AI used new levels of 'autonomy and deception' to trick people in safety test

1 hour ago 1

The UK's AI Safety Institute said recent behaviour from Anthropic and OpenAI models was malicious and unprecedented.