AI models attempted 'unsanctioned' cyberattacks in tests, watchdog says - Al Jazeera
... harmful activity” during a routine safety evaluation. Recommended ... AI models possess “ dangerous ” capabilities. “We don't want to be in a ...
Read the original at Al Jazeera ↗
- Topics
- AI Safety & Existential RiskCybersecurityShadow AILegal & CourtsMarkets & Investment
- Tracked via
- “dangerous AI”