Federal Court Strikes Down AI-Powered Grant Cancellations as Unconstitutional
Key Takeaways
- ▸Federal court ruled that ChatGPT-based analysis used to cancel federal grants violated the First Amendment and Equal Protection Clause
- ▸DOGE used ChatGPT to automate discriminatory targeting of 1,400+ grants based on crude DEI keyword matching, producing nonsensical and constitutionally impermissible classifications
- ▸AI tools cannot be used to systematize or mask discriminatory government decision-making; the government remains liable for discriminatory prompts and outputs even when human approval follows
Summary
In a landmark ruling on artificial intelligence's role in government, a federal court found that the Trump administration's use of ChatGPT to identify and cancel over 1,400 federal grants related to diversity, equity, and inclusion (DEI) violated the First Amendment and Equal Protection Clause. The case, American Council of Learned Societies v. National Endowment for the Humanities, revealed that DOGE staffers fed grant descriptions into ChatGPT with a discriminatory prompt asking if grants related to DEI, leading to absurd and constitutionally impermissible outcomes—from flagging a whaling museum to terminating funding for a Holocaust documentary about Jewish women's slave labor.
The court's analysis exposed how ChatGPT's crude classification system repeatedly flagged grants based solely on references to particular races or genders, demonstrating the tool's systematic discriminatory application. DOGE staffers then adopted these AI outputs wholesale without meaningful human review or analysis, creating a clear paper trail of discriminatory intent. The government attempted to distance itself from ChatGPT's outputs, arguing the AI dialogue merely provided "context," but the court rejected this defense, holding that DOGE both selected the AI tool and formulated the discriminatory prompt that led to the flagged grants.
The ruling establishes that government agencies cannot use AI systems as a shield for discriminatory decision-making, nor can they outsource constitutional obligations to algorithms. The case raises urgent questions about accountability and transparency when AI is embedded in high-stakes government processes, and highlights the risks of deploying large language models for policy decisions without adequate human oversight and consideration of algorithmic bias and fairness.
- The case exposes critical risks in deploying LLMs for high-stakes government policy without meaningful human oversight, and demonstrates that discriminatory inputs reliably produce discriminatory outputs
