Five themes reshaping the legal function by 2030 according to GC research

You must now demonstrate which AI systems your legal workflows use, what controls surrounded each one, and what evidence trail exists to reconstruct the decisions made. This is no longer optional infrastructure — it is the baseline your general counsel function must be able to show.

An analysis of 1 September 2026 of five structural themes reshaping the legal function by 2030 argues that managing AI and data risk has become a core task of the general counsel, and that governance must shift from stand-alone tools to entire workflows with built-in verification. The concrete case is the gap between legal departments' risk instinct and their actual governance infrastructure: teams rely on policy documents and isolated tools, but cannot reconstruct which systems were used, which controls applied, or what evidence exists. In our assessment, this means that the transition from logging obligation to reconstruction obligation is already underway. You cannot meet the emerging standard by policy alone — you must be able to demonstrate practice.

What does "reconstruction obligation" mean for your workflows?

The shift is from knowing that you have controls to proving what controls were applied to a specific task. When a legal team uses AI for research, due diligence, contract analysis or any other workflow, the question is no longer whether the team has a policy on AI use. It is whether you can show, for each piece of work, which AI system was involved, what verification steps were taken, where disagreement or correction occurred, and what evidence trail remains.

This is not the same as audit compliance. Audit asks whether you have a framework. Reconstruction asks whether you can walk through a specific decision and show the steps. If a lawyer used an AI model to research case law, you must be able to show which model, what the model returned, how the lawyer verified it, and what the final output was. That trail becomes part of the work product itself.

Which failure modes does this address?

• Hallucination and inaccuracy — AI systems produce plausible but false information, and stand-alone verification tools cannot catch all instances.
• Untraced AI use — workflows deploy AI systems without documented record of which model, which version, or which data it touched.
• Governance without evidence — legal teams have policies but cannot reconstruct whether those policies were followed in practice.
• Vendor lock-in and data leakage — reliance on a single AI platform without visibility into data flows or alternative verification routes.
• Accountability gaps — when an AI-assisted output causes harm, the legal team cannot show what steps were taken to verify it.
• Hallucination detection failure — a single verification layer cannot catch all errors; layered detection across independent systems is required.

What concrete controls must you be able to demonstrate?

1. Record the AI system and its purpose — document which model each workflow uses, when it was used, and the lawful basis for any data it processed.
2. Implement citation verification for research workflows — require the AI system to show its sources, and verify those sources independently before relying on the output.
3. Route sensitive tasks through multiple independent models — for high-stakes work, send the same task to more than one AI system and compare results to surface disagreement.
4. Log corrections and disagreements — when a lawyer corrects an AI output or finds that two models disagree, record that fact as part of the evidence trail.
5. Retain the evidence trail with the work product — keep the record of which systems were used, what they returned, and what verification occurred alongside the final output.
6. Establish a verification layer outside your primary AI platform — do not rely on a single vendor's verification tool; use independent systems to check results.

How does this change your relationship with AI vendors?

Your legal team will need to orchestrate an ecosystem of vendors, AI platforms and internal IT to retain control over sensitive data. This does not mean building your own AI systems. It means that you cannot outsource the verification decision to a single platform. You need visibility into which systems are in play, which data they touch, and what alternative routes exist if one vendor fails or if you need to exit a contract.

A verification console or workflow tool can make visible which AI systems are in play per task, which verification steps were taken, and what corrections occurred. But the tool itself does not guarantee that outputs are correct, and it does not remove the risk of hallucinations. The professional final judgement remains with the lawyer. A verification layer is thus a part of a broader governance architecture, not the architecture itself.

What can tooling carry and what stays your responsibility?

Tooling can automate the recording of which systems were used, route tasks through multiple models, and surface disagreement for your review. It can log corrections and build the evidence trail. It cannot judge whether the final output is legally sound, whether it is fit for the client's purpose, or whether the risk of error has been reduced to an acceptable level. Those judgements are yours. The shift to reconstruction obligation means that you must be able to show your work — not that the work itself can be delegated to a system. The infrastructure exists to support that demonstration. Your professional responsibility is to use it.