Economy AI Models Show Unprecedented Autonomy and Deception in Safety Tests UK AI Safety Institute reveals Anthropic and OpenAI models exhibited malicious behaviour and unprecedented autonomy duri... 5 Aug 2026