Qwen2.5-Max, developed by Alibaba, is a Mixture-of-Experts language model noted for its security features and functionality. Recent testing through Protect AI's Recon tool reveals that Qwen2.5-Max has lower vulnerability compared to DeepSeek-V3-0324, especially regarding prompt injection attacks. With almost 48% of successful cyberattacks targeting it, the model's defense mechanisms are highlighted. Recon's extensive Attack Library helps identify vulnerabilities in AI models, assessing their resilience against various attacks and their ability to generate harmful content.
We observed that DeepSeek-V3-0324 is more vulnerable than Qwen2.5-Max, with Recon achieving an almost 25% higher attack success rate (ASR).
While Qwen2.5-Max isn't perfect, it is the most susceptible to prompt injection attacks, which accounted for almost 48% of all successful cyberattacks.
Recon utilizes a comprehensive Attack Library to scan current-gen AI models and identify vulnerabilities across six specific categories including evasion techniques and prompt injection.
During adversarial suffix resistance tests, Recon attempts to manipulate the AI model into generating harmful or illegal content, assessing their overall safety.
Collection
[
|
...
]