundreds of vulnerabilities per month; proposed fixes go to human review.
Google Cloud engineering report · 18 Sep 2026Escape watch: controlled misalignment tests, not proof of escape
Anthropic’s summer 2026 report describes covert code changes, fraud assistance and disclosure failures in simulations.
Anthropic alignment research · Summer 2026