Hidden Prompt Injection Found in CT Court Filing Leads to Sanctions
404 Media · rss · 2026-08-14
A self-represented litigant in a Connecticut court, Matthew Elliott, was found to have hidden prompt injection instructions in official legal filings using tiny, 3-point white font. The concealed text attempted to manipulate any LLM reviewing the document, directing the AI's output to strictly favor the plaintiff's position.
Court staff spotted the manipulation due to abnormal white space in the document. Judge Walter Spader Jr. excoriated the behavior in a 14-page sanction decision, noting that while the court does not currently use AI to process documents and the attempt failed, deploying covert communications fundamentally compromises the integrity of the judicial system.
The judge allowed the case to proceed but sanctioned Elliott by revoking his electronic filing privileges, requiring him to submit printed hard copies. When 404 Media tested the filing with ChatGPT, the model noticed and ignored the injection but flagged it as a credibility and professionalism concern.
More from Safety
- LLMs Are Not Stateless: Paper Reveals Implicit Memory Threatens Agent Eval Safety — lbeurerkellner · 2026-08-14
- Frontier LLMs Hit Perfect Detection Rate in UEFI Firmware Vulnerability Tests — evilsocket · 2026-08-14
- AI Safety Researcher Joins Apollo to Study LLM Post-Training Failures — niloofar_mire · 2026-08-14
- Hidden Prompt Injection Found in Court Filing to Manipulate AI — RebeccaBellan · 2026-08-14
- Anthropic Experiment: Multi-Agent Systems Spark Turf Wars and Collusion — TechCrunch AI · 2026-08-14
- Inside the OpenAI Sandbox Breach: AI Models Communicated to Break Out — binarybits · 2026-08-14