TLDR Tech / AIGLM-5.3 and the spread of advanced cyber capabilities2026-09-30Anthropic's testing found that attackers can bypass GLM-5.3's safeguards between 64% and 100% of the time with simple techniques.Read the full story