免费 AI 安全测试
1 分•作者: aiagentlover•9 个月前
我和我的联合创始人开发了一个 AI 红队测试平台，希望在融资前邀请 5-10 家公司进行测试。我们正在通过实际案例研究来验证我们的方法，作为回报，您将获得一份全面的安全审计报告。
我们专注于那些真正能攻破生产环境中 AI 系统的攻击：
* 提示词注入攻击（直接/间接）和越狱
* 工具滥用和 RAG 数据泄露
* 身份伪造和角色扮演攻击
* 通过文档上传进行的 CSV/HTML 注入
* 语音系统操控和基于音频的攻击
您将获得一份完整的报告，其中包含具体的复现步骤、详细的缓解措施，并且在您实施修复后，我们会进行重新测试。如果需要，我们还可以将发现的问题映射到合规框架（如 OWASP Top 10 for LLMs、NIST AI RMF、欧盟 AI 法案等）。我们只需要您提供一个访问端点，并允许我们使用您的匿名化结果作为案例研究。整个过程大约需要 2-3 周。如果您正在生产环境中运行 AI/LLM 系统并希望进行安全审查，请私信我。
查看原文
My co-founder and I built an AI red teaming platform and want 5-10 companies to test it on before trying to go fundraise. We're validating our approach with real-world case studies, and you'd get a comprehensive security audit in return.<p>We focus on the stuff that actually breaks AI systems in production:<p>Prompt injection attacks (direct/indirect) and jailbreaks<p>Tool abuse and RAG data exfiltration<p>Identity manipulation and role-playing exploits<p>CSV/HTML injection through document uploads<p>Voice system manipulation and audio-based attacks<p>You'd get a full report with concrete reproduction steps, specific mitigations, and we'll do a retest after you implement fixes. We can also map findings to compliance frameworks (OWASP Top 10 for LLMs, NIST AI RMF, EU AI Act, etc.) if that's useful. All we need is access to an endpoint and permission to use your anonymized results as a case study. The whole process takes about 2-3 weeks. If you're running AI/LLM systems in production and want a security review, shoot me a DM.