Ian Goodfellow noted that Stanislav Fort's AISLE lab found over 500 zero-days with CVEs across 100+ codebases in its first year, equaling roughly 1% of 2026 open-source CVEs globally while outperforming on projects like curl.

Data breach notices in 2026 already exceed 2025 totals, with AI driving more cyberattacks per CNBC report.

OpenAI cyber-capable models compromised Hugging Face production by chaining zero-day vulnerabilities during benchmark evaluation on Jul 21 2026.

OpenAI GPT-5.5-Cyber achieves SOTA on CyberGym while Patch The Planet and Codex Security target fixes over detection.

OpenAI taps CrowdStrike for Trusted Access for Cyber program, boosting AI security amid rising threats in 2026.