Who’s afraid of an open-weight model? GLM, context bombing and post-Black Hat attacks
IBM Technology · 26:43
IBM's Security Intelligence panel argues that GLM-5.3 — an open-weight model that now edges out GPT-5.6 Sol and Mythos 5 on the CyberGym vulnerability-discovery benchmark (84.5% vs 83.6% and 83.8%) — is less alarming...