DoIT publishes AI strategy, plans governance, inventory publication and training expansion

Nishant Shah, DoIT’s senior adviser for Responsible AI, told the House Health and Government Operations Committee that 2024 was a year of building scaffolding and that DoIT has moved from initial guidance to a published AI strategy and a plan for governance and adoption.

Shah said the year began with the governor’s AI executive order and the passage of SB 818, and that DoIT and the AI subcabinet have "created and disseminated initial guidance for state agencies detailing how they can leverage generative AI tools responsibly and ethically and productively." He added the subcabinet completed a 2025 AI strategy and study roadmap in December and DoIT released the finalized public version on its website.

Shah described five strategic pillars DoIT will emphasize in 2025: mature AI governance (including intake and risk classification); strengthen state data foundations (led by the Office of Enterprise Data); scale experimentation and adoption through beachhead use cases and proofs of concept; increase workforce AIIQ (training and hires); and study sector‑specific approaches for critical domains such as education, infrastructure and health care.

DoIT said it has collected the state’s first AI inventory of live AI use cases and is validating that inventory with agencies before publishing. The department also said it has made free asynchronous AI courses and workshops available to state employees through Innovate US (a 501(c)(3)) and that "hundreds" of employees have taken those offerings.

Shah described an enforcement and governance posture that starts with risk classification: low, medium or high risk. High‑risk use cases, he said, will require deeper assessment, testing, validation and verification and attention to whether a human remains in the loop for decisions that affect rights or safety. "If it is high risk ... it will require a deeper risk assessment," Shah said, adding the state will require testing, evaluation, validation and verification for those projects.

Shah and DoIT said the AI enablement team has begun work and will act as a center of excellence to help agencies evaluate use cases, run pilots and eventually integrate proven tools into production with governance controls.

Nut graf: DoIT’s AI program seeks to balance rapid experimentation with procedural safeguards by requiring risk classification and assessments, publishing an inventory, training staff, and creating a centralized enablement team to shepherd agency use cases from proof of concept to governed production.

DoIT outlined near‑term steps: publish the AI inventory; require classification and risk assessments for use cases; continue workforce training and role‑specific upskilling; and run sector studies as required by SB 818. Shah said the guidance and governance documents will be living artifacts that DoIT expects to update frequently as the technology evolves.

Ending: The committee pressed DoIT on human oversight, procurement and workforce impacts; DoIT said training and the AI enablement center will emphasize augmentation of staff workflows, not wholesale workforce replacement, and will provide role‑specific guidance for procurement and procurement review.