S07599 would create a new framework in New York for the use of automated decision-making tools by government agencies. The bill defines automated decision-making tools broadly to include software using algorithms, computational models, or artificial intelligence to automate, support, or replace human decision-making, while excluding basic office software and certain internal administrative tools. It requires state agencies that use such tools to publicly disclose them on agency websites, and it directs agencies to prepare recurring impact assessments before use and at least every two years thereafter, as well as before any material change to a tool.
The required impact assessments are extensive. They must evaluate the tool’s objectives, underlying algorithms and training data, accuracy, fairness, bias and discrimination, cybersecurity and privacy risks, public health and safety risks, foreseeable misuse, and how sensitive personal data is handled. If an assessment finds discriminatory or biased outcomes, the agency must stop using the tool and any information produced by it. The bill also requires submission of assessments to the governor and legislative leaders, publication of assessments on agency websites, and allows limited redactions for safety, privacy, and security reasons.
The bill also creates a statewide inventory of state automated decision-making tools to be maintained by the Office of Information Technology Services and posted on the state open data website. In addition, agencies that already use such tools must provide the legislature with disclosures about those tools, including vendors, purpose, start date, and any impact assessments. The bill amends education and civil service laws to make clear that use of artificial intelligence systems and automated decision-making tools cannot diminish collective bargaining rights, displace employees, transfer existing duties to machines, or alter civil service status and related employment protections.
Overall, the sentiment reflected in the bill’s progress is strongly favorable and bipartisan. It passed the Senate Internet and Technology Committee 6-1, then passed the Senate 59-0 and the Assembly 143-2, indicating broad support for transparency, oversight, and guardrails around government use of AI. The bill’s structure suggests a policy approach that is cautious but not prohibitive: it allows agency use of automated tools, but only with disclosure, human review, and accountability requirements.
The main points of contention appear to center on the scope of disclosure, the burden of recurring impact assessments, and the possibility that some information may need to be redacted for security or privacy reasons. The labor-related provisions also indicate concern about automation replacing public employees or weakening collective bargaining rights, especially in education and civil service settings. Supporters likely view these protections as necessary safeguards, while any opposition would likely focus on administrative burden, operational flexibility, and the risk that disclosure requirements could expose sensitive government systems.
The bill would add a new article to the state technology law governing automated decision-making by government agencies, require statewide inventory and public disclosure of such tools, and mandate recurring impact assessments with specific testing for bias, privacy, cybersecurity, and safety risks. It also amends the education law and civil service law to protect employee rights, collective bargaining relationships, and civil service status from being undermined by AI or automated decision-making systems. Affected parties include state and local government agencies, the Office of Information Technology Services, public employers, school districts and education entities, public authorities, and public-sector employees and unions.
The legislative record shows strong support for the bill. It advanced out of committee with only one dissenting vote and then passed both chambers overwhelmingly, including unanimous Senate floor approval and near-unanimous Assembly approval. That voting pattern suggests broad agreement that government use of automated decision-making should be transparent, reviewed for bias and risk, and constrained by human oversight and labor protections.
The most notable areas of contention are the practical and policy limits of disclosure and oversight. Agencies may need to balance transparency against protecting cybersecurity, privacy, and sensitive operational information, which the bill addresses through redaction provisions. Another likely point of debate is the labor-protection language, which explicitly bars automation from displacing workers, reducing hours or benefits, or transferring existing duties to AI systems; this reflects concern from public-sector employees and unions about job loss and erosion of bargaining rights. Any opposition would likely come from those worried about compliance costs, administrative complexity, or reduced flexibility in deploying AI tools.