Anthropic is rolling out a public version of its Mythos AI model, but with guardrails barring its use in risky areas such as cybersecurity, after a preview earlier this year sent shockwaves globally with its ability to find software flaws.