Generic AI gives you a starting point. Legal and compliance work needs a finish line.

01

The finish line problem

Adoption is no longer the debate. By 2026, generative AI use across professional services had nearly doubled year-on-year — from 22% to 40% of organisations — with corporate legal departments alone jumping from 23% to 47% adoption in a single year (Thomson Reuters 2026 AI in Professional Services Report). Among lawyers already using AI tools, 77% use it for document review, 74% for legal research, and 74% to summarise documents (Thomson Reuters).

That is the starting point, and it is genuinely useful. But three things sit right next to that adoption curve that every APAC legal and compliance leader should read twice. Despite record adoption, 48% of general counsel say AI's impact on their organisation "hasn't fully materialised yet," and only 26% strongly agree that AI-integration success metrics are even well defined (Diligent GC Risk Index 2026; KPMG). In June 2026, the 9th U.S. Circuit Court of Appeals sanctioned two lawyers US$2,500 each and suspended their right to practise before the court for six months after briefs containing AI-"hallucinated," nonexistent case citations — its first ruling on AI accountability. The court was blunt: "attorneys do not need cutting-edge technology to fabricate citations" (Reuters). And two-thirds of general counsel report spending more time on enterprise-wide risk and compliance than a year ago, while 65% agree AI's output-accuracy questions raise real concerns about its appropriateness for legal work (Diligent; KPMG).

A finish line is the point at which a piece of work becomes accountable — filed with a regulator, signed by a director, relied upon in court, or acted on by a customer. Generic AI can get any legal or compliance workflow to perhaps 70–80% complete. It cannot:

• Verify against the specific statute, notice or code in force in the specific APAC jurisdiction on the specific date of use.
• Take personal or corporate accountability for the output — no AI model can be named as the filer, signatory, or reporting officer.
• Exercise privilege, discretion and escalation judgment — knowing when a matter must go to outside counsel, the board, or a regulator, rather than being handled at desk level.
• Hold the relationship — with MAS, the PDPC, SIAC, HKIAC, a regulator, a counterparty's counsel, or a whistleblower who needs to trust a real person.

The industry has enthusiastically adopted a tool that produces a first draft faster than ever, while quietly admitting it still needs a trained person to cross the finish line.

02

Financial crime, AML & sanctions

1. 1

   AML/CFT customer due diligence and ongoing monitoring

   AI-assisted screening can triage thousands of customer records and flag anomalies in transaction patterns. The finish line is the escalation decision: under MAS's AML/CFT framework, a financial institution that fails to properly identify, verify or report is liable for fines of up to S$1 million per offence — a liability that sits with a named compliance officer, not a model (MAS).

2. 2

   Sanctions and export control screening

   Generic AI can run automated classification against OFAC, BIS Entity List and equivalent APAC watchlists. The finish line is the licensing and end-use judgment call — assessing intermediaries, re-export risk and dual-use classification, which requires trade-compliance expertise, not pattern matching (U.S. Department of Commerce; IBA).

3. 3

   Suspicious transaction reporting (STR) and regulator escalation

   AI can draft the narrative from flagged data. The finish line is the human MLRO decision to file — a statutory obligation that carries personal liability and cannot be delegated to a tool.

03

Data privacy & cyber

1. 4

   Data Protection Impact Assessments (DPIAs)

   AI can populate risk registers against the PDPA's 11 core obligations. The finish line is the proportionality judgment — whether a given data flow's risk is "acceptable" given business context, which the PDPC's own guidance frames as a matter of organisational accountability, not a checklist score (PDPC).

2. 5

   Cross-border data breach notification

   Notification triggers vary sharply across APAC — Indonesia requires notice within 72 hours, Singapore requires a "serious harm" assessment within 30 days, and Australia, India, Japan and the Philippines each set different thresholds. AI can draft the incident timeline; only a lawyer who knows which regime governs a multi-country breach can decide where and when to notify (Bird & Bird APAC Privacy Law Review; Global Law Experts).

3. 6

   Third-party and vendor data-risk due diligence

   AI can summarise a vendor's security questionnaire in seconds. The finish line is the negotiated allocation of liability in the contract — the clause that determines who actually pays when the vendor's system fails.

04

Governance, board & company secretarial

1. 7

   Board and GRC reporting

   AI-generated dashboards are now common, yet 79% of general counsel still lack confidence that their board reporting strikes the right balance between clarity and overload, and only 19% report fully integrated GRC systems (Diligent GC Risk Index 2026). The finish line is a general counsel who can translate the dashboard into a board-level risk narrative and recommendation.

2. 8

   Corporate governance code disclosure

   Singapore-listed companies must disclose, and explain any departure from, the Code of Corporate Governance (MAS). AI can draft the "comply" language; only a company secretary or governance counsel can write a defensible "explain."

3. 9

   Cross-border entity management and regulatory filings

   Annual filings, beneficial-ownership registers and local statutory returns differ by country and change constantly across APAC. AI can track deadlines; a licensed company secretary or corporate services professional has to actually lodge, attest and remain liable for the filing.

05

Disputes & investigations

1. 10

   Litigation document review and e-discovery

   AI-assisted review is now one of the most mature legal AI use cases, cutting document-review time meaningfully (Thomson Reuters). The finish line is privilege review and the final relevance call — both of which carry professional and ethical liability if wrong, as recent AI-hallucination sanctions have made painfully clear (Reuters).

2. 11

   Arbitration case management (SIAC / HKIAC)

   Platforms like HKIAC's Case Connect digitise procedural steps, and the SIAC Rules 2025 introduced a new Expedited Procedure for low-value, low-complexity disputes (Mayer Brown; HKIAC Case Connect). The finish line is advocacy and tribunal strategy — no software argues a case.

3. 12

   Whistleblower intake and internal investigation

   AI can triage and categorise incoming reports within the recommended 24–48 hour ownership window. The finish line is the privilege determination at intake and the credibility assessment during interviews — decisions that determine whether a company can later claim cooperation credit with a regulator.

06

Corporate, commercial & IP

1. 13

   Contract lifecycle management (CLM)

   AI now handles clause extraction, metadata tagging and version control at scale. The finish line is risk-weighted negotiation — deciding which clause to hold the line on, and with which counterparty, which is a commercial judgment call no template library makes for you.

2. 14

   M&A legal due diligence

   AI can flag change-of-control clauses and red-flag issues across a data room in hours instead of weeks. The finish line is materiality judgment — deciding which flagged issue actually changes valuation or deal structure, which requires deal experience, not document recall.

3. 15

   IP portfolio and filing management

   AI can monitor renewal deadlines and search prior art across jurisdictions. The finish line is the strategic filing decision — where to file, when to enforce, and when a mark or patent is worth defending in a specific APAC market.

4. 16

   Competition and antitrust merger-control filings

   AI can pull market-share data points quickly. The finish line is the substantive competition analysis and the regulator negotiation — filings that, if wrong, can unwind a transaction.

07

People, ESG & regulatory change

1. 17

   Multi-jurisdiction employment law compliance

   Employers operating across APAC must navigate divergent — and rapidly evolving — rules on everything from termination to mandatory workplace harassment prevention in markets including Australia, China, Hong Kong, Japan and Taiwan (HSF Kramer; The Employer Report). AI can draft a policy template; only local employment counsel can confirm it survives contact with the actual regulator.

2. 18

   Anti-bribery and corruption (ABC) programme management

   Enforcement intensity varies sharply across APAC — Singapore, New Zealand and Australia have markedly stronger governance infrastructure than several regional peers, which changes how a multinational should calibrate its third-party risk programme country by country (BDO). AI can screen a distributor list; a compliance officer with regional enforcement knowledge decides where to actually walk away from the relationship.

3. 19

   ESG and sustainability regulatory reporting

   APAC's ESG reporting landscape is moving fast — Japan's SSBJ standards become mandatory for large Prime Market-listed companies from fiscal years starting April 2026, and China's mandatory reporting population is expanding by index membership (Socious; ESG Pedia). AI can populate disclosure templates; a sustainability or ESG counsel has to defend the assurance-level judgment behind the numbers.

4. 20

   Regulatory horizon-scanning and change management

   AI can monitor and summarise new notices, guidelines and consultation papers across regulators the moment they publish. The finish line is deciding what actually changes internal policy, training and sign-off authority — and communicating that change to a board that already cites "volume and complexity of regulations" as the single biggest driver of role complexity for 39% of general counsel (KPMG 2026 Global General Counsel Outlook).

08

The pattern across all 20

Look across every one of these workflows and the same shape repeats: AI compresses the research, drafting, screening and summarising layer — the 70–80% that used to eat a junior associate's week. What's left, every time, is the 20–30% that carries legal liability, regulatory accountability, or a relationship that has to be trusted by a human on the other side. That remaining layer is not a smaller problem than it used to be. If anything, KPMG's data shows it is a more concentrated one: 65% of general counsel now say output-accuracy questions raise real concerns about AI's appropriateness for legal work at all (KPMG).

We are not in the business of placing people who can prompt a model — every general counsel's team can already do that. We are in the business of placing the people who can take a model's 70% and turn it into the 100% a regulator, a board, or a court will actually accept.— Rita Chou, Founder, Arctiva Partners