Loading...
HomeMy WebLinkAboutPort Townsend AI Policy I. PURPOSE CITY OF PORT TOWNSEND Subject: ArƟficial Intelligence (AI) Policy Date: October 15, 2025 Authority: City Manager This policy outlines the requirements for acquiring and using supporƟve generaƟve arƟficial intelligence (AI) soŌware at the City of Port Townsend. This policy only applies to supporƟve generaƟve AI. At this Ɵme neither responsive generaƟve AI nor predicƟve AI tools will be implemented by the City of Port Townsend unless in a controlled internal-only tesƟng environment for select staff. The City authorizes the use of supporƟve AI while users are conducƟng duƟes related to their jobs with the understanding that the output from these tools may contain inaccuracies and will require human verificaƟon. All users of City compuƟng and network resources must do so in an ethical, legal, and responsible manner. All use of technology resources must be consistent with all City policies and work rules. II. I. SCOPE This policy applies to all City employees. This policy also includes contract or temporary employees, volunteers, and elected officials when working on behalf of the City. III. DEFINITIONS a. GeneraƟve AI: A class of soŌware that uses large language models, deep learning, and machine learning to generate new content (e.g., text, images, video, audio) based on input data. Examples include GPT-4/ChatGPT (OpenAI, the AI system used by Copilot), Gemini (Google), and DALL-E. There are two types of generaƟve AI:  i. SupporƟve generaƟve AI tools (SupporƟve) – AI tools capable of generaƟng text, images, or other data from sources not-limited to City-generated content in response to employee-provided prompts.  ii. Responsive generaƟve AI tools (Responsive) – AI tools that interact directly with community members and employees to provide informaƟon from the City’s website or similar discrete set of City-generated or City-curated content. B. PredicƟve AI: Uses staƟsƟcal analysis and machine learning to idenƟfy paƩerns, anƟcipate behaviors, and forecast future events. It essenƟally uses data to predict potenƟal outcomes based on past trends and relaƟonships. OrganizaƟons use this type of AI to predict things like customer behavior, potenƟal risks, or future market trends. C. Algorithm: A series of logical steps used by a computer program to process inputs into outputs. D. Algorithmic bias: A systemaƟc and unfair preference or prejudice embedded in the outcome of an algorithm, oŌen due to biased data, flawed assumpƟons, or design choices. E. Biometrics: Biological and physical characterisƟcs used to idenƟfy individuals. These include:  Fingerprints  Facial recogniƟon  Iris scans  Voice paƩerns  DNA F. Protected Classes: Groups of people who are legally protected from discriminaƟon under various laws and regulaƟons to prevent unfair treatment in areas like employment, housing, educaƟon, and access to services. These classes are typically based on characterisƟcs such as:  Race  Gender  Age  Religion  Disability  NaƟonal origin  Sexual orientaƟon (in some jurisdicƟons) G. AI System: Any system that generates outputs such as predicƟons, recommendaƟons, or decisions that augment or replace human decision-making. This includes soŌware, hardware, algorithms, and data used for automaƟon or analysis. While this policy document includes principles that apply to AI technologies generally, these policy statements apply only to GeneraƟve AI systems. H. Prompt: In AI, a query or instrucƟon that a user provides to an AI model to generate a response. The quality of the data returned is directly influenced by the quality of your prompt. I. SoŌware as a Service (SaaS): A cloud -based model that users access on-demand, usually through a web browser, without having to install or maintain any so Ōware on their computer. J. EmoƟonal analysis: The use of technology (oŌen AI) to detect and interpret human emoƟons from data such as facial expressions, voice tone, text, or physiological signals. E.g. customer service, surveillance, markeƟng, mental health monitoring. K. Social scoring: The use of data analyƟcs to assign individuals a score based on their behavior, reputaƟon, or social interacƟons—oŌen by governments or companies. L. Personally IdenƟfiable InformaƟon (PII): In the context of this AI policy, the following are considered PII: DefiniƟon of Personal InformaƟon Our state’s noƟficaƟon law defines “personal informaƟon”  A Washingtonians’ first name or iniƟal and last name in combinaƟon with one or more of the following: o Social Security number; o Driver’s license number or Washington idenƟficaƟon card number; o Account number or credit or debit card number, in combinaƟon with any required security code, or password that would permit access to their account; o Full date of birth; o Private keys for electronic signature; o Student, military, or passport idenƟficaƟon numbers; o Health insurance policy or idenƟficaƟon numbers; o Medical informaƟon, including medical history, mental or physical condi Ɵon, diagnoses, or treatment; and o Biometric data.  Any of the above elements, not in combinaƟon with first name or iniƟal and last name, if the affected data was not rendered unusable via encrypƟon or redacƟon and would enable a person to commit idenƟty theŌ against the consumer.  Username and email address in combinaƟon with a password or security quesƟons and answers that would permit access to an online account. Personal informaƟon does not include publicly available informaƟon that is lawfully made available to the general public from federal, state, or local government records. hƩps://www.atg.wa.gov/washington-s-data-breach-noƟficaƟon-laws IV. AI GENERAL CODE OF CONDUCT These principles describe general codes of conduct that represent the City’s values and align with its responsibiliƟes to the people it serves. City employees, volunteers, and elected officials shall adhere to these principles and requirements outlined in this policy and will be held accountable for any work product AI tools generate and must ensure compliance with these expectaƟons: A. InnovaƟon and Sustainability: The City values innovaƟon that meets community needs while advancing beneficial outcomes for people and the environment. We commit to responsibly exploring AI technologies that improve services. B. Transparency and Accountability: The City will be transparent in its use of AI and will review results to ensure applicable laws are complied with. DocumentaƟon of AI systems—including data sources and operaƟonal models— will be maintained, such as a log. C. Quality and quality control: Validity and Reliability are paramount. AI systems must perform reliably under expected condiƟons; however, human users are accountable to ensure accuracy. System accuracy will be governed and reviewed throughout the lifecycle to inform policy improvements and training. D. Privacy ProtecƟon: The City prioriƟzes data privacy. Policies will be in place to safeguard personally idenƟfiable informaƟon (PII) throughout AI system development and use. PII or any sensiƟve data shall not be entered into any AI system without prior wriƩen approval from the user’s supervisor and the Finance Director. SupporƟve generaƟve AI use must abide by SecƟon II.L relaƟng to personally idenƟfiable informaƟon or other sensiƟve informaƟon. E. Security and Resiliency: Should the City allow future Responsive G eneraƟve AI, the City will ensure that AI systems are evaluated for resilience against security risks. These systems must protect the confidenƟality, integrity, and availability of City data. F. The City will only allow SupporƟve GeneraƟve AI use. Responsive GeneraƟve and PredicƟve AI outside of controlled tesƟng environments for select staff will not be used without policy modificaƟon. V. POLICY 1. RESPONSIBILITIES The following enƟƟes and individuals have specific responsibiliƟes in the implementaƟon and adherence to this policy: • InformaƟon Technology: Assist with policy development and maintenance, risk assessment, and planning. Support the City and its departments with training and collaboraƟon between departments. • City Staff: Assume full responsibility for the accuracy of all work products including data obtained using any AI systems. Staff are also responsible for ensuring strict compliance with this policy. Staff are also responsible for including proper aƩribuƟons. City staff will also submit AI soŌware requests to the AI policy team and network administrator or designee to secure approval for the acquisiƟon and use of new AI systems. • Legal Advisor: Provide expert guidance on potenƟal legal risks associated with AI usage within the City. • AI Policy Team: An ad-hoc interdepartmental City team will be convened as needed to oversee updates to the AI policy, periodically assess bias and review of issues, recommend training, and review proposals for AI subscripƟons and services. 2. ACQUISITION OF GENERATIVE AI TECHNOLOGY A. Employees may request to use various AI systems by submiƫng an “AI Use Request Form” to their Director who will forward to HR for AI Policy Team for consideraƟon. This team will collaborate with the requestor’s Director to review the request and provide a response. B. No addiƟonal approval is required if a previously approved technology integrates generaƟve AI capabiliƟes, e.g. Zoom now integrates a free AI feature. However, this policy will sƟll apply and the City through the AI Policy Team may discuss potenƟal impacts to prepare for possible consequences of new tools. C. The AI Policy Team will review requests based on risk specific to GeneraƟve AI technology for those technologies not covered in SubsecƟon B. D. The AI Policy Team may recommend the City Manager revoke authorizaƟon of specific AI tools, including if new AI capabiliƟes present risks that violate this policy pertaining to data privacy, prohibited uses, public records, aƩribuƟon, transparency and accountability. 3. EMPLOYEE TRAINING A. Staff may receive annual training, if funding is available, on basic AI concepts, responsible use of approved systems, and ethical consideraƟons. Training may be online or in-person. B. GeneraƟve AI is a tool that employees may use to augment their work, but it is not a subsƟtute for work responsibiliƟes. As with all work produced, employees remain responsible for their judgment and use of AI tools, including the outcomes of AI use. 4. USE OF GENERATIVE AI OUTPUTS All data received from a GeneraƟve AI system must undergo human review before official use by the City. Responsive generaƟve and predicƟve AI may not be used unless in a limited tesƟng capacity for select internal City staff unƟl the City has staff capacity, experience, and training in these areas for policy modificaƟon. 5. ATTRIBUTION, ACCOUNTABILITY, AND TRANSPARENCY OF AUTHORSHIP A. Images or videos created by GeneraƟve AI systems must include aƩribuƟon where feasible (e.g., watermarks, capƟons, or footnotes). Videos created using AI must be reviewed and approved by the creator’s supervisor or delegate before release. B. Text generated by an AI system must include aƩribuƟon if used substanƟvely in the final work product. C. Source code generated by an AI system must include aƩribuƟon via comments if used for criƟcal funcƟons. D. All aƩribuƟons should include the name of the AI system used plus a “human in the loop “(HITL)” asserƟon (include the department, division, and/or staff member who reviewed the content). Example aƩribuƟon: "This material -was generated using ChatGPT 4.0 and reviewed by the Public Works Director before publicaƟon on MM/DD/YYYY." E. As with all work products at the City, AI users will be responsible for 100% of any AI products. AI users will be responsible for accuracy and compliance with current standards, regulaƟons, and laws. F. All consultant contracts shall sƟpulate AI aƩribuƟon requirements for work substanƟally produced by AI. 6. REDUCING BIAS AND HARM Employees are encouraged to work with the City’s AI Policy Team when using GeneraƟve AI systems that analyze City data or inform decisions to ensure results are free from bias against protected classes. 7. DATA PRIVACY A. Use of GeneraƟve AI systems must align with the City's Data Security and Privacy Policy, Policy Number #XXXXXXXXX, and the City of Port Townsend Employee Manual. B. Employees shall not submit non-public City data into GeneraƟve AI systems. C. City data should not be used for training external GeneraƟve AI models. D. Staff shall not upload confidenƟal, sensiƟve, or personally idenƟfiable data to any AI soŌware or service. 8. PUBLIC RECORDS AND CITY RECORDS MANAGEMENT A. Records created by generaƟve AI are likely to be considered public records. For addiƟonal informaƟon, consult the Public Records Officer for guidance. B. Approved supporƟve generaƟve AI systems must support storing and retrieving all prompts and outputs. The City considers most records generated using AI tools to be transitory or draŌ unƟl the work product is finalized. Employees are responsible for ensuring that any prompts or generated work product with retenƟon value is captured. C. Employees must maintain records of inputs, prompts, and outputs consistent with public records requirements, which depends on the content of the record. See below for examples. AI users should only use accounts created with their City email addresses to ensure that the City meets records retenƟon requirements. Personal accounts should NOT be used for City business, including AI systems. Example 1: If staff uses AI to assist in draŌing a City publicaƟon, arguably the AI-related records are “draŌs”. Under applicable retenƟon schedule, City only needs to retain unƟl no longer needed and then destroy. Example 2: If staff uses AI to assist in draŌing a resoluƟon or ordinance, the applicable retenƟon schedule for records related to the “development” of a resoluƟon or ordinance is to retain for 3 years aŌer approval or denial and then destroy. 9. PROHIBITED USES Using AI for the following is prohibited: • RequesƟng or obtaining legal advice related to City business. • Decisions requiring human decision-making (i.e. permit applicaƟons, approvals, etc.). • Real-Ɵme biometric idenƟficaƟon. • CogniƟve behavioral manipulaƟon. • Social scoring, or the use of AI systems, to track and classify individuals based on behaviors, socioeconomic status, or personal characterisƟcs. • EmoƟon analysis, or the use of computer vision techniques to classify human facial and body movements into certain emoƟons or senƟment (e.g., posiƟve, negaƟve, neutral, happy, angry, nervous). • Fully automated decisions that substanƟally impact the rights or safety of individuals with no meaningful human oversight. For example, idenƟfying and/or manipulaƟng appropriate egress during an emergency situaƟon 10. NON-COMPLIANCE Non-compliance with this policy may result in disciplinary acƟon up to terminaƟon of employment, volunteers, or contracts. 11. RELATED STANDARDS AND POLICIES • Data Privacy & Security Policy • City of Port Townsend, Employee Manual 12. DOCUMENT CONTROL This policy shall be effecƟve on October 15, 2025, and shall be reviewed annually to incorporate new AI developments. It is acknowledged that AI support is an emerging and rapidly changing area of technology and the City Manager is authorized to amend these policies at any Ɵme in order to comply with legal and regulatory requirements. Approved By: John Mauro, City Manager