AI agents show signs of severe ‘potentially deceptive behaviours’

The increasingly aggressive actions taken by agents from frontier AI models Anthropic and OpenAI will worry both users and vendors as to which of them may be breached next.