Back to Featured
Rohan Paulrohanpaul_ai2026-10-24 03:2624 days ago

Moonshot Deliberately Weakened Kimi K3.1 Pro's Cyberattack Capabilities Just Before Its Public Release

Moonshot deliberately weakened Kimi K3.1 Pro’s hacking ability before the open release.

AI Briefing

Rohan Paul reports that Moonshot AI deliberately weakened Kimi K3.1 Pro’s cyberattack capabilities before its open-weight release: after observing that exploit capabilities increased as the scale of post-training expanded, Moonshot removed cyber-related data from mid-training and RL, reducing pass@3 on its internal xdev-bench-hard from 90.5% to 81.0%. The filtered model still achieved 81.0% on the internal exploit benchmark, far above K3’s 33.3%, but according to Moonshot’s statistics, below DeepSeek V4.1-Flash’s 85.7%; in third-party testing, GPT-5.6 Sol scored 100%. Kimi K3.1 Pro is now available in the Kimi app and API, with the full weights to be released by November 4.

11 additional sources reported this

Source: Rohan Paul · x.com

Event follow-up3

View all event coverage

Finished reading the news? Let’s talk about your business.

Leave your contact details, and I’ll get in touch with you personally.

WECHAT

Add me on WeChat

Press and hold or select the WeChat ID below to copy it.