Vermont Tested AI for State IT Security
The state government entered a six-month agreement with Anthropic to pressure-test IT systems and cybersecurity policies.
Updated on Sept. 30, 2026 in Artificial Intelligence
Live Poll
Should government agencies provide more transparency when testing new technologies like artificial intelligence?
Vermont lawmakers held a hearing regarding the state's use of artificial intelligence following a six-month evaluation agreement with Anthropic. The state government is utilizing AI tools within isolated test and development environments to identify system vulnerabilities.
Why it matters
The program aims to bolster state cybersecurity defenses against cyberattacks by leveraging advanced language models for system stress testing. This initiative marks an effort to integrate commercial AI safety protocols into public-sector IT infrastructure.
The state government is leveraging Anthropic’s models under a six-month, no-cost agreement for testing and development. The initiative focuses on simulating cyberattacks to pressure-test existing security policies and identify weaknesses in IT infrastructure.
The players
Vermont Agency of Digital Services
The state department responsible for managing IT infrastructure, cybersecurity policy, and digital system procurement for Vermont.
Anthropic
An artificial intelligence lab focused on the development of large language models and AI safety research, known for its State, Local, Tribal and Territorial program.
Laura Sibilia
A member of the Vermont House of Representatives who participated in legislative oversight regarding state technology agreements.
The details
The Vermont Agency of Digital Services uses artificial intelligence to identify vulnerabilities within existing IT systems and evaluate the resilience of cybersecurity policies. The tools function in restricted test and development environments, allowing researchers to simulate threats and assess defensive responses without exposing production systems. The agency is applying these technologies to identify potential gaps in infrastructure that could be exploited during a cyberattack.
Timeline
June 2026: The agreement with Anthropic was formally executed.
Early August 2026: Rep. Laura Sibilia was informed of the state agreement during a tech summit.
September 30, 2026: Lawmakers convened for testimony regarding the government's use of the technology.
December 2026: The evaluation period for the artificial intelligence agreement is scheduled to conclude.
The Tech Race
Vermont's pilot program positions the state within a growing trend of public-sector entities testing commercial AI to improve cybersecurity resilience. This effort aligns with broader national research initiatives aimed at codifying how large language models can safely interact with government IT systems.
The program currently operates in isolated development environments and does not impact public-facing digital services or data. Residents may see changes in future state procurement policies for security tools after the evaluation ends in December 2026.
The takeaway
Vermont's evaluation highlights the growing state-level effort to stress-test critical infrastructure against modern cyber threats. Watch for the December 2026 project conclusion to see if the state moves toward formal procurement or integration of these tools into live IT systems.
Further reading
For more on the development of these systems, see our Artificial Intelligence coverage.
Source note: This article includes information reported by My NBC5.
Live Poll
Should government agencies provide more transparency when testing new technologies like artificial intelligence?
