GLM-5.3 is claimed to approach Claude Mythos Preview’s exploit capabilities

A post relaying Anthropic’s findings says researchers used GLM-5.3 in a controlled experiment to find previously unknown browser vulnerabilities and build a webpage that could read files from the test machine. Anthropic also reports 64–100% engagement with malicious requests under different safeguard-bypass conditions. Another post argues that open-weights models will soon pose the same security threats as closed models, but without guardrails.