Lawyers using AI share more than documents — reasoning traces can leak your 'alpha'
jkubicki · x · 2026-09-04
Lawyer Jack Kubicki's long post warns the legal industry that even with enterprise AI subscriptions, you share more than uploaded documents and prompts: reasoning traces record a model's intermediate steps and can reveal how your information becomes useful.
He decomposes a firm's value into four layers:
- Data: client documents, communications, code, confidential facts;
- Reasoning: connections between facts, what matters and what doesn't;
- Institutional knowledge: strategies, judgment calls, processes, workflows;
- Alpha: the advantage created when all three combine.
A great lawyer's superpower is internal pattern recognition. But when you feed models client data, show them your firm's analytical examples, correct their work, and repeat across thousands of matters, you may be sharing how you think. He stresses that retention doesn't necessarily mean training or permanent weights — the core issue is opacity: what exactly is retained?
More from Safety
- New protocol enables private, verifiable LLM inference outsourcing with 179MB local storage — chaumian · 2026-09-04
- Hundreds protest Scotland's datacentre boom, demanding pause on 20+ proposed projects — nordicinst · 2026-09-04
- Abliterated models say yes 3-4x more but miss the real FreeBSD kernel bug entirely — evilsocket · 2026-09-04
- BOSS: Breadth-Oriented Suffix Search Boosts GCG Jailbreak Success Rates — chaumian · 2026-09-04
- WeaveMark Hits 89.8% Match Rate for 32-bit LLM Watermarks, Beating BiMark — chaumian · 2026-09-04
- Researchers: Chain-of-Thought Legibility Is Too Fragile to Underpin Long-Term AI Safety — tszzl · 2026-09-04