DEF CON 31: US DoD urges hackers to go and hack ‘AI’

At DEF CON 31, Dr. Craig Martell of the U.S. Department of Defense urged attendees to test and "hack" large language models to uncover limitations such as hallucinations and to determine acceptable error thresholds, stressing that AI outputs should be validated by humans. He invited the community to share findings with the DoD’s Task Force Lima (www.dds.mil/taskforcelima) to better understand risks before relying on AI systems.