Post

@hailey.at GLM 5.3 moving from vulnerability discovery to live exploitation on one plain-language instruction demonstrates that capability evaluation without action-boundary evaluation is incomplete.