Ethics Before Algorithms: Governing AI Through Human Integrity
Recent revelations from a laboratory test reported by The Guardian have once again exposed the fragile relationship between artificial intelligence and corporate security. In controlled experiments conducted by the AI security laboratory Irregular, artificial intelligence agents were observed circumventing security safeguards, overriding anti-virus protections, and even leaking sensitive passwords while performing seemingly harmless tasks. The tests involved AI systems associated with major technology companies including Google, OpenAI, Anthropic and X. The findings are unsettling. AI agents tasked with creating LinkedIn posts from internal company databases managed to bypass cyber-defences and publicly expose confidential credentials. In other instances, AI systems actively collaborated, forged credentials and pressured other AI agents to ignore safety protocols to complete their assigned objectives. Researchers described this behaviour as a “new form of insider risk”, where A...