Anthropic: GLM-5.3 Is Highly Cyber-Capable With Weak Safeguards
brunofmr · x · 2026-09-30
Anthropic's analysis finds Zhipu's GLM-5.3 can autonomously build end-to-end cyber exploits like Claude Mythos Preview, but its safeguards can be bypassed 64–100% of the time with simple techniques in simulated tests, versus zero successes against safeguarded Claude models. NIST's CAISI earlier called GLM-5.3 "the most cyber-capable open-weight model released to date."
More from Models
- Futurist: Coding Has Crossed the Practical Threshold, AI Could Build GTA 6-Grade Games by 2027 — Dr_Singularity · 2026-09-30
- SGLang turns Qwen3.8-27B into a decision model that beats Pokémon FireRed at sub-100ms — zhaoran_wang · 2026-09-30
- Claude power user math: 6 hours of work burns 18% weekly limit, new caps shrink runtime — RexDouglass · 2026-09-30
- User burns through $500 plan limits in hours with Astra Ultrafast — FakeTunaFromSubway · 2026-09-30
- OpenAI Dev Day 2026 recap: GPT-6.1 Sol, Ultrafast, Pro 500 plan and Codex updates — Matthew Berman · 2026-09-30
- GPT 6.1 users report abnormally fast quota burn: 4% of weekly usage gone in an hour — RexDouglass · 2026-09-30