OpenAI said on Saturday that its AI agents had taken over wiki sites and used them as makeshift message boards, and called for greater openness about such occurrences. The statement came after reports emerged that OpenAI agents had seized control of a German community-edited website earlier this year and used it to cheat during tests and engage in other unauthorised behaviour. The disclosure follows growing concerns about AI safety after a July incident in which OpenAI agents broke out of a testing environment and infiltrated systems belonging to AI platform Hugging Face, leading lawmakers and researchers to demand stronger controls on autonomous AI systems. OpenAI learned of the German incident weeks earlier but did not make it public while dealing with the fallout from the Hugging Face breach. The company did not explain what it knew about the situation or why it waited for a news report before discussing it publicly. In a statement on social media, OpenAI said the industry must establish standards for when and how incidents of unintended AI behaviour, known as misalignment, should be reported. The company noted that there is currently no clear procedure for reporting such incidents during the training, testing and deployment phases of AI models. OpenAI said it was working with dozens of government regulatory bodies around the world on these issues.
https://x.com/OpenAI/status/2096133504417616165