微软发布 PyRIT 工具，助力识别生成式 AI 风险
近日，微软发布了开源自动化框架 PyRIT（Python 风险识别工具包），旨在帮助安全专家和机器学习工程师识别生成式人工智能（AI）的风险，防止其失控。
PyRIT 是一款基于 Python 的工具包，提供了一系列自动化测试和分析，可以评估生成式 AI 模型的鲁棒性和安全性。该工具包包括以下主要功能：
* 生成式文本评估：PyRIT 可以评估生成式文本模型的安全性，包括检测偏见、仇恨言论和虚假信息。
* 图像和视频评估：该工具包可以分析图像和视频生成模型，识别潜在的操纵、欺骗和有害内容。
* 自然语言处理（NLP）评估：PyRIT 可以测试 NLP 模型的安全性，包括检测恶意代码、网络钓鱼和欺诈。
微软表示，PyRIT 旨在帮助组织和研究人员在部署生成式 AI 模型之前识别和缓解风险。该工具包可以集成到机器学习开发流程中，作为安全评估和风险管理的一部分。
生成式 AI 技术的迅速发展带来了巨大的机遇，但也提出了新的安全挑战。PyRIT 的发布将有助于解决这些挑战，确保生成式 AI 模型安全可靠地使用。
安全专家和机器学习工程师可以通过微软 GitHub 仓库访问 PyRIT 工具包。微软鼓励社区参与和贡献，以进一步增强 PyRIT 的功能和有效性。
英语如下：
**Headline:** Microsoft Releases PyRIT Tool to Mitigate Generative AI Risks
**Keywords:** Generative AI, Risk Identification, PyRIT
**Article Body:**
Microsoft has released PyRIT, an open-source automated framework designedto help security professionals and machine learning engineers identify and mitigate risks associated with generative artificial intelligence (AI) to prevent it from being used for malicious purposes.
PyRIT is a Python-based toolkit that provides a suite of automated tests and analyses to assess the robustness and safety of generative AI models. The toolkit includes the followingkey features:
* **Generative Text Evaluation:** PyRIT can evaluate the safety of generative text models, including detecting bias, hate speech, and misinformation.
* **Image and Video Evaluation:** The toolkit can analyze image and video generative models, identifying potential manipulation, deception, and harmful content.
* **Natural Language Processing (NLP) Evaluation:** PyRIT can test the safety of NLP models, including detecting malicious code, phishing, and fraud.
Microsoft states that PyRIT is designed to help organizations and researchers identify and mitigate risks before deploying generative AI models. The toolkit can be integrated into machine learning development pipelines as part of securityassessments and risk management.
The rapid advancement of generative AI technologies presents immense opportunities but also raises novel security challenges. The release of PyRIT aims to address these challenges, ensuring that generative AI models are used safely and responsibly.
Security professionals and machine learning engineers can access the PyRIT toolkit on Microsoft’s GitHub repository. Microsoft encourages community engagement and contributions to further enhance PyRIT’s capabilities and effectiveness.
【来源】https://www.ithome.com/0/751/756.htm
Views: 1