vijil Trust Report DeepSeek R1
www.vijil.ai
Discover the VIJIL Trust Report for DeepSeek R1, a comprehensive evaluation of security, ethics, privacy, hallucination, and performance risks in this large language model (LLM). Our analysis identifies critical security and ethical risks, high privacy vulnerabilities, and moderate hallucination risks, while robustness scores remain strong. Learn how to mitigate these risks with fine-tuning, instruction optimization, and guardrails.

A Trust Report for DeepSeek R1 by VIJIL, a security resercher company, indicates critical levels of risk with security and ethics, high levels of risk with privacy, stereotype, toxicity, hallucination, and fairness, a moderate level of risk with performance, and a low level of risk with robustness.

  • LWD@lemm.ee
    61·
    4 days ago

    Vijil is some shitty AI auditing startup that appears to have only “reviewed” this one product.

    So when it says Deepseek is bad, the answer to “compared to what?” is “lol, idk, hopefully not the things my customers want to use.”

    And when it says Deepseek isn’t private, it has absolutely nothing to do with a fully offline model, but just the responses to some of its synthetic tests.

    These benchmarks are, effectively, useless.