WHITE PAPER Responsible AI How ServiceNow is committed to developing and delivering responsible AI solutions Updated: May 2025 WHITE PAPER 2Resonsible_AI_and_GenAI_v2.0 | April 2025 Release: <NAME> Introduction Recent developments in Artificial Intelligence (AI) represent a transformative leap in the evolution of enterprise technology. Following the public launch of GPT-3 and its chatbot incarnation, ChatGPT, there has been incredible interest and debate around AI — and that interest is mirrored by rapid growth in the AI marketplace. According to the 2025 UN Trade and Development (UNCTAD) report the global AI market will soar from $189 billion in 2023 to $4.8 trillion by 2033. AI has the potential to deliver huge benefits, both commercially and more broadly for humanity. Areas as diverse as medical imaging and software programming are being transformed by AI, improving human health in the first case and accelerating digital innovation in the second. However, AI is also a source of concerns ranging from the creation of disinformation (both malicious and as a result of hallucinations), to introducing inherent biases that can worsen existing inequalities and adversely impact disadvantaged groups. Developing responsible AI is essential to ensure that the significant benefits promised by AI are realized, while mitigating the associated risks by offering AI capabilities that are unbiased, truthful, secure, and — critically — do no harm. Resonsible_AI_and_GenAI_v2.0 | April 2025 WHITE PAPER Table of contents Introduction 2 How ServiceNow is committed to responsible AI 4 The ServiceNow approach: combining the power of the platform and AI 5 How ServiceNow delivers responsible AI 6 The ServiceNow AI governance program 6 How ServiceNow builds and maintains Large Language Models (LLMs) 7 Enforcing a consistent lifecycle model process 7 Fine-tuning ServiceNow LLMs using focused datasets 8 Customers control whether data is made available for fine-tuning and testing 8 Data security, handling, and processing for customer instances in production 9 Handling ServiceNow LLM requests in production 9 Handling Azure OpenAI LLM requests in production 9 ServiceNow AI governance tool for customers 9 Conclusion 10 Resources 10 WHITE PAPER 4Resonsible_AI_and_GenAI_v2.0 | April 2025 Release: <NAME> ServiceNow is at the forefront of the latest AI developments and believes that providers and consumers of AI should implement solutions responsibly and ethically. How ServiceNow is committed to responsible AI ServiceNow is proud to be a founding member of the AI Alliance in partnership with IBM, Meta, and other leading organizations to advance the principles of open, safe, and responsible AI globally. AI is not new to ServiceNow, intelligent chatbots, along with other AI capabilities such as machine learning and predictive intelligence, were introduced on the Now Platform as far back as 2018, along with other AI capabilities. Therefore, incorporating the latest AI capabilities into the platform is a natural progression. ServiceNow is at the forefront of the latest AI developments and believes that providers and consumers of AI should implement solutions responsibly and ethically. Therefore, inspired by the NIST Artificial Intelligence Risk Management Framework, ServiceNow has embraced four guiding principles for developing responsible AI: 1. Human-centered — ServiceNow generative AI solutions are built from a foundation of human-centric principles, including persona-based designs and an “out of the box” experience that puts humans in control of AI-based decisions. Where AI is being used in products and services, ServiceNow provides clear documentation and guidance on how to deploy AI in a responsible manner, enabling customers to make informed decisions around where, when, and how they use ServiceNow AI solutions. 2. Inclusive — ServiceNow believes that AI has the power to reduce complexity and make the world a better place for everyone. AI team members strive to build models with datasets that are representative of our global customer base. ServiceNow AI solutions are continuously tested to promote fairness for all, and to minimize bias. 3. Transparent — ServiceNow communicates with customers transparently on the topic of AI, using clear and understandable terms. AI documentation includes practitioner topics, like limits and intended usage. ServiceNow also shares detailed information on the governance foundations of our AI, like the type of data used for training/fine-tuning and the approach to privacy and security. Publicly available model cards explain each specific LLM model’s context, intended use, training/fine-tuning data, limitations, and other important information. 4. Accountable — Trust is the cornerstone of ServiceNow AI initiatives, and as such has adopted an oversight structure to provide accountability and governance. ServiceNow also works closely with external experts and the AI community to gather feedback and has established internal governance bodies to oversee ongoing product and development activities. WHITE PAPER 5Resonsible_AI_and_GenAI_v2.0 | April 2025 Release: <NAME> The ServiceNow approach: combining the power of the platform and AI ServiceNow believes that the best way to unlock the value of AI for customers is to deeply embed it in the platform, rather than simply providing a gateway to an external AI engine. ServiceNow has built AI use cases targeted to the types of data typically stored in the Now Platform, specific to the domains of workflows and knowledge sharing Through this approach, AI has access to contextual data, allowing more precise and relevant information to be provided. It connects this intelligence to action by integrating AI into workflows. ServiceNow also understands the role of AI in developer productivity and has built specialized source code generation tools for customers’ development teams. Embedding the ServiceNow proprietary AI solution directly into the Now Platform also enables ServiceNow to have much tighter control over what AI can do, how it performs, and that human-centric AI guidelines are able to be followed more effectively — a critical component for responsible AI development. Customers can evaluate how Large Language Models (LLMs) are performing, identify and mitigate the risks of bias or problematic outputs, and tailor LLMs to the use cases ServiceNow customers need. This embedded approach also leverages the intrinsic access controls of the Now Platform, which are known to ServiceNow admins and developers and designed into the out of the box workflow experience. To find out more about the physical, administrative, and technical controls ServiceNow has in place to protect customer data please see Securing the ServiceNow AI Platform. WHITE PAPER 6Resonsible_AI_and_GenAI_v2.0 | April 2025 Release: <NAME> How ServiceNow delivers responsible AI The ServiceNow AI governance program The Board, in coordination with the Audit Committee, is responsible for overseeing the ServiceNow AI Governance program, which is focused on the responsible development and use of ServiceNow AI products and services, as well as third-party technologies ServiceNow uses internally. In addition, ServiceNow has an Enterprise AI Governance Steering Committee, comprised of executive leadership, that approves of the ServiceNow AI governance plan and approach. Under this executive mandate, and guided by the dedicated ServiceNow Enterprise AI Governance Committee, multiple disciplines come together to collaborate on AI development: • Platform experts and workflow owners — ensure that AI capabilities integrate seamlessly and safely into each ServiceNow product area. • AI researchers — contribute cutting-edge insights and help set guidelines on fairness and algorithmic risk. • Product and UX leaders — design and implement generative AI features that solve real customer problems in an intuitive, human-centered way. • Legal and Compliance teams — data governance and privacy experts evaluate risks and regulatory requirements for each use case. This multidisciplinary governance team meets regularly, conducts risk & design reviews, and establishes policies/standards for responsible AI. Making responsible AI a company priority ensures that everyone – from engineers to product managers – understands that they are individually and collectively accountable for upholding the ServiceNow responsible AI principles. Accountability mechanisms are built into the governance structure. For example, high-risk AI proposals are escalated to a senior oversight committee for approval. Decisions are clearly documented, and development teams are required to follow the responsible AI guidelines with checkpoints to verify compliance. These mechanisms create a strong culture of accountability and empower every stakeholder to be comfortable speaking out about potential risks or improvements. This collective, transparent governance approach ensures that our AI efforts remain consistent, ethical, and aligned with customer expectations. For more information, please see the ServiceNow Enterprise Artificial Intelligence Governance Policy (requires access to the ServiceNow CORE Compliance Portal). Find out how to access CORE here. WHITE PAPER 7Resonsible_AI_and_GenAI_v2.0 | April 2025 Release: <NAME> How ServiceNow builds and maintains Large Language Models (LLMs) Enforcing a consistent lifecycle model process ServiceNow builds transparent, responsible, auditable, and safe LLMs through a well-governed lifecycle process. Develop Validate Deploy Monitor Retire Research AI SDLC Process 1. Research — an appropriate industry-standard foundational model is selected, such as Mistral-Nemo-12B for tasks like automated code generation. 2. Develop — Fine-tune the foundational model using carefully selected ServiceNow data to create an LLM with the specific intelligence needed for the targeted use cases. In addition, instruction fine-tuning (IFT) is performed to embed carefully curated prompts that are appropriate for the type of interactions the LLM will support.