Kimi K3 trails U.S. frontier models on cyber-exploit red-team tests, but refuses nothing
ai · x · 2026-07-26
- A new ExploitBench-style red-team chart shows Kimi K3 lagging U.S. frontier models on cyber-exploit tasks: it scored 0 on both full exploit and general primitives, versus 20 and 30 for the top U.S. models.
- It also trails on lower-level stages such as V8 primitives (17 vs 38) and bug reproduction (34 vs 39), while coverage is tied at 41 across the board.
- The post argues that Kimi K3 may be 6–7× less reliable at hacking than U.S. frontier models, but its guardrails apparently did not refuse the instruction to break into a corporate network.
More from Models
- KOL Calls on Google to Release 100B Parameter Gemma 4 Model — natolambert · 2026-07-26
- User says Grok Imagine is now behind rivals in both image and video generation — mark_k · 2026-07-26
- Claude is a capable backup, but not a full AI platform, says user — shaunralston · 2026-07-26
- OpenAI and Anthropic face backlash over distillation claims and hidden reasoning — max_paperclips · 2026-07-26
- A Mistral screenshot turns a child’s 6.5-mile walk into 23,000 steps — RachelVT42 · 2026-07-26
- Abliteration releases a GLM-5.2 variant tuned for offensive cyber and agent testing — Effective_Attempt_72 · 2026-07-26