AISLE finds 500 plus zero days with small LLM systems
According to a post by Ian Goodfellow citing AISLE founder Stanislav Fort on X, AISLE discovered 500 plus zero days with CVEs across 100 plus codebases in a year by boosting small, fast LLMs with strong surrounding systems.
SourceAnalysis
According to Ian Goodfellow on X, Stanislav Fort's AI security research lab AISLE discovered more than 500 zero-days with CVEs across over 100 codebases in its first year.
This output equals roughly 1 percent of 2026 open-source CVEs globally from one team, and on hardened codebases such as curl the lab has repeatedly matched or outperformed frontier AI labs in zero-day discovery, showing that the surrounding system matters more than model size and that small, cheap, fast models can deliver valuable security results.
Ian Goodfellow
@goodfellow_ianGAN inventor and DeepMind researcher who co-authored the definitive deep learning textbook while championing public health initiatives.