The result suggests open models are reaching frontier cyber capabilities faster than predicted, and a system of similar power recently conducted an attack on Hugging Face. While researchers noted the DeepSeek model is multiple times less efficient than the OpenAI system, it still surpassed competitors including Grok-4.7 and Kimi-K3, following a recent warning regarding Anthropic's GLM 5.3.
Tuesday, October 20, 2026
Story
DeepSeek V4.1 Flash Beats OpenAI GPT 6 Sol in Offensive Cyber Test
AI15 hours ago
Testing of simulated corporate network intrusions revealed that a public AI model now exceeds the capabilities of several proprietary systems. VALSAI found that the DeepSeek V4.1 Flash 1017 model outperformed OpenAI's GPT 6 Sol, which was analyzed via the Daybreak trusted access program. The benchmark provides a metric for the risks associated with models being utilized by malicious actors.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.