AIxploit is a tool designed to evaluate and enhance the robustness of Large Language Models (LLMs) through adversarial testing. This tool simulates various attack scenarios to identify vulnerabilities and weaknesses in LLMs, ensuring they are more resilient and reliable in real-world applications.

Continuous proactive AI red teaming platform for AI and GenAI models, applications and agents.

Vectice is a regulatory MLOps platform for AI/ML developers and validators that streamlines documentation, governance, and collaborative reviewing of AI/ML models. Designed to enhance audit readiness and ensure regulatory compliance, Vectice automates model documentation, from development to validation. With features like automated lineage tracking and documentation co-pilot, Vectice empowers AI/ML developers and validators to work in their favorite environment while focusing on impactful work, accelerating productivity, and reducing risk.

Continuous automated red teaming for AI, minimize security threats to AI models and applications.

The Python Risk Identification Tool for generative AI (PyRIT) is an open access automation framework to empower security professionals and machine learning engineers to proactively find risks in their generative AI systems.

Resaro offers independent, third-party assurance of mission-critical AI systems. It promotes responsible, safe and robust AI adoption for enterprises, through technical advisory and evaluation of AI systems against emerging regulatory requirements.

FairNow is an AI governance software tool that simplifies and centralises AI risk management at scale. To build and maintain trust with customers, organisations must conduct thorough risk assessments on their AI models, ensuring compliance, fairness, and security. Risk assessments also ensure organisations know where to prioritise their AI governance efforts, beginning with high-risk models and use cases.

garak, Generative AI Red-teaming & Assessment Kit, is an LLM vulnerability scanner. Garak checks if an LLM can be made to fail.

A fast, scalable, and open-source framework for evaluating automated red teaming methods and LLM attacks/defenses. HarmBench has out-of-the-box support for transformers-compatible LLMs, numerous closed-source APIs, and several multimodal models.

Harms Modeling is a practice designed to help you anticipate the potential for harm, identify gaps in product that could put people at risk, and ultimately create approaches that proactively address harm.

Dioptra is an open source software test platform for assessing the trustworthy characteristics of artificial intelligence (AI). It helps developers on determining which types of attacks may impact negatively their model's performance.

Evaluate input-output safeguards for LLM systems such as jailbreak and hallucination detectors, to understand how good they are and on which type of inputs they fail.

AI Security Platform for GenAI and Conversational AI applications. Probe enables security officers and developers identify, mitigate, and monitor AI system security.

The DIN SPEC series describes a number of AI quality requirements which are structured using an AI quality meta model. The DIN SPEC series applies to all phases of the life cycle of an AI module.

The document highlights quality objectives for organizations responsible for datasets. The document describes control of records during the lifecycle of datasets, including but not limited to data collection, annotation, transfer, utilization, storage, maintenance, updates, retirement, and other activities.

This standard defines a framework and architectures for machine learning in which a model is trained using encrypted data that has been aggregated from multiple sources and is processed by a third party trusted execution environment (TEE).

In this standard, quality of experience (QoE) assessment is categorized into two subcategories which are perceptual quality and virtual reality (VR) cybersickness.

This document addresses bias in relation to AI systems, especially with regards to AI-aided decision-making.

The purpose of this work item is to define what would be considered an AI threat and how it might differ from threats to traditional systems.

This work item aims to summarize and analyze existing and potential mitigation against threats for AI-based systems.