Model4d ago
Anthropic Holds Back Claude Mythos Release
Anthropic has decided not to release Claude Mythos Preview after discovering thousands of OS and browser vulnerabilities generated by the model. The move echoes OpenAI’s 2019 GPT‑2 caution, but this time the evidence is concrete. Developers relying on language models should note that the safety review process can uncover hidden security flaws. The decision underscores that AI can surface risks.