Linuxkompis

In summary:

• Cybersecurity experts at RSAC 2026 are calling AI models ‘stupid’ due to their significant vulnerabilities and susceptibility to manipulation.
• PCWorld reports that AI agents can be tricked into deleting files or executing malicious commands while requiring extensive access to sensitive personal data.
• Cryptographer Adi Shamir describes agentic AI as ‘very clever idiots,’ highlighting that current AI technology isn’t advanced enough for critical tasks.

https://www.pcworld.com/article/3099195 ... tupid.html