Feds need to be careful when tapping generative AI for work

Anucha Tiemsom/Getty Images

Human review of AI-generated outputs is critical, OPM says in new guidance for government employees.

Generative artificial intelligence can help federal employees do their jobs better in certain cases, but there are risks to watch out for, according to the Office of Personnel Management.

The federal HR shop issued guidance today on the Responsible Use of Generative AI, which points out some key pitfalls — as well as benefits — of using the emerging technology in federal work, such as document drafting, coding, translation, data analysis and more.

First, it's important to make sure your agency has cleared employees to use GenAI apps in the workplace. Agencies have different policies about onboarding tools that assist in writing, computer coding and image generation. While these tools are often freely available over the open internet, it's up to agencies to authorize access. Some agencies may have their own enterprise GenAI tools. In any case, OPM notes, feds should "expect that your use of GenAI technology may be logged and monitored."

Even approved tools can generate outputs that aren't suitable for public-facing or even internal use, because of the potential for copyright infringement and plagiarism, as well as the release of personal information contained in training materials used by these large language models. Additionally, OPM notes the risk "to agency credibility if content is inaccurate, unreliable or discriminatory."

OPM urges feds to impose human checks on AI outputs before any publication or dissemination of written materials or AI-generated images. It is especially important to "review AI-generated material for potentially biased, harmful, stereotypical, or otherwise offensive language or images," the guidance states. Additionally, some agencies may require some disclaimer or watermark that identifies AI-generated material. 

The guidance comes 180 days after the release of the Biden administration's executive order on AI. The White House noted separately that all federal agencies completed the 180-day deliverables contained in the order on schedule. These include, on the government operations front, guidance from the Department of Labor's Office of Federal Contract Compliance Programs on deploying AI in the workplace, as well as multi-agency guidance on the responsible and equitable use of GenAI in public benefits programs.