Philadelphia officially has a new AI policy for municipal workers that restricts the use of consumer tools and mandates strict risk assessments for new software.\u00a0<\/p>\n
However, the infrastructure needed to actually enforce these rules appears to be missing, according to public records obtained by Technical.ly via a right-to-know request.<\/p>\n
The new policies follow a contentious City Council hearing last fall, where council members grilled OIT and the mayor\u2019s office representatives over their handling of AI.\u00a0<\/p>\n
At the end of March, the city\u2019s Office of Innovation and Technology (OIT)<\/a> published an AI use policy for city departments. The guidelines outline approval standards for AI tools<\/a>, rules for employee use of generative AI and plans to manage the technology\u2019s impact on city services.<\/p>\n The policy repeatedly mentions an \u201cAI governance committee\u201d<\/a> that\u2019s responsible for approving tools and ensuring compliance, but it\u2019s not clear if it exists. In response to Technical.ly\u2019s public records request, the city explicitly stated there are no records regarding the committee.<\/p>\n Councilmember Rue Landau, chair of City Council\u2019s Technology and the Information Services Committee, hosted a hearing last fall on how the city is using AI and its plans for future policy. She acknowledged the lingering uncertainty surrounding the governance committee\u2019s structure moving forward.<\/p>\n \u201cThere are some important questions to sort through,\u201d Landau told Technical.ly in June, \u201clike how this policy will be put into practice, how it will be implemented by law-enforcement agencies and what the makeup of the governance committee will be.\u201d<\/p>\n OIT officials did not respond to repeated requests for clarification on this issue.<\/p>\n What the new compliance policy actually requires<\/p>\n The city\u2019s AI policy focuses heavily on vendor vetting and risk mitigation<\/a>.<\/p>\n OIT, in partnership with the Law Department, requires \u201call departments, agencies, employees and contractors\u201d to track, log and justify every tool they plan to use.<\/p>\n That includes an inventory of active AI systems kept by the governance committee and regular evaluation of tools to ensure they meet the city\u2019s standards. OIT plans to update these procedures as technology evolves and will publish them \u201cwherever feasible.\u201d<\/p>\n The framework also mandates equity impact tests to measure how AI systems affect residents and employees.\u00a0<\/p>\n OIT, the Law Department and the AI governance committee are meant to subject all new software to risk and security assessments to align with city data standards, while reviewing third-party contracts to guarantee data security and transparency.<\/p>\n The new guidelines restrict employees to only use licensed \u201centerprise\u201d versions of tools and ban unapproved \u201cconsumer\u201d generative AI. The text does not specify which apps belong in which category.\u00a0<\/p>\n Employees using approved generative tools must review all outputs for mistakes and bias.<\/p>\n To ensure compliance, the city mandated AI training starting April 1, according to CIO Melissa Scott at a budget hearing in March<\/a>. OIT and the Law Department hold the power to halt the use of any non-compliant tool.<\/p>\n At the hearing, Scott also explained the core principles behind the policy\u2019s specifics.<\/p>\n \u201cThe AI policy \u2026 is anchored in three core principles,\u201d Scott said. \u201cEthical utilization, robust governance frameworks [and] data security emphasizing bias reductions, human oversight and stringent data protection measures.\u201d\u00a0<\/p>\n Addressing public pushback and ethical questions<\/p>\n