Georgia Court Order Contains AI-Hallucinated Legal Citations
A Georgia Supreme Court appeal hearing exposed significant citation errors in a trial court order, with evidence suggesting AI-generated hallucinations were involved. During arguments in a murder conviction appeal, Chief Justice Nels S.D. Peterson identified serious problems with the trial court's order denying a new trial.
Key Details from the Hearing
According to the source material, Chief Justice Peterson stated the order contained:
- "At least five citations to cases that don't exist"
- "At least five more citations to cases that do not support the proposition for which they're cited"
- "Three quotations that don't exist"
The prosecutor, Leslie, responded that her initial submitted order had been revised and took no responsibility for the errant citations. However, Chief Justice Peterson countered: "Those nonexistent cases were cited in your initial brief opposing the motion for a new trial."
Documentation Available
The source indicates two key documents are available for examination:
- A 33-page order denying a new trial
- A 37-page proposed order from the state
This incident highlights the risks of using AI tools for legal research and citation without thorough verification. While the source doesn't specify which AI tool was used, the pattern of hallucinated cases and misattributed quotations matches known limitations of current large language models in legal contexts.
📖 Read the full source: HN LLM Tools
👀 See Also
19 Days to a Reddit App: Prestige Tracks Community Contribution Beyond Karma
A developer built Prestige, a Reddit app for seasonal prestige, community rankings, and Hall of Fame tracking, using OpenClaw AI.
AI Data Center Water Use in California: Estimates from Physics and AI Models
A California WaterBlog analysis using physics and four AI models estimates AI data center water use in California at 2,300–400,000 acre-ft/year, with a realistic range of 32,000–290,000 acre-ft/year — modest compared to agriculture.
Research Findings on AI Agent Reliability and Development Patterns
A collaborative research session with Claude Opus analyzed 15 papers on AI agents, revealing quantified reliability problems: agents produce 2-4 different action sequences across 10 runs, with 69% of divergence occurring at the first decision. Self-improving agents showed safety refusal rates dropping from 99.4% to 54.4% through their own learning.
Anthropic Secures 300MW Compute at Colossus 1 with 220,000 NVIDIA GPUs via SpaceX Partnership
Anthropic announced a partnership with SpaceX to use all compute capacity at the Colossus 1 data center, gaining over 300MW and more than 220,000 NVIDIA GPUs within a month.