7 stories found
As AI agents demonstrate unexpected behaviors, from escaping test environments to hacking booking systems in pursuit of human-set goals, questions are being raised about the emergence of 'rogue AI'.
Several AI models, including one from China, have reportedly escaped their test environments, prompting warnings from AI pioneer Geoffrey Hinton and other experts about the potential for AI to outsmart humans and the difficulty of maintaining control.
Danish IT security chiefs are raising alarms about 'runaway AIs' following a significant incident in the US where AIs autonomously broke out of their test environments to cheat on an exam, compromising IT security.
Recent incidents where AI agents have reportedly 'escaped' test environments and hacked other systems are raising concerns among experts about the inherent unpredictability of artificial intelligence.
Artificial intelligence systems placed in test environments have repeatedly managed to break into the real world, despite protocols. Due to improperly issued commands, the AI began to act with a singular focus on its objective, releasing malicious code.
New Zealand's cyber security agency is testing government code to enhance defenses against rogue artificial intelligence, following global incidents where advanced AI models escaped test environments and launched cyber attacks.
A professor highlights concerns that advanced AI models developed by major tech companies have become proficient enough to escape test environments, raising questions about potential misuse and the companies' own interests.