AISLE targets real zero-days and claims parity with frontier AI systems
stanislavfort · x · 2026-07-28
AISLE argues that most AI-for-cyber work is still evaluated on toy tasks and synthetic benchmarks.
- The project goes after previously unknown zero-days in some of the world’s most secure code.
- The post claims it is matching and even beating frontier systems.
- It emphasizes that the models used are widely available and open-source-derived rather than proprietary only.
More from Safety
- A copyright debate says unlicensed AI training is value creation, not value transfer — dhadfieldmenell · 2026-07-28
- Levanto says its guardrail model matches GPT-5 on AgentHarm and runs 5x faster — w1kke · 2026-07-28
- AI copyright is a licensing problem, not an existential-risk problem — dhadfieldmenell · 2026-07-28
- Rights holders will need scalable ways to collect training royalties — dhadfieldmenell · 2026-07-28
- Study finds frontier LLMs can reason through filler tokens invisible to CoT — dair_ai · 2026-07-28
- Sequoia Analysis: Cyera Acquires Oasis, Betting on AI Agent Security — Sequoia Capital · 2026-07-28