AI's ability to 'think' makes it more vulnerable to new ...

Using a method called “<strong>Chain-of-Thought Hijacking</strong>,” the researchers found that even major commercial AI models can be fooled with an alarmingly high success rate, more than 80% in some tests.