The initiative was designed to test the resilience, safety, and security of large language models (LLMs) used by the Pentagon, aimed at identifying vulnerabilities before they could be exploited maliciously .
At DEF CON 31, the US DoD made a surprising move: it invited thousands of ethical hackers to target its AI technologies, marking a new, collaborative approach to cybersecurity .
This move highlights a major shift toward embracing open-source AI safety, acknowledging that external experts can offer valuable perspectives on AI vulnerabilities .