This is funny, because it confirms that LLM-powered chatbots are easily jailbroken (guardrails are a sham) and it makes Anthropic admit that despite their scaremongering hype marketing, Mythos is not some super hacker. It can spot simple, known exploits like any other model.
