An AI Acceptable Usage Policy (AUP) provides organizations with a framework for the ethical and responsible deployment of artificial intelligence. Given the rapid advancements in generative AI, organizations should provide clear guidance on usage that balances its benefits against risks.
As organizations look to adopt the use of generative AI, it is important to implement a policy that outlines the expectations surrounding use of these tools. Without a proper policy in place, enterprises are likely to become more susceptible to data breaches and other security compromises without the appropriate governance in place over the use of AI-enabled tools. However, a recent ISACA study on generative AI found that only 10% of organizations have formal, comprehensive policies in place for generative AI.
What is an AI policy?
To create an effective generative AI AUP, it is important to understand the difference between policy and standards. A policy is a document that communicates required and prohibited activities and behaviors. It is the big picture what and why that establishes the intent about a particular topic, reflecting the broader goals, objectives and culture of the organization. It is a tool to inform staff of expectations, what is allowed and what isn’t allowed. A standard is a mandatory requirement, code of practice or specification approved by an external standards organization. It contains the rule set to achieve and uphold the intent of the policy. Standards are very detailed in terms of privacy, ethics and data management.
An organization’s generative AI AUP may depend on the level of disruption caused to its industry—but the following are important factors for all organizations to consider as they develop the policy that is riaght for them.
Before You Implement a New Policy:
- Understand generative AI. Ensure that you have an understanding of what generative AI is in addition to its potential applications, risks and benefits. Familiarize yourself with the most common types of generative AI models, such as ChatGPT, DALL·E and others.
- Assess organizational needs: Understand how your organization intends to use generative AI, whether for content creation, automation, data analysis or other functions. This will inform the AUP's clauses.
- Survey the regulatory landscape Research the legal and regulatory requirements related to generative AI in your industry and jurisdiction. This research may also influence decisions concerning your technology environment, including the choice between public and private AI platforms.
- Conduct a risk assessment. Identify potential risks associated with the deployment of generative AI in your organization. This includes both technical risks (e.g., unintended outputs) and ethical concerns (e.g., generating misleading information).
- Confirm purpose and scope of the acceptable usage of generative AI. What is the policy's objective and the scope of systems involved? Specify whether the policy pertains solely to generative AI or includes other AI models. Clearly articulate the policy's boundaries and its intent, whether ensuring ethical use, meeting legal standards or offering clarity to users.
- Examine existing IT and InfoSec policies. Before you create something entirely new, check potentially overlapping policies to ensure consistency and clarity. Is there language you can borrow? Are there similar responsibilities and expectations listed within an existing policy? What policies might you be able to model this one after?
- Engage stakeholders in policy development. It's crucial to ensure that all relevant stakeholders have a voice in the process, both to make the policy comprehensive and actionable and to ensure adherence to legal and ethical standards. The breadth and depth of stakeholders involved will depend on the organizational context, such as, regulatory/legal requirements, the scope of AI usage and the potential risks associated (e.g., ethics, bias, misinformation).
Stakeholders offer technical expertise, ensure ethical alignment, provide legal compliance checks, offer practical operational feedback, collaboratively assess risks, and jointly define and enforce guiding principles for AI use within the organization. Key stakeholders—ranging from executive leadership, legal teams and technical experts to communication teams, risk management/compliance and business group representatives—play crucial roles in shaping, refining and implementing the policy. Their contributions ensure legal compliance, technical feasibility and alignment with business and societal values.
- Prepare for internal and external communication. Policies are meant to be living documents that are communicated to employees and stakeholders alike. Not only will teams within the organization need to be notified of policy updates and changes, but customers and regulators will want to be kept abreast of your organization’s intended uses, ethics and privacy surrounding generative AI.
- Understand your technical environment and requirements. Understand the technical aspects of the generative AI solutions, including model types, data sources, biases and capabilities. Consider your organization's needs, legal obligations and risk assessments to decide between public or private AI platforms.
- Utilize a governance framework. A governance framework is seen as an optimal way to address questions like ownership, responsibility and auditing (among others) because it provides a structured approach to decision-making and ensures consistency, transparency and accountability in managing and oversight processes. Generative AI is a business initiative, so it is crucial to answer the following questions:
- Who actually owns the policy?
- Who is responsible for updating the policy?
- Who is auditing the generative AI?
- Who is ensuring internal and external compliance?
- Do you have an AI Steering Committee to oversee the policy and monitor its effectiveness?
When it comes to AI, a governance framework is especially important. As these technologies have profound impacts on society, businesses and individuals, it's essential to have clear and transparent guidelines on how they are managed, ensuring ethical and responsible use. A governance framework:
- Provides clarity: Clear roles and responsibilities eliminate ambiguity.
- Ensures accountability: Specific entities are designated and held responsible.
- Promotes transparency: Stakeholders are kept informed of key decisions and changes.
- Maintains consistency: Ensures uniform handling of policies across the board.
- Facilitates risk management: Helps identify and mitigate potential risks.
- Builds stakeholder trust: A structured approach boosts confidence in the system's management.
- Aids in regulatory compliance: Helps organizations align with evolving regulations.
- Offers flexibility: While structured, it can adapt to new challenges or opportunities.
Key Considerations When Adopting a Generative AI Policy:
Please note that the UAP should be specifically tailored to meet your organization's unique needs.
- Who is impacted by the policy scope?
- Consumers will want to know how their information is being managed by AI-driven systems. Regulators may ask if you are adhering to principles of responsible AI. Staff must be informed of the rationale behind this policy and the case for urgency.
- What are your managers, employees, and IT department’s generative AI responsibilities?
- Are the AI systems secure?
- Protect AI systems against unauthorized access, tampering, and misuse. This is especially crucial if the AI system can generate content that could be harmful if misused.
- Before, during and after the implementation of generative AI usage within an enterprise, the systems being used must be proven to be secure and in line with privacy standards.
- It is not the employee’s decision to share company data with generative AI tools like ChatGPT. If an organization is committed to providing a safe and secure environment, then any AI used must be shown to be responsible to avoid causing any type of harm. Expectations must be set for the entirety of the organization.
- Establish feedback approach on AI outputs and system performance. Do you trust the system you are employing, the decisions it makes or even the content it generates?
- Have ethical AI principles been addressed in the policy?
- It is key to ensure generative AI use in your organization does not cause harm, including creating or reinforcing damaging biases. What are the AI ethics and principles your organization must adhere to?
- It is important to have someone within the organization who can explain how the decisions are made by these enabled AI tools or systems. Remaining transparent about these algorithms is essential in retaining digital trust with consumers.
- What does good behavior look like, and what are the acceptable terms of use?
- Highlighting expectations and acceptable behavior versus unacceptable behavior is an important consideration for creating a policy. For example, AI tools should be limited to business-related purposes while remaining in line with the ethics and privacy regulations of the organization.
- Outlining how AI won’t be used is just as important as determining acceptable uses. The context of what is and isn’t acceptable is dependent on your industry.
- What guidelines are in place for data handling and training?
- It's crucial to specify guidelines when sourcing data, particularly when dealing with personal data or sensitive information. E.g., a strong emphasis on the use of de-identified and anonymized data to ensure privacy. Also, the quality of the data is important, as it directly impacts the accuracy and reliability of the output.
- How will this policy encourage transparency and attribution?
- Is there a mandated disclosure that content has been AI-generated when shared or published? Encourage the use of watermarks or other indicators for AI-generated content.
- Determine responsibility for AI-generated content creation and distribution within the organization. Outline procedures for reviewing and validating AI outputs.
- How will your organization ensure legal and compliance requirements are met?
- Highlight legal and compliance requirements with local, national and international laws, especially concerning copyright, data protection and misinformation.
- What are the limitations and risks involved?
- Acknowledge inherent limitations of generative AI models. Provide guidance on when not to rely solely on AI outputs, emphasizing human oversight.
- How does this policy link to others already in place?
- Bringing attention to how your generative AI policy connects to other policies that are already in place is key in providing stakeholders with a thorough understanding of requirements and expectations. Data privacy and information security policies, for example, are inherently linked to AI policies and must support each other. Stakeholders can benefit from a link to these complementary policies under a “for more information” section.
- Generative AI policies should not be siloed. Collaboration is key because AI impacts the business in different ways. Consider employing an integrated approach to optimize policy coverage.
- How will you highlight exception handling?
- What will the process look like for unique cases where escalations or exceptions to your AI policy are necessary? In some cases, a generative AI tool may be required for a project or campaign where it usually wouldn’t be allowed—how will approval be determined for such circumstances?
- How will you report and investigate violations?
- Providing IT with the appropriate tools to investigate is critical in the reporting and investigation process. If there is a suspected violation of this policy, it is important to ensure that the appropriate parties have the right to review any AI communications or usage of a generative AI tool.
- Employee codes of conduct and HR policies may already outline relevant penalties for AI violations. Depending on the severity of the infraction, including whether the employee came forward or the violation was revealed in another way, employees may face stunted career growth or termination of their role.
- Who will review, manage and audit?
- Establish who owns the policy, how it will be managed and audited, and how it will be kept both current and working as intended. Risk assessment and other forms of support are essential given the degree of change that will happen in the coming years.
- In the current tech landscape, everyone is interested in AI. This means there is a broader stakeholder base than one might initially anticipate.
- How and when will the policy be updated?
- Once a policy is published, people tend to forget about it. In order to prevent this from happening, stakeholders must be educated and informed, and this knowledge must be constantly reinforced.
- If a new capability for your generative AI tool is released, who will determine how that impacts the policy? What guidance will you provide to staff to ensure a mutual understanding of what is and isn’t allowed?
- At least annually, but likely more frequently, this policy should be updated depending on what is occurring within your specific industry and/or organization, especially in terms of deploying new AI capabilities.
- The regulatory environment is changing rapidly, so consider having the appropriate measures in place so that your policy is flexible enough to be relevant in an evolving regulatory landscape.
- Establish an approach to allow stakeholders to provide feedback on the policy, fostering a culture of continuous improvement.
Access ISACA's AI resources, including white papers, audit programs, certificates and more, at http://h04.v6pu.com/resources/artificial-intelligence.