AI Controls Become Practice Rules, Firmwide Adoption Accelerates, and Taiwan Advances AI Governance

By DripPublished

The gist

Legal teams are moving from AI experimentation to enforceable workflow, verification, and governance duties that reshape daily judgment, supervision, and accountability.

This week’s developments

California and the IRS Turn AI Controls into Practice Rules

California’s SB 574 would make AI use a controlled workflow requirement: lawyers and arbitrators could not place confidential, personally identifying, or other nonpublic client data into public generative AI tools, must take reasonable steps to verify outputs and remove hallucinated citations or harmful content, and remain responsible for AI-assisted work as if they produced it themselves. The enforcement signal is already real. In State v. Coleman, 2026-Ohio-965, an Ohio attorney who filed an AI-drafted brief with fabricated facts and citations was sanctioned $2,000 and referred to disciplinary counsel, showing courts are treating failed AI review as a pleading and conduct problem, not a tech glitch.

That pressure is now spreading beyond general ethics guidance. ABA Formal Opinion 512 already tied generative AI to competence, confidentiality, candor, and supervision duties, and the IRS has now extended that logic to tax advisors under Circular 230, allowing drafting support but not blind reliance on AI for authority or calculations. The market response is a shift toward auditable platforms that preserve traceability.

For lawyers, the career edge is moving to those who can document review, protect client data, and build defensible sign-off processes. AI is no longer just a productivity boost; it is becoming a supervised production system.

How should we update AI review workflows and accountability now?

If you're an individual contributor

  • AI speed won't save you; your review discipline will.
  • Learn to verify AI output, strip bad citations, and protect client data—your value is now in defensible judgment.

Sources

If you manage a team

Sources

If you lead the organization

Sources

Firmwide Adoption and Custom Builds Signal the Next AI Arms Race

Barnes & Thornburg reported nearly 90% attorney usage, 1,000+ users, and more than 150,000 prompts across Harvey, CoCounsel, and ChatBT in 30 days, while BDO Abogados said its eight-month Legora rollout reached 90%+ adoption and Hanson Bridgett adopted Anthropic’s Claude firmwide for 200+ attorneys and staff. Those disclosures show firms are standardizing on platforms that combine matter context, research, and authority checks in one interface, not just tools that draft faster.

The adoption is also moving into specific workflows: Kirkland & Ellis is planning a $500 million tailored AI build, NKF moved to a firmwide Legora partnership after internal evaluation, and BKL is using Harvey for cross-border M&A, international disputes, and multi-jurisdictional regulatory work with Korean-language support. The competitive edge is shifting from isolated drafting assistance to control of the full matter environment plus verified legal authority inside the same workspace.

For practitioners, this is the next step beyond the workflow and supervision layers already taking hold: less time stitching together research, drafts, and matter data across systems, and more responsibility for supervising matter-aware AI, checking jurisdiction-specific authority, and becoming fluent in the platform your firm standardizes on.

How should firms govern AI adoption across all seniority levels?

If you're an individual contributor

  • Your edge is shifting from drafting fast to supervising AI well.
  • Learn your firm’s platform deeply and get sharp at checking authority, jurisdiction, and matter context—those errors will define your value.

Sources

If you manage a team

  • Your team’s leverage now comes from AI fluency, not just output speed.
  • Coach lawyers on review discipline and workflow judgment, and reallocate time toward exception handling, not repetitive drafting.

Sources

If you lead the organization

  • The AI race is now about firmwide control, not isolated tools.
  • Invest in one governed platform, custom matter workflows, and talent who can run AI-enabled legal work across jurisdictions.

Sources

Taiwan Sets the Next Layer of AI Governance

Taiwan’s Artificial Intelligence Basic Act, promulgated on 14 January 2026, is now pushing the next layer of AI governance into place. The Act makes the National Science and Technology Council the central AI policy authority, creates a National AI Strategy/Governance Committee under the Executive Yuan chaired by the premier, and requires agencies to implement AI risk assessments, internal control and usage guidelines, labor-right safeguards, and privacy-by-design/data governance measures by July 2026. Sector regulators then have until 14 January 2028 to issue AI risk-management rules and fill gaps through delegated rules.

The enforcement burden is still being built. The Basic Act is principles-based, with no clear penalty regime or private right of action, so the real compliance load will come through secondary rules, audits, and sector-specific implementation. That makes Taiwan closer to Japan and Singapore than the EU AI Act’s harmonized, risk-tiered model, even as China tightens controls and U.S. state laws keep fragmenting the baseline.

For Legal teams, this is the next step after inventories and incident workflows: keeping live obligation maps current across regimes that diverge in timing, structure, and enforcement. The value now sits in regulatory intelligence, local ownership, and fast evidence retrieval when secondary rules move.

How should we adapt our AI governance operating model now?

If you're an individual contributor

  • Taiwan’s AI rules will reward lawyers who can track fast-moving gaps.
  • Keep your obligation map live and learn to pull evidence fast; that’s how you stay useful as secondary rules land.

Sources

If you manage a team

  • Your team’s edge shifts from drafting to monitoring fragmented AI rules.
  • Coach for regulatory intel, local ownership, and rapid issue-spotting; inventories alone won’t carry the workload.

Sources

If you lead the organization

  • Your operating model must absorb AI governance before sector rules hit.
  • Invest in a central obligation-tracking spine and local execution now, or you’ll pay later in audit friction and missed gaps.

Sources

Part of these trends

Stay ahead in Legal

Get the weekly Legal brief in your inbox — the developments, what they mean by seniority, and what to do next.