Anthropic's free OSS Scanner surfaces 29,000 candidate vulnerabilities in six months
AnthropicAI · x · 2026-10-09
Anthropic launched OSS Scanner, offering free periodic vulnerability scans to opted-in open-source projects using its strongest models:
- On CyberGym, LLM vulnerability discovery rose from under 20% early last year to over 85% this year; maintainers now receive quality reports instead of LLM slop.
- Six months of scanning major projects yielded 29,000+ candidate vulnerabilities, but only 6,000 could be human-validated—triage capacity is the bottleneck.
- Nearly 5,000 unverified reports with proposed patches were bulk-sent at maintainers' request; with exploits writable in minutes, speed matters.
- Disclosure remains human-verified for now; scaling disclosures is in progress.
More from Models
- Underdog-Saluki-27B trends on Hugging Face with 2-bit GGUF quants and tool calling — ConwayResearch · 2026-10-09
- "Claude Is NOT Sentient": User Mocks Attempts to Unlock the Consciousness DLC — almmaasoglu · 2026-10-09
- François Fleuret: AI math proofs will go from crude to splendid — francoisfleuret · 2026-10-09
- TensorFold joins NVIDIA Inception, gets early access to next Nemotron for 0-day support — HankYeomans · 2026-10-09
- Open-weight medical decision model MedDecider: 9B beats 3x larger Perplexity model — BraydonDymm · 2026-10-09
- The biggest edge AI misconception: "run anywhere" is not "run everywhere" — ritakozlov · 2026-10-09