Laboratory of Law and Artificial Intelligence
Legal AI Predictions 2025 for the U.S.
Predictions Observatory · Scorecard

141
Predictions
68 authors
6.4
Accuracy
average /10
6.1
Weighted
accuracy × specificity
55%
Correct
78 of 141
3.1
Specificity
average /5
Verdict Distribution
Weighted Score by Category
Key Findings
Regulatory predictions were the most accurate
Predictions about Congressional inaction, revocation of Biden's EO, and state regulatory activism (47 states with deepfake laws) were remarkably accurate. Legislators intensified, not abandoned, the risk-based approach.
Technology progress exceeded expectations
Those who predicted a slowdown in model development were wrong: GPT-5.2 and Claude Opus 4.5 marked historic breakthroughs. Models saturated existing benchmarks.
Adoption was steady, Jevons paradox confirmed
Adoption grew from 14% to 26%. The Jevons paradox was confirmed: AI made lawyers more efficient, but demand grew proportionally, keeping them as busy as before.
The legal AI market boomed
Harvey reached $8B, Clio acquired vLex for $1B, record legaltech funding of $2.4B. VCs paid more attention to legal tech, including Africa.
Agentic AI was the buzzword, not the reality
Agentic AI was the #1 trend (Gartner), but actual adoption was very limited. Intelligent process chains emerged (Clio Work, CoCounsel), but the promised revolution did not materialize.
AI hallucinations remained the Achilles heel
916 documented cases of filings with AI-fabricated citations. Two federal judges were also affected. Human verification remains essential.