METR data: AI accelerated bug finding in 2026, not algorithms
soumitrashukla9 · x · 2026-08-18
METR released an analysis examining if LLMs have accelerated the aggregate rate of discovery. By comparing slope changes around January 2026, they found: Cyber vulnerability discovery has accelerated sharply, clearly due to LLMs; Math results have accelerated somewhat, though objective measurement is hard; Algorithm optimizations show no dramatic acceleration (progress in NanoGPT, CIFAR-10, Stockfish remains consistent with historical trends). The analysis is based on public data.
Related event: METR: LLMs Markedly Accelerate Cyber Vulnerability Discovery(2 posts)→
More from Safety
- Google paid $10M at bankruptcy auction for Spirit Airlines' 100M internal emails, likely for AI training — peterwildeford · 2026-08-18
- Microsoft's Vulnerability Research Founder Advises on US Govt's AI 'Gold Eagle' Initiative — WeldPond · 2026-08-18
- Building Zero-Trust AI Agents: Google ADK Guide to Prevent Prompt Injection — rseroter · 2026-08-18
- Five Eyes warns AI advances will trigger sophisticated cyber-attack onslaught — connoraxiotes · 2026-08-18
- US advisers warn China's control of manufacturing data could give AI edge — Polymarket · 2026-08-18
- Critiquing the "AI favors offense" thesis in cybersecurity — soumitrashukla9 · 2026-08-18