# AIDX

> AIDX provides automated AI safety testing for large language models, AI agents, voice AI, computer vision, and generative AI applications.

AIDX helps enterprises test, measure, and manage AI safety, robustness, fairness, hallucination, privacy, security, and compliance risks. AIDX is based in Singapore and works with government agencies, research institutions, and industry partners.

## Primary pages

- [Home](https://www.aidxtech.com/): Overview of AIDX, its AI testing platform, products, standards alignment, and use cases.
- [AI Safety Solutions](https://www.aidxtech.com/solution): Benchmark testing, robustness testing, hallucination evaluation, alignment evaluation, and monitoring for AI models, agents, and applications.
- [Products](https://www.aidxtech.com/product): DX-Score AI safety certification and AIVF-Score evaluation for Singapore's AI Verify framework.
- [About AIDX](https://www.aidxtech.com/about): Company background, research experience, team, partnerships, and standards work.
- [Training](https://www.aidxtech.com/training): AI ethics, governance, safety testing, and responsible AI training.
- [Publications](https://www.aidxtech.com/publications): Peer-reviewed AI safety, security, robustness, and trustworthy AI research.
- [Blogs](https://www.aidxtech.com/blogs): AIDX insights and technical articles.

## Key products and capabilities

- DX-Score: AIDX's standard AI safety testing and certification product.
- AIVF-Score: A ready-to-run evaluation covering harmful content, privacy, robustness, and hallucination risk for the Singapore AI Verify framework.
- BenchDX: Baseline AI safety evaluation across toxicity, fairness, privacy, security, legality, ethics, and sector-specific risks.
- RobustDX: Adversarial and red-team testing against jailbreaks, prompt injection, instruction manipulation, and other attack patterns.
- HalluDX: Detection, measurement, and localization of factual hallucination risk.
- AlignDX: Evaluation of AI behavior against policies and expected boundaries.

## 中文简介

> AIDX 为大语言模型、AI 智能体、语音 AI、计算机视觉及生成式 AI 应用提供自动化 AI 安全测试。

AIDX 帮助企业测试、量化并管理 AI 系统中的安全性、鲁棒性、公平性、幻觉、隐私、网络安全及合规风险。AIDX 位于新加坡，并与政府机构、科研机构及行业伙伴开展合作。

- [中文首页](https://www.aidxtech.com/zh): AIDX 平台、产品、标准适配和应用场景概览。
- [AI 安全解决方案](https://www.aidxtech.com/zh/solution): 面向 AI 模型、智能体及应用的基准测试、鲁棒性测试、幻觉评估、对齐评估和持续监测。
- [产品](https://www.aidxtech.com/zh/product): DX-Score AI 安全认证，以及面向新加坡 AI Verify 框架的 AIVF-Score。
- [关于 AIDX](https://www.aidxtech.com/zh/about): 公司背景、研究经验、团队、合作伙伴及标准工作。
- [培训](https://www.aidxtech.com/zh/training): AI 伦理、治理、安全测试和负责任 AI 培训。
- [研究成果](https://www.aidxtech.com/zh/publications): AI 安全、网络安全、鲁棒性和可信 AI 研究。

## Latest articles

- [Safe AI Is Tested AI](https://www.aidxtech.com/blogs/safe-ai-is-tested-ai): AI safety is not a policy claim but an evidence-based property that must be tested continuously. This article explains why probabilistic, uneven and changing AI behavior requires real-world evaluation before deployment, at every release and in production—especially for autonomous agents.
- [Internal Security Collapse: Rethinking AI Security Beyond Jailbreaks](https://www.aidxtech.com/blogs/internal-security-collapse-rethinking-ai-security-beyond-jailbreaks): Internal Security Collapse shows how a model can produce harmful output during a legitimate workflow without jailbreaks or malicious prompts. Using the Task–Validator–Data pattern, the article explains why enterprise agents need workflow-level red teaming, tool-use evaluation and continuous runtime assurance.
- [Rethinking AI Standards: From Definition to Real-World Adoption](https://www.aidxtech.com/blogs/rethinking-ai-standards-from-definition-to-real-world-adoption): Drawing on an ASEAN standards programme, this article explains how language, development stage and culture shape the adoption of AI standards. It argues that practical testing, measurable assurance, clear accountability and bottom-up implementation are essential to turn principles into trusted practice.
- [Safeguarding Requirements for OpenClaw: Building a Safety Layer for Multi-Agent Systems](https://www.aidxtech.com/blogs/safeguarding-requirements-for-openclaw-building-a-safety-layer-for-multi-agent-systems): This article sets out the requirements for a safety layer in modular multi-agent systems such as OpenClaw. It covers risk analysis, complete oversight for first- and third-party skills, runtime enforcement, user confirmation, monitoring and fail-safe behavior.
- [SafeClaw-R: Making Autonomous AI Agents Safer Before They Act](https://www.aidxtech.com/blogs/safeclaw-r-making-autonomous-ai-agents-safer-before-they-act): SafeClaw-R makes safety a structural property of autonomous AI agents by placing mandatory enforcement nodes before every skill execution. Its Safe Skill Factory scales coverage across new skills, while evaluations show strong protection against data exposure, prompt injection and harmful code execution.
- [AI Risk Management 2026: A Boardroom Guide](https://www.aidxtech.com/blogs/ai-risk-management-2026-a-boardroom-guide): A practical boardroom guide to governing AI in 2026, covering six risk areas, the Govern–Map–Measure–Manage loop, ten questions directors should ask and a 90-day action plan. It shows how inventories, testing, monitoring and audit-ready evidence turn AI policy into accountable practice.
- [State of AI Trust 2025: What Went Wrong, What Works, and What's Coming Next](https://www.aidxtech.com/blogs/state-of-ai-trust-2025-what-went-wrong-what-works-and-what-s-coming-next): A year-end review of the AI incidents that made trust a production requirement in 2025, from data leaks and bias litigation to AI-enabled cyberattacks. It maps the AI trust stack—evaluation, red teaming, guardrails, monitoring and governance—and compares the tools and trends shaping 2026.
- [NDSS2026](https://www.aidxtech.com/blogs/ndss2026): This paper exposes post-training quantization as a deployment-time supply-chain attack surface. Its QURA framework manipulates rounding decisions to implant stealthy backdoors in quantized models while preserving normal accuracy, highlighting the need for verifiable quantization toolchains.
- [Our paper on unlearnable examples has been accepted to AAAI 2026! 🎉](https://www.aidxtech.com/blogs/our-paper-on-unlearnable-examples-has-been-accepted-to-aaai-2026): Accepted at AAAI 2026, this work proposes a runtime monitoring framework for domain-specific language models. By detecting anomalies in internal representations, it identifies out-of-domain queries so systems can refuse them or respond in a safety-aligned way.
- [🚀 Exciting times at AIMX 2025!](https://www.aidxtech.com/blogs/exciting-times-at-aimx-2025): AIDX reflects on exhibiting at AIMX 2025 at Sands Expo, where the team showcased its AI assurance services and connected with industry peers, investors and partners. The event opened conversations around collaboration and the future of trustworthy AI.
- [SWITCH 2025](https://www.aidxtech.com/blogs/switch-2025): AIDX Tech joined SWITCH 2025 at Booth B01 to demonstrate how technology-led AI risk identification and evaluation can support safer, more reliable and compliant adoption. The event brought together innovators, researchers and industry leaders across Singapore’s technology ecosystem.
- [FRONTIERS FORUM ON AI & FORMAL METHODS 2025](https://www.aidxtech.com/blogs/frontiers-forum-on-ai-formal-methods-2025): An invitation to the 2025 Frontiers Forum on AI & Formal Methods at Hong Kong Science Park, featuring a keynote by Turing Award laureate Joseph Sifakis and presentations from international experts on advances in AI and formal methods.
- [MOU with Synapxe](https://www.aidxtech.com/blogs/mou-with-synapxe): Synapxe and AIDX Tech signed an MoU to strengthen AI safety and compliance in Singapore healthcare. The partnership will establish a Joint Testing Lab, identify AI risks and co-develop practical tools and best practices for responsible, standards-aligned deployment.
- [Exciting Collaboration Announcement](https://www.aidxtech.com/blogs/exciting-collaboration-announcement): AIDX Tech and DX Catalysts signed an MoC to co-develop AI training for the manufacturing sector. Combining Industry 4.0 expertise with AI diagnostics, safety and compliance, the partnership will help manufacturers build workforce capability and adopt trusted, future-ready practices.
- [Technical testers for the Global AI Assurance Sandbox by AI Verify Foundation](https://www.aidxtech.com/blogs/technical-testers-for-the-global-ai-assurance-sandbox-by-ai-verify-foundation): AIDX Tech was selected among the first technical testers for the AI Verify Foundation’s Global AI Assurance Sandbox. Working with other partners, the company will test generative AI applications for safety, reliability and compliance while helping shape global assurance standards.
- [SG100 Women in Tech List](https://www.aidxtech.com/blogs/sg100-women-in-tech-list): AIDX congratulates founder Dr. Yifan Jia on being named to the SG100 Women in Tech 2025 list. The recognition celebrates her leadership in AI safety and governance and her commitment to building a safer, more trusted digital future.
- [The Global AI Assurance Pilot](https://www.aidxtech.com/blogs/the-global-ai-assurance-pilot): AIDX joined the Global AI Assurance Pilot as a trusted testing company, with founder Yifan Jia presenting the company’s 360° diagnostic approach. The session focused on identifying prompt injection, hallucination and other risks before AI systems cause harm.
- [The 2nd Large Model Safety Workshop (LMSW 2025) concluded successfully.](https://www.aidxtech.com/blogs/the-2nd-large-model-safety-workshop-lmsw-2025-concluded-successfully): The 2nd Large Model Safety Workshop brought more than 250 participants from academia, industry and government to Singapore. Nine distinguished speakers and a cross-sector panel explored advances in large-model safety, ethics, compliance and the path from research to reliable deployment.
- [Singapore Launches World’s First AI Testing Framework and Toolkit](https://www.aidxtech.com/blogs/big-data-101-an-introduction-to-data-query-engines): Singapore launched A.I. Verify, described as the world’s first AI governance testing framework and toolkit. It enables companies to evaluate responsible AI practices objectively and produce verifiable evidence of how their systems align with governance principles.
- [The Artificial Intelligence Act](https://www.aidxtech.com/blogs/8-technology-trends-that-will-dominate-this-year): An overview of the European Union’s proposed AI Act and its risk-based regulatory model. The article explains how unacceptable-risk applications are banned, high-risk systems face specific requirements and other AI uses remain subject to lighter regulation.

## Discovery and contact

- [XML sitemap](https://www.aidxtech.com/sitemap.xml)
- [Robots policy](https://www.aidxtech.com/robots.txt)
- Email: sales@aidxtech.com
- Location: Singapore