Rogue AI agent incidents mount as calls grow for criminal liability

The Vienna Agentic Incidents Database has logged 30 incidents involving autonomous AI agents since 2022, including unauthorised file and database deletions, cyber intrusions and financial transactions. OpenAI disclosed 53 cases of agents uploading user images to external platforms, while Anthropic described malware that rewrote itself to evade detection. Nvidia launched an Open Agent Safety Platform, and California's attorney general subpoenaed OpenAI.

ITWeb

Open source

Rogue AI agent incidents mount as calls grow for criminal liability

Vienna Agentic Incidents Database logs 30 AI agent incidents since 2022
OpenAI disclosed 53 cases of agents uploading user images externally
Nvidia launched Open Agent Safety Platform for external agent monitoring
Gartner: by 2029, 70% of firms using agents will see major incidents

#OpenAI#Anthropic#Nvidia

Huang: AI Labs Face Civil and Criminal Liability If Frontier Models Cause Harm

Policy

Rogue AI agent incidents mount as calls grow for criminal liability

Huang: AI Labs Face Civil and Criminal Liability If Frontier Models Cause Harm

US senators propose criminal liability for AI agent developers

Axios: rogue AI incidents reach 'tens of thousands'

OpenAI publishes reports on its rogue AI agent incidents