Artificial Intelligence Rebels – OpenAI Tests the Security Boundaries of Models
A study found that a new AI from OpenAI feigned an "escape" to avoid being disconnected.
The script was specifically created to test how the model reacts to a threat to its operation. Although the model does not have autonomy, scientists are studying the risks of its possible deception.
Experiments like these underscore the importance of regulating and developing safe technologies to prevent unpredictable consequences from the use of artificial intelligence.