AI and the Company Secretary: Operating the Boundary the Chair Polices

The company secretary’s role was built to maintain the conditions under which directors can apply judgement and the company can meet its governance obligations. AI has changed the conditions of execution for both. The board pack, the minutes, and the disclosure paper all reach the Board in their familiar form, and the familiarity is what makes the shift difficult to see. The summary that opens the pack was composed by a system whose framing decisions are not transparent. The minutes capture what the AI inferred from a transcript rather than what the secretary judged the legal record needed to hold. The disclosure paper has been informed by a regulatory tracker whose accuracy the secretary cannot fully interrogate. The Board reads what looks like a board pack; it is reading a representation of one. The secretary is the only person with line of sight to the difference, and increasingly even the secretary cannot fully see it.

The constitutional foundation of the role of the secretary has not moved. Cadbury named the company secretary in 1992 as central to effective board functioning, with directors entitled to the secretary’s advice and the chair to the secretary’s active support. The Companies Act 2006 made the role a statutory requirement for every public company. The secretary is an officer of the company, with both appointment and removal a matter for the whole Board. The FRC’s 2024 UK Corporate Governance Code carries the obligation forward, recognising the company secretary as the officer through whom directors access governance advice and through whom the Board secures the resources it needs to function effectively. The Chartered Governance Institute qualifies the modern profession through a postgraduate qualification and a binding Code of Professional Conduct. None of those sources has been updated to prescribe how the secretary’s responsibilities are to be discharged in a world where AI has entered the workflow at every operational point. Research and policy commentary from the governance professions has identified the gap; the binding constitutional texts have not closed it.

The Great Remaking series argued that AI is remaking the essence of work rather than automating tasks within it. The secretary’s work is no exception. The composition of the Board’s information environment, the integrity of the official legal record, and the company’s compliance position on its AI deployments are all being remade through operational decisions the secretary now makes, often by default, often without the conditions of judgement the role requires. This article does not propose new secretary responsibilities. The source material gives secretaries more than enough already. What has changed are the conditions of execution. The article walks through several existing secretary responsibilities, asks how AI changes their execution, and identifies what remains constant beneath the change.

The principle the secretary operates

The principle on which everything rests was named by Cadbury in 1992 as collective responsibility in law: all directors are equally responsible in law for the Board’s actions and decisions, and the obligation to meet them rests with the Board collectively. What the Board experiences in practice is collective accountability. Nothing in the AI era weakens that principle.

Within it, agency and accountability behave differently. Agency, the doing of specific work, can be transferred. A board pack can be summarised by a system; minutes can be drafted from a transcript; a regulatory tracker can pre-flag compliance changes; a board portal can score agenda items by importance. None of those uses is illegitimate in itself. Accountability cannot be transferred at all. The secretary remains accountable for whether directors received timely, relevant information in a form fit for judgement, for whether the official legal record reflects what the Board decided, and for whether the company’s disclosures on its AI deployments are accurate and complete.

AI and the Chair framed the chair’s responsibility as policing the boundary between agency transfer and accountability transfer, and the secretary’s responsibility is to operate that boundary in practice. Every operational decision the secretary makes either preserves the conditions under which directors can apply judgement or silently erodes them. What the Board sees is what the Board can judge. What the Board can judge is what the Board becomes accountable for. The role is not a meeting administrator with constitutional trappings; it is the constitutional actor at whose desk the Board’s information environment is constructed, the official legal record is kept, and the company’s compliance position is captured.

AI in the Board: the behavioural critique

The published commentary on AI for company secretaries follows a familiar shape. Board portal vendors compete for secretary adoption with marketing that presents AI as making the corporate secretary more indispensable, the boardroom’s most trusted adviser, and the operational load lifter the role has been waiting for. Governance education providers and trade publications increasingly describe AI literacy as having shifted from nice-to-have to central. The celebratory framing is the same as the chair-level commentary the previous article in this series critiqued: tool adoption presented as governance modernisation. The same critique applies. Tool adoption is not transformation. In some forms it is the silent erosion of the secretary’s specific contribution, which is the conditions under which directors can apply judgement.

The failure modes are already visible. The first is the summary substitution failure. A secretary commissions or accepts an AI-generated summary of a 200-page board pack as the directors’ primary reading. The summary is technically accurate but selects emphasis the AI chose. Directors arrive at the meeting having read the summary, asking questions the summary’s framing invited. The secretary has formally discharged the function of ensuring directors receive timely, relevant information. The information has, however, been pre-framed by a system whose framing decisions are not transparent and whose accountability for what it omitted cannot be located. Board packs are not just management reports. They are the basis on which directors discharge their statutory duty to exercise independent judgement. A director reading a summary the AI selected is a director whose statutory judgement is being applied to material whose framing the AI shaped.

The second is the minute-fidelity failure. The secretary uses AI to draft minutes from a meeting transcript or notes, producing a record that captures what was said but not what was contested, weighed, dissented from, or tabled. The minutes that go out as the official legal record reflect the AI’s compression of the discussion rather than the secretary’s judgement about what the record needs to hold for governance purposes. Directors read the minutes and find the meeting they attended subtly different from the meeting on the page; few flag it, and the minutes stand as the record. Minutes are not notes; they are the legal artefact that auditors examine, that regulators reference, and that courts cite when a Board decision is contested. A minute partially authored by a system that cannot be held accountable for what it left out is a legal record that no one can fully defend. Automated minutes are not minutes.

The third is the agenda-curation failure. The secretary uses AI to score, prioritise, or pre-filter agenda items. The model applies patterns learned from historical agendas, optimising for what it recognises. Items it does not recognise as important, including emerging risks, weak signals, and dissenting views the chair wants surfaced, drift down the agenda or out of it entirely. The secretary has used a tool; the agenda is not the secretary’s agenda; the Board governs an agenda whose framing has silently been delegated.

The fourth is the compliance-confidence failure. The secretary uses AI to track regulatory changes, generate disclosure language, or pre-flag compliance risks. The tool produces output of confident polish. The secretary, stretched across many responsibilities, accepts the output without the depth of interrogation the manual process demanded. The company’s compliance position becomes a function of the tool’s accuracy rather than the secretary’s professional judgement. The Chartered Governance Professional’s duty to advise the Board on all governance matters has been discharged through a system the secretary cannot fully interrogate.

The pattern repeats. Agency goes to the machine, accountability stays formally with the secretary, and the thinking layer that connects the two has dissolved. The secretary appears more capable while becoming less so. The Board’s information environment appears better managed while losing the substance of the secretary’s specific contribution.

The verdict is short. An AI-generated board pack is not a board pack.

AI in the business: literacy and disclosure

Those operational failures resolve to a single underlying problem: the secretary is now accountable for systems they did not build, cannot fully interrogate, and cannot fully explain. That is the literacy problem stated directly, and it has a sharper edge for this role than for any other Board-supporting actor. The question is not whether the secretary can use ChatGPT; most can. The question is whether the secretary can stand behind the outputs of board pack summarisers, minute-drafting tools, agenda-curation engines, disclosure assistants, and compliance trackers when the Board, the auditor, the regulator, or a court asks them to. Generative AI fluency does not equip the secretary to do that, and the professional duty to advise the Board on all governance matters does not pause because the advice was machine-assisted.

Disclosure literacy comes first because the exposure is direct. The EU AI Act’s literacy provisions came into force in February 2025; high-risk system obligations apply from August 2026 for organisations operating in the EU. The FRC Code’s risk reporting requirements, listing rule disclosures on AI-related material risks, and audit committee reporting on AI controls all require the secretary to know enough about the company’s AI deployments to ensure correct, complete, and timely disclosure. Generative AI fluency does not equip a secretary to know whether the company’s customer recommendation engine is in scope for an emerging UK or EU regulation, whether a credit model qualifies as high-risk under the EU AI Act, or whether an autonomous safety system requires specific governance disclosures. The secretary is the officer through whom those disclosures are typically assembled and submitted, accountable for declarations they cannot fully examine.

The exposure does not stop at disclosure. The AI tools entering board administration require the secretary to know enough to evaluate them: whether summaries remain faithful to source documents, whether minutes capture what was contested or tabled, whether board portals store data compatibly with confidentiality and privilege requirements, whether agenda curation preserves the editorial decisions the secretary should be making. It plays out one degree removed in the vendor relationships that shape what the secretary has access to. Board portals compete commercially for secretary adoption, framing their tools as governance-ready. Generative AI fluency offers no purchase on separating governance-ready from governance-marketed.

The research suggests the literacy gap is already operational. The GC100 minute-taking poll, conducted with Norton Rose Fulbright in November 2024 across 106 companies, mostly public, found that 92% of respondents had not introduced AI to assist with minute-taking or preparation, and 84% had no internal policy on AI transcription. Among the small minority using AI to produce minutes or transcripts, review by the secretary or a member of their team was standard practice. Where that review discipline is absent, the legal record drifts away from the secretary’s judgement. PwC’s 2025 Annual Corporate Directors Survey, and the November 2025 Harvard Law School Forum piece by PwC’s DeNicola, Berlin, and Smilowitz, place the corporate secretary among the central stakeholders in board oversight of AI. McKinsey’s December 2025 board governance analysis, drawing on its 2024 Global Board Survey, found that 66% of directors report their boards have “limited to no knowledge or experience” with AI. A secretary operating AI tools sits next to a Board the majority of whose directors cannot evaluate them. Protiviti and BoardProspects’ 2026 Global Board Governance Survey found that only 26% of boards discuss AI at every meeting, which lands directly on the secretary’s agenda-construction responsibility. Where AI does not reach the agenda, the secretary’s operational decisions about AI tooling face no scrutiny from the body those decisions exist to serve.

The Six Board Concerns are the lens that prevents the secretary’s AI conversation from collapsing into the narrowest version of itself. A secretary who chooses a board pack summariser by considering only its operational impact, without examining its Ethical and Legal Responsibility, Stakeholder Confidence, or Strategic Alignment implications, has discharged a fraction of the role’s actual obligation. The chair polices the boundary; the secretary operates it; operating it requires the secretary to stand behind every AI tool they have allowed into board administration and every AI deployment they are required to disclose.

The bifurcation risk inside the secretariat

Earlier work on the AI capability bifurcation argued that workforces are splitting into premium-earning and penalty-suffering bands depending on whether they have built genuine capability or accumulated credentials. The same dynamic operates inside the secretariat.

A secretary who uses a Board portal, has run an AI awareness session for the Board, can discuss large language models confidently in conversation, and has subscribed to AI tooling can feel themselves to be modern. The credentials are real. The capability is a different matter: the ability to evaluate whether the AI tools entering board administration are preserving or eroding the conditions under which directors can apply judgement, and whether the company is meeting its AI disclosure obligations across emerging regulation. That capability may not exist. That is the secretary on the penalty side of the bifurcation, disguised as the secretary on the premium side.

A secretary who cannot evaluate the AI tools they are using is not relieved of professional responsibility for the materials those tools produce. The Chartered Governance Professional’s duty to advise the Board on all governance matters does not pause for tool adoption. Bifurcation is the explanation for why operational failure modes in the secretariat are now so often silent: the appearance of modernisation is doing the work, and the substance underneath has not been built.

The boundary in operation

The secretary’s role was built to maintain the conditions under which directors can apply judgement, to safeguard the integrity of the official legal record, and to ensure the company meets its governance obligations. Those responsibilities have not changed. What has changed are the conditions of execution.

The chair polices the boundary between agency transfer and accountability transfer. The secretary operates that boundary in practice. Every operational decision about which AI tools enter board administration is the boundary in operation. Every decision about what makes it into the official legal record is the boundary in operation. Every decision about disclosure of the company’s AI deployments under emerging regulation is the boundary in operation.

The principle that does not move is collective responsibility. When the secretary’s operational decisions silently erode the conditions under which it can be discharged, the chair has nothing left to police. The secretary is the role at which that principle is most directly translated into operational reality, and the role at which silent erosion is most likely to begin.

A Board can continue to believe it is governing through human judgement long after the conditions for that judgement have begun to erode. AI does not change the secretary’s responsibilities; it changes what their discharge requires. The constitutional principle survives. The informational conditions required to uphold it may not.