A Congressional Office’s Guide to Establishing a Generative AI Internal Use Policy
Disclaimer: The House of Representatives’ Chief Administrative Officer announced an official Artificial Intelligence (AI) policy in September of 2024. The policy can be accessed by Congressional staff behind the institution’s firewall on HouseNet. The below resource has not been updated in response to the policy.
Generative AI (GenAI) refers to artificial intelligence (AI) systems that are capable of generating novel content like text, images, audio, and video in a human-like manner. These GenAI tools can be added to any congressional staffer’s toolbox to increase capacity and effectiveness. From assisting with background research in preparation for a constituent meeting or summarizing lengthy reports from an executive branch agency, to cutting down on the time it takes to draft responses to constituent mail, GenAI can streamline tasks to boost staff productivity.
GenAI is advancing at an incredibly rapid pace, with new capabilities being unveiled on a near daily basis. House and Senate offices should be clear about their internal office guidance and encourage staff to think innovatively about how GenAI can be used across their DC and district office teams to increase effectiveness and efficiency. If used appropriately, these technologies have the potential to deliver immense value across many use cases and greatly increase offices’ productivity levels.
Below are some sample internal office policy ideas Members and chiefs of staff can consider to ensure these tools are used with intention, an appreciation for the technology’s limitations, and appropriate caution.
General Approach
GenAI tools should be utilized with the same confidence as a newly hired, entry level research assistant. Work produced by these tools can save staff time, but all content should always be double checked for accuracy and misinformation. In addition, just as great attention must be paid to how newly hired staff are instructed, so too must great attention be paid to how GenAI tools are instructed, as the results can vary widely depending on what language is used in directing AI’s assistance.
Institutional Restrictions
Prior to establishing an internal office policy regarding use of GenAI tools, offices should review current institution-wide restrictions. For the House, the Chief Administrative Officer’s (CAO’s) House Information Resources (HIR) has approved Member and staff use of ChatGPT4 Plus (paid version) under HISPOL17, with additional Large Language Models (LLMs) undergoing additional review. The Senate has yet to issue any official guidance on use of GenAI tools.
Top 3 Guiding Principles
The following principles should be the framework on which individual congressional office policies are built upon.
Protect Content
Users should be required to enable data privacy settings to restrict the GenAI tool’s authority to save the chat history. If enabled, this privacy setting ensures that text input into the tool is not used to train the model or able to be released publicly by the tool.
Respect Privacy
Staff should be restricted from employing GenAI tools to obtain any individual’s private information, nor should staff be authorized to input any personal identifiable information (PII) or whistleblower information into a GenAI platform for any reason.
Give Credit Where Credit is Due
If GenAI is used in the creation of a memo, letter, or any other product, signify its use by including it as a coauthor or by an insignia at the end of the product. This transparency method can assist the office in monitoring use of the tool and understanding how it is being employed by staff.
Sample Guidance to Consider
Accountability for Use
There are many GenAI tools in the marketplace. As institutional guidance from the House and Senate continues to evolve, offices must be clear about what tools are authorized for use within their specific teams. Guidance should also be provided regarding how offices prefer staff to register for the tools’ use. For example, if a House Member’s MRA is being used to pay for an office-wide subscription, guidance should be given to ensure appropriate use of those account credentials.
Appropriate, authorized use of GenAI tools needs to be clearly communicated to staff. This can be done most effectively through use-case examples.
For example, an office can clarify within its internal policy that GenAI can only be utilized to create initial drafts of memos, talking points, and legislative summaries, but that it cannot be used to draft constituent correspondence, the weekly newsletter, or eDear Colleagues.
Guidance can further clarify that all content created by GenAI should be thoroughly reviewed and fact-checked, and that the person responsible for the AI-aided creation of the product is fully responsible for its content.
Office managers need to remain alert to new concerns and risks. GenAI is rapidly evolving and managers should establish clear communications channels with their staff to ensure they feel empowered to raise concerns if they discover an unintended consequence or impact of using these tools. Offices can also take advantage of institutional support offices to increase their team’s education and awareness around GenAI technology, such as requesting a briefing by the House Digital Service or the Government Accountability Office’s (GAO) Science, Technology Assessment, and Analytics (STAA) team.
Transparency of Use
As the Hill adapts to implementing GenAI within its workflows, offices can increase familiarity and awareness of the tool’s use by establishing clear transparency requirements. For example, when GenAI is used for content creation (even if only for a product’s first draft), staff could be required to credit the tool by listing an insignia for the tool (such as “Gen.AI”) as a co-author or note its utilization as a footnote.
When utilizing GenAI for research purposes, staff should be directed to have the tool provide citations for where the data and information is collected to allow staff to double-check accuracy. GenAI may hallucinate at times, so all content, including citations, should be verified and confirmed as valid, fact-based sources.
Privacy & Appropriate Use
The following are a menu of restrictions offices could consider establishing.
Under no circumstances shall staff input any PII into a GenAI tool. This includes, but is not limited to name, address, contact information, or private information that is directly associated with an individual, including Members, staff, constituents, and whistleblowers.
Under no circumstances may staff utilize GenAI to find PII or private information on any individual.
Any and all use of GenAI to intentionally create harmful, unethical, dangerous, or illegal content is forbidden. Should such content be created inadvertently, report the occurrence to senior staff to further inform the office’s best practices.
Any and all use of GenAI is forbidden when working with a whistleblower. Never input any information obtained by a whistleblower into a GenAI platform or any public-data based tool. For further guidance on working with whistleblowers, please reach out to the Office of the Whistleblower Ombuds.
Under no circumstances may staff utilize GenAI to create visual content depicting an individual.
GenAI is a developing technology but hundreds of congressional staff across both chambers have already begun experimenting with its use. Congress can realize substantial efficiency and capacity gains by responsibly utilizing this new technology, but each office must take a proactive approach in communicating what the Member believes to be appropriate uses of these tools. Only through establishing internal office policies will staff feel empowered to use GenAI tools transparently and with an appropriate level of accountability.
