Delaware’s gen AI policy allows ChatGPT, but not DeepSeek
Delaware’s Department of Technology and Information adopted a policy this month that outlines how and when state employees can use generative artificial intelligence tools, and permits them to use many public and enterprise AI models, but not ones considered to be potentially malicious, like DeepSeek.
The policy, dated July 3, provides state employees with a framework to follow when considering the use of generative AI for state functions and data sharing. It prohibits state employees uploading any sensitive data owned by the state into public AI tools as that data could be made public or used to train the models.
The guidance also encourages employees to independently verify all output to avoid promoting false, biased or harmful results.
The guidance limits use to state purposes, but employees are approved to use public versions of AI tools like like ChatGPT and Gemini, and enterprise versions such as Microsoft Copilot and Amazon Bedrock. The state prohibits use of “malicious AI models” like RedNote or DeepSeek, both of which are developed by Chinese companies.
The policy prohibits using AI tools on personal devices or personal accounts to circumvent state policies or safeguards. It also urges employees to act responsibly with regard to AI and to label any AI-generated content on internal or outward-facing communications.
The policy also states that agencies are “strongly encouraged” to require employees to complete training before granting them permission to use generative AI tools for work purposes, to mitigate cybersecurity risks.
