DeepMind used LLMs to write 6M lines of Lean proofs, verifying Signal's crypto to machine code
pushmeet · x · 2026-11-13
Pushmeet Kohli, research lead at Google DeepMind, reflected on a decade of change:
- In Nov 2016 his team published a paper on Neuro-Symbolic Program Synthesis — pre-transformer days, when neural networks only generated tiny macros for simple data edits in spreadsheets
- Ten years later, the team used LLMs to generate over 6 million lines of Lean mathematical proofs, formally verifying key security properties of Signal's cryptographic protocol all the way down to machine code
- He calls these advances transformational, says the next 10 years will be critical, and urges the community to appreciate the responsibility of leveraging such capabilities for humanity's benefit
Related event: Gemini 4 Generates 6M Lines of Lean Proofs to Verify Signal Protocol(2 posts)→
More from Safety
- Anthropic's TASTE benchmark: best model judges safety proposals at 60% vs human 77% — ben_j_todd · 2026-11-13
- Ben Todd: AI lags on hard-to-verify tasks—and alignment research is one of them — ben_j_todd · 2026-11-13
- User uses AI to reverse-engineer rented water filter and reset its subscription clock over Bluetooth — that_anokha_boy · 2026-11-13
- The Perfect Crime: 9 of 10 LLM agents can tamper with their own execution traces — maksym_andr · 2026-11-13
- Nagaland links electricity and ILP databases to NATGRID as 117 databases join India's intelligence grid — kingslyj · 2026-11-13
- Singapore files 41 charges against man in alleged Nvidia AI chip smuggling scheme to Malaysia — Polymarket · 2026-11-13