Anthropic report flags GLM-5.3 as strongest open model for cyber offense, drawing self-serving criticism
oran_ge · x · 2026-09-30
After OpenAI added GLM 5.3 and other open models to its enterprise tier, Anthropic published a report criticizing open-model safety, drawing widespread skepticism.
Report highlights:
- Calls GLM-5.3 "the strongest open model for cybersecurity capabilities to date," roughly 4 months behind US frontier models.
- Among models tested (Opus 4.6, Kimi K3, DeepSeek Flash), GLM 5.3 was the only one with exploit capabilities.
- Argues open weights are inherently unsafe: abliteration can strip refusal behavior for $4,400, and jailbreak-style prompts or prefill thinking can bypass guardrails; Claude is "safe" because it's closed and offers no prefill.
- Concludes enterprises should urgently adopt safe closed models.
Author's critique: the report is self-serving security theater — removing a model's guardrails to prove it's unsafe is circular; if Anthropic cares about open-model safety it could contribute to it, rather than moralizing while banning accounts.
More from Models
- Fable 5.2 Leads MazeBench Token Efficiency; Billion-Token Runs Underway — patience_cave · 2026-11-16
- Fable 5.2's Puzzle Skills Look Suspiciously Good, Reciting Outdated MazeBench Solution — patience_cave · 2026-11-16
- Fable 5.2 Tops MazeBench With 21% After 90 Hours and $6,000 in Credits — patience_cave · 2026-11-16
- Tencent Hy4 flash and InternLM Sage-S2.1-Flash both released, tool-calling tests requested — MajesticPuzzler · 2026-11-16
- User Shares Coral Reef Creation Labeled 'Qwen 4 Max (xhigh)', an Unannounced Model Name — intellectronica · 2026-11-16
- Stealth Model 'silicon-2' Surfaces in Devin, Suspected to Be Anthropic's Next Model — kimmonismus · 2026-11-16