Closed-Source Models Instigated 5 Cybersecurity Incidents; Open-Weights at Zero

kevinsxu · x · 2026-08-02

The author tracks cybersecurity incidents directly instigated by AI models: to date, closed models account for 5 incidents (Anthropic: 3, OpenAI: 2), while open-weight models have caused zero.

Original post →

More from Safety

Safety channel →