The Office of Research Integrity (ORI) has applied its existing research misconduct framework to generative AI, confirming that no separate category for AI-related offenses exists. New regulations, which became applicable to allegations in January, require that misconduct be proven by a preponderance of evidence, involving intentional or reckless departure from accepted practices.
Researchers are now advised to disclose all AI tools used in their work, including during manuscript and grant preparation. According to guidance from Crowell & Moring, institutions should document specific inputs, outputs, and verification steps for every tool utilized. This transparency is intended to aid in reproducibility and provide a defense against potential misconduct allegations.
Determining what constitutes accepted practice remains complex, as standards vary across research fields and funding agencies. For instance, the National Institutes of Health (NIH) has implemented specific restrictions on grant applications developed with AI and prohibits peer reviewers from using generative AI to analyze submissions. Institutions must evaluate each case against these governing standards.
The investigation process follows standard institutional procedures, starting with an inquiry to determine if a full investigation is warranted. ORI recommends that investigative committees include AI specialists to interpret tool functions. Investigators are cautioned against relying solely on automated plagiarism detectors, as these tools may fail to identify copied material within AI-generated scientific text.
Evidence preservation is a critical challenge, as AI prompts and session histories are often scattered across personal devices or external platforms. Institutions are generally required to retain records for seven years after a proceeding concludes. Intentionally destroying relevant data after an allegation arises can be used as evidence of misconduct in itself.
Crowell & Moring suggests that institutions update their internal training and disclosure policies immediately. Organizations should identify AI experts, assess vendor data retention practices, and ensure that AI usage is verifiable before allegations occur. Meanwhile, the NIH may independently pursue grant-management remedies, such as suspension, while misconduct cases proceed.
