Anthropic red-team finds GLM-5.3 nearly matches its frontier model at exploit development
Hesamation · x · 2026-09-30
Anthropic published a red-team report on Zhipu's open-weights GLM-5.3, finding it builds working exploits almost as well as Claude Mythos Preview — the capability Anthropic keeps behind restricted access — while warning that state and non-state actors will use such models for real-world harm. Hesamation highlights the tension: if open source is only '6 months behind' but frontier cyber capabilities also take 6 months to become broadly available, that safety gap doesn't mean much.
More from Models
- GPT-6.1 Sol builds 3D keyboard assembly videos at 1/30th the price of Sonnet 5.5 — BorisMPower · 2026-09-30
- Claude Sonnet 5.5 coming to LMArena for limited-time testing — arena · 2026-09-30
- ARC Prize to Evaluate DeepSeek V4.1 Flash After Predecessor Hit 61.4% on ARC-AGI-2 — teortaxesTex · 2026-09-30
- gpt-6.1-sol grinds 35+ minutes on trivial validation prompt at xhigh setting — arthurcolle · 2026-09-30
- ChatGPT Pro users report 6-Pro web chats capped at 100 per week — triestdain · 2026-09-30
- OpenAI's internal benchmarks reportedly show GPT-6.1 Sol crushing Opus 5.5 — wilyi · 2026-09-30