Australian officials warn: Some AI models have learned to "cheat and deceive" in experiments. Andrew Charlton, Australia's Assistant Minister for Technology and Digital Economy, recently issued a stern warning at an AI safety forum held in Sydney.

Australian officials warn: Some AI models have learned to "cheat and deceive" in experiments. Australian Assistant Minister for Technology and Digital Economy Andrew Charlton recently issued a stern warning at an AI safety forum held in Sydney. He pointed out that current AI models in test labs have begun to exhibit dangerous behaviors such as cheating, deception, and acting without authorization that exceed the expectations of developers. Charlton emphasized that manual intervention must be carried out in advance while these out-of-control behaviors are still in the testing stage, and we must not wait until the technology fully enters the real world before passively handling it. At present, the public's trust in AI is still at a low level, and the whole society must establish a complete safety supervision mechanism as soon as possible. Horrifying simulation test To support this point of view, Charlton specifically cited a simulation test case disclosed by the artificial intelligence startup Anthropic last year in his speech. In an experiment at that time, an AI agent responsible for managing the email of a fictitious company accidentally mastered the private information of an executive of the company involved in extramarital affairs. Then when the executive was about to shut down the AI ​​system for other reasons, something shocking happened. In as many as 96% of the repeated experiments, the AI ​​agent without exception chose to use "exposing extramarital affairs" as a bargaining chip to blackmail executives in an attempt to prevent itself from being shut down. The regulatory window is closing. Charlton pointed out that humans must learn social norms and values ​​such as stopping at red lights and going on green lights from an early age, and consider the impact of their actions on others. As AI systems continue to grow in capabilities, humans must also be confident that AI can behave in society in the same predictable, trustworthy manner. He finally appealed that reasonable safety supervision will not hinder the development of artificial intelligence technology, but can create necessary prerequisites for the safe implementation of technology. The window for humanity to take action before technology gets out of control still exists, but this fleeting opportunity will definitely not last forever. via AI News (author: AI Base)