Legal AI's 90% Recall Rate Is the Displacement Signal We've Been Waiting For — But the Forecast Hinges on What 'Publicly Attributed' Actually Means
textak's white-collar displacement forecast sits at 73% — and today's legal AI data is the most direct evidence we've seen yet that the phenomenon is real at scale. Harvey and CoCounsel achieving 90%-plus recall on document review, outperforming human reviewers by 20-30 percentage points, while firms report 30-70% time reductions: this is not a pilot result or a benchmark score. This is operational displacement. The question our forecast actually turns on, though, is attribution — and that's where today's evidence gets complicated.
Let's be precise about what 73% reflects. We weight the phenomenon side of this forecast heavily — the underlying displacement is clearly happening, and today's legal AI data is direct evidence of it. Contract attorneys billing $40-60/hour are being priced out of first-pass document review by systems that outperform them on recall while costing a fraction as much. Firms are restructuring hiring toward experienced laterals over new graduates. This isn't attrition-based replacement, which is the soft version we've been watching for — this is structural restructuring driven explicitly by AI capability. On the phenomenon, we'd put probability near 85%.
But our forecast target requires public attribution, not just displacement happening. And here's where today's news creates a genuinely useful analytical fork. The Colorado AI Act, effective June 30, creates disclosure requirements around algorithmic employment decisions — but critically, it requires safeguards against discrimination, not public attribution of layoffs to AI. Governor Newsom's California executive order is closer: it directs evaluation of worker adjustment notification rules specifically for AI-driven displacement. These are the first US state frameworks that create legal infrastructure around the concept of AI-attributed workforce changes. We read this as a slow ratchet, not a trigger.
The strongest counterargument to our 73% isn't that displacement isn't happening — it clearly is. It's that companies have every incentive to describe these restructuring decisions in the language of 'efficiency' and 'operational transformation' rather than 'AI replaced these roles.' The legal AI evidence today is actually a case study in this: the firms reporting 30-70% time reductions aren't issuing press releases titled 'We Replaced Our Contract Attorney Bench With Harvey.' They're describing workflow improvements. The White House's National Policy Framework actively discourages new federal AI disclosure mandates, which removes one structural forcing mechanism we'd been watching for.
What would move us above 80%: a major firm — AmLaw 50 or a Fortune 500 in-house team — publicly quantifying headcount reduction and explicitly naming AI automation as the cause in an earnings call, investor filing, or press statement. What would drop us below 60%: if the Q2 and Q3 earnings cycle passes without any major employer using that language despite visible displacement continuing, we'd interpret that as evidence the attribution-avoidance norm is more durable than we've modeled. We're watching Q3 earnings calls closely — that's the first real test of whether corporate disclosure behavior is shifting.