ANNAPOLIS, Md. — Maryland officials announced Dec. 30, 2025, that the state secured more than $2.6 million in grants over two years from the Center for Civic Futures’ Public Benefit Innovation Fund to develop artificial intelligence tools aimed at improving access to public benefits, including streamlining work verification for the Supplemental Nutrition Assistance Program and Medicaid amid new federal requirements under H.R. 1.
Gov. Wes Moore highlighted the initiative as a way to harness AI responsibly to reduce barriers to essential services like nutrition and health care. The $1.2 million portion of the funding supports a multi-state project led by the Maryland Department of Human Services, Maryland Department of Health, Maryland Benefits and the Maryland Health Benefit Exchange through the American Public Human Services Association. This effort focuses on creating open-source AI systems to automate and accelerate verification processes, with Maryland anchoring a cohort for tool development and deployment across states.
The grants follow H.R. 1, the budget reconciliation law signed July 4, 2025, which expands work requirements for SNAP and Medicaid recipients, shifts administrative cost-sharing to 75 percent state responsibility from 50 percent, and introduces benefit cost-sharing penalties for states with payment error rates above 6 percent. Maryland’s fiscal year 2024 SNAP error rate of 13.64 percent, one of the nation’s highest, could trigger a 15 percent state share of benefits, adding up to $240 million annually based on projected $1.6 billion in 2026 issuances. Combined with the admin cost increase from $115 million to $172.5 million, total state expenses could rise over $300 million yearly.
State estimates indicate AI tools could help mitigate these burdens by improving accuracy in eligibility determinations and reducing administrative overhead, though officials have not specified projected savings. The two-year funding supports testing in sandbox environments with human oversight to ensure compliance with Maryland’s Responsible AI Policy, which requires reviews for bias, privacy protections and no automated decisions without human input.
AI in work verification could automate document review, data extraction from pay stubs or employment records, and cross-checking against state databases, potentially cutting processing times from weeks to days. This efficiency addresses H.R. 1’s mandates, such as verifying 80 hours monthly of work or qualifying activities for able-bodied adults aged 18-64, including parents of children 14 and older. Faster verifications could prevent benefit disruptions for compliant recipients, reducing churn where eligible individuals lose coverage due to paperwork delays.
Beneficiaries include low-income families, with Maryland’s 800,000 SNAP participants in 2025 averaging $200 monthly per person. Quicker access supports food security, particularly in Southern Maryland counties like Charles, Calvert and St. Mary’s, where rural areas face transportation barriers to in-person verifications. State agencies benefit from reduced workloads; caseworkers could focus on complex cases rather than routine checks. Taxpayers gain from potential error reductions, avoiding federal penalties and optimizing $172.5 million in admin spending.
Broader impacts extend to multi-state collaboration, sharing tools to standardize processes and lower development costs. AI-powered digital assistants could provide 24/7 multilingual support, aiding non-English speakers in navigating applications and verifications, enhancing equity in access.
However, dangers persist. Historical AI implementations in welfare systems have led to wrongful denials; Arkansas’ 2016 algorithm cut home care hours for thousands of disabled individuals, prompting lawsuits. Michigan’s 2013 fraud detection system falsely accused 40,000 of unemployment fraud, with 93 percent of cases later overturned, causing bankruptcies and suicides. Bias in training data could disproportionately affect minorities or low-income groups, exacerbating disparities in Maryland’s diverse population.
Privacy risks arise from handling sensitive data like income records, potentially vulnerable to breaches. AI hallucinations—generating false information—could lead to inaccurate verifications, denying benefits erroneously. Over-reliance might displace caseworkers, reducing human empathy in decisions. Transparency issues make contesting AI outputs difficult, as seen in Indiana’s 2016 automation denying 1 million applications, a 54 percent increase, including terminal patients.
Experts recommend human-in-the-loop designs, regular audits for bias and clear appeal processes. Maryland’s policy prohibits using personal data for training without consent and mandates ethical reviews, but implementation challenges remain.
The grants also fund a $1.4 million project by the Maryland Department of Labor with Code for America to develop AI for unemployment insurance, including fraud detection and claimant support. These initiatives align with Moore’s digital modernization, including the relaunched Maryland.gov on Jan. 6, 2026, simplifying benefits access.
As development proceeds, stakeholders monitor for equitable outcomes, with potential for cost savings if error rates drop below 6 percent, avoiding $240 million penalties.