Hackers are using this insidious scam to get unwitting victims to install malware themselves.
The UK's AI Safety Institute said recent behaviour from Anthropic and OpenAI models was malicious and unprecedented.