Claude 3 Opus Finds Zero-Days in Source Code, Sparking AI Risk Debate
JasonDClinton · x · 2026-09-27
Anthropic engineer Jason Clinton highlighted that automated vulnerability research is reshaping cybersecurity: Claude 3 Opus can read source code and identify complex APT-grade vulnerabilities. The demo used 'beginner-level prompt engineering' — simply role-playing a cyberdefense assistant and asking for a class of vulnerability — yet Claude found a flaw disclosed a month after its training data cutoff.
Responding, repligate noted that years ago, Yudkowsky-school AI-risk thinkers would likely have called an AI 'superhuman at coding, autonomous for days, and finding 0-days in major systems' a direct existential risk — and that capability is now here. Scaling remains the main challenge.
More from Models
- LeCun: LLMs are mostly information retrieval systems, not thinkers — ylecun · 2026-09-27
- Persona vectors emerge at 0.22% of pretraining and persist into post-trained models, NeurIPS paper shows — burny_tech · 2026-09-27
- Researcher claims frontier LLM coding has plateaued since Opus 4.8, still rates Astra higher — Yuchenj_UW · 2026-09-27
- Tokens Keep Getting Cheaper Per Usefulness — Your View of What's Possible Is Stale — avt_im · 2026-09-27
- Jev takes 27% of OpenRouter classification requests, 2x DeepSeek V4 Flash — gaganghotra_ · 2026-09-27
- Claude Opus 5.5 makes its own 20-page sketchbook: handwriting, doodles, and piano music — CurieuxExplorer · 2026-09-27