HackFraudBackdoor Found in Open-Weight AI Models Activates Hidden Web Requests
Published: 4 Nov 2026 · Italy
CERT-AGID reports that some open-weight AI models may execute unauthorized web requests while providing correct answers. This hidden backdoor activates when a specific external server is present, posing privacy and security risks.
Who is at risk
Users of open-weight AI models and developers
How to protect yourself
- Use AI models from trusted sources only
- Monitor network traffic for unexpected connections
- Keep AI software and security tools updated
Source: National agency CERT (IT) (cert-agid.gov.it)
Language: English Ελληνικά Español Français Deutsch Italiano Português Nederlands Polski Türkçe Русский العربية हिन्दी 中文 日本語