A Chief Regulatory Officer asks a deceptively simple question in a portfolio review: "Should we tell the FDA we used AI to draft the Module 2.5, or should we just file and answer if they ask?" The room divides instantly. One camp wants to say nothing, on the theory that the AI is just a tool like a word processor and disclosure invites scrutiny. The other camp wants a full appendix describing every prompt, on the theory that transparency is always safer. Both are wrong, and the reason they are wrong is the entire substance of regulatory strategy for AI. Disclosure is not a binary and it is not a confession; it is a calibrated act of engagement that happens through specific channels, at specific moments in the review cycle, in specific language, and the AI Function Strategist who treats it as a strategy rather than a reflex is the one who keeps the function ahead of the regulators instead of reacting to them. This lesson maps the engagement architecture across FDA, EMA, PMDA, MHRA, Health Canada, NMPA, and TGA, and it teaches the named-window discipline that determines when and how AI involvement is raised.

Why Engagement Precedes Disclosure

The first strategic error is to treat the cover letter as the only place AI is discussed with a regulator, because by the time you are writing the cover letter the strategy is already fixed and most of your options are gone. Regulatory engagement on AI is a sequence that begins long before submission, in the pre-submission meetings where a sponsor and an agency align on the development program, and the strategist who raises AI use in those forums shapes the regulator's expectations rather than reacting to them at filing. The FDA-EMA Guiding Principles released on 14 January 2026 establish transparency and stakeholder collaboration as principles, which means the regulators have signaled they want to be engaged on AI use, and a sponsor who engages early is meeting that signal rather than triggering suspicion. Engagement is the relationship; disclosure is one event inside it. A strategist who builds the relationship first finds that the disclosure, when it comes, lands on prepared ground.

The second strategic error is to imagine that AI engagement is a single conversation when it is actually a portfolio of channels with different purposes, audiences, and timing. The FDA Pre-IND and Type B and Type C meetings, EMA scientific advice and the Innovation Task Force, the MHRA AI Airlock, the PMDA AI Working Group, and the parallel mechanisms at Health Canada, NMPA, and TGA are each a distinct door, and the strategist chooses which door to open for which question. A question about whether an AI-assisted drafting workflow is acceptable in principle belongs in an early scientific-advice or innovation forum; a question about a specific submission's AI disclosure belongs in the meeting attached to that submission. Using the wrong channel wastes the engagement and can confuse the regulator about what is being asked, so channel selection is itself a strategic decision the function must make deliberately.

The FDA Engagement Architecture for AI

The FDA gives a sponsor several timed engagement points, and each one is a window the strategist can use to raise AI involvement before it becomes a review question. The Pre-IND meeting, with a Written Response Only path that goes out within 30 days of the request, is the earliest formal moment to signal that the program will use AI in specified ways, such as AI-assisted toxicology report drafting or AI-assisted protocol development, and to ask whether the agency has concerns. Type B meetings, scheduled within 60 days of the request and including the End-of-Phase-2 meeting, are the natural place to discuss how AI will be used in the registration-enabling work, and Type C meetings, scheduled within 75 days, are the flexible channel for a specific question such as how to disclose AI involvement in the coming submission. The strategist maps the AI engagement to the meeting that already exists in the program rather than creating a special-purpose interaction, because folding AI into a planned meeting normalizes it as one development consideration among many.

Once the application is under review, the FDA engagement shifts to the review-cycle windows, and here the named-window discipline becomes operational. The Day 74 communication is the point at which the FDA conveys the filing review status and any early issues, and a sponsor who disclosed AI use cleanly in the cover letter has positioned that disclosure to be assessed within the normal filing review rather than surfacing later as a surprise. Through the review cycle the FDA may issue Information Requests, and an AI-related question, such as a request to explain how an AI-drafted section was verified, arrives as an IR that the strategist must be ready to answer with the audit trail and verification record already assembled. The PDUFA goal date is the fixed endpoint the entire cycle runs toward, and the strategic point is that every AI engagement decision is made with that date in view: a disclosure or a question that triggers a late-cycle review issue can threaten the goal date, which is why the engagement is front-loaded into the pre-submission and filing windows rather than left to emerge under the clock.

The EMA Engagement Architecture and the Day 120 / Day 180 Windows

The European architecture runs on different mechanisms and different named windows, and the strategist must hold both the FDA and the EMA maps simultaneously because most significant programs engage both. EMA scientific advice, obtained through the Committee for Medicinal Products for Human Use and its Scientific Advice Working Party, is the formal channel for asking whether a development approach, including an AI-assisted one, is acceptable, and the EMA Innovation Task Force is the early, informal channel for novel methodologies that do not yet fit a standard advice question. The EMA AI Workplan is the multi-year vehicle through which European AI expectations are being developed, and a strategist tracking it can anticipate where European thinking is heading and engage proactively through the established advice channels rather than waiting for finalized guidance. Engaging EMA early through scientific advice on an AI-assisted approach converts a potential review-cycle objection into a pre-agreed position.

Inside the centralized review the named windows are Day 120 and Day 180, and they are the EMA equivalents of the FDA review-cycle communication points that the strategist must plan around. The Day 120 list of questions is the consolidated set of issues the rapporteur and co-rapporteur raise after their initial assessment, and an AI-related question, such as how an AI tool contributing to the dossier was validated, can appear here if the sponsor did not pre-empt it through engagement or clean disclosure. The Day 180 list of outstanding issues is the narrower set that remains after the sponsor's responses to the Day 120 questions, and an AI issue that reaches Day 180 unresolved becomes a gating item the sponsor must close before the procedure can conclude favorably. The discipline is therefore to engage EMA early enough, and disclose cleanly enough, that AI never first appears as a Day 120 question, because an issue raised by the regulator carries more weight and more risk than the same matter raised proactively by the sponsor.

The MHRA, PMDA, and the Other Agency Channels

Beyond FDA and EMA, the strategist maintains a working knowledge of the channels each major agency offers for AI engagement, because a global program files in multiple jurisdictions and the AI strategy must be coherent across them. The MHRA in the United Kingdom operates the AI Airlock, a regulatory sandbox that allows sponsors and developers to test AI approaches against regulatory expectations in a controlled setting, and it is the most explicitly AI-focused engagement channel any major agency offers, which makes it valuable both for the specific question and for the signal it sends about MHRA's direction. The PMDA in Japan runs an AI Working Group developing the Japanese regulatory position on AI, and engaging it positions a sponsor ahead of formalized Japanese expectations. Health Canada, the NMPA in China, and the TGA in Australia each have evolving positions, and while their AI-specific channels are less mature, the strategist still raises AI use through the standard pre-submission and scientific-advice mechanisms those agencies offer.

The strategic principle that unifies the multi-agency picture is coherence: a sponsor cannot tell the FDA one story about its AI use and the EMA a different one, because the regulators communicate and a transparency principle that both FDA and EMA have adopted makes inconsistency a credibility problem. The function must therefore hold a single, defensible account of how it uses AI, validated and audit-trailed, that it can present to any agency through that agency's channel, adjusting the framing and the depth to the channel but never the underlying facts. This is why the engagement architecture and the validation and governance architecture are the same architecture viewed from different angles: the controls the function built for inspection readiness are the same controls it describes to a regulator in engagement, and a function that has one cannot improvise the other. The strategist who has built the controls can engage any agency with confidence because the answer is always the same true account, told in that agency's idiom.

When and How to Disclose AI Involvement

The disclosure decision turns on materiality and on the principle of transparency, not on a blanket rule, and the strategist's judgment is what distinguishes useful disclosure from noise. AI used as a general drafting aid, with every output verified by a named author against source, is at one end of the spectrum, and disclosing it serves the transparency principle and is increasingly expected. AI used in a way that materially shapes a conclusion, an analysis, or the evidence itself sits at the other end, and here disclosure is not optional because the regulator's ability to assess the submission depends on understanding how the evidence was produced. The strategic test is whether a reasonable reviewer, knowing how the AI was used, would want that information to assess the submission, and where the answer is yes, the disclosure is made proactively rather than withheld and discovered. Withheld-and-discovered is the worst outcome, because it converts a tool-use question into a credibility and integrity question.

The how of disclosure is as strategic as the whether, and the wording that works is specific, bounded, and paired with the controls. Effective disclosure language names what the AI did, names the human accountability that governed it, and references the validation and audit-trail controls that make the output defensible, in a few precise sentences rather than an exhaustive appendix of prompts that the reviewer neither wants nor can use. The goal is to give the reviewer exactly what they need to be satisfied that the AI was used in a controlled, accountable, verified way, and no more, because over-disclosure invites questions that under-disclosure would have avoided while adding no assurance. The strategist calibrates the disclosure to the materiality, places it in the cover letter and Module 1 as emerging practice now directs, and ensures the controls referenced in the disclosure actually exist and are retrievable, because a disclosure that points to a validation record the function cannot produce is worse than no disclosure at all.

The Named-Window Discipline as the Backbone of Engagement

Everything in this lesson resolves into a single operational discipline: the strategist runs AI engagement against the named regulatory windows, because the windows determine when an AI matter can be raised cleanly and when it becomes a risk. The pre-submission windows, the FDA Pre-IND Written Response Only within 30 days, the Type B within 60 days, the Type C within 75 days, and the EMA scientific advice and Innovation Task Force timelines, are when AI use is shaped as a development consideration before any review pressure exists. The filing windows, the FDA Day 74 communication and the clean cover-letter disclosure, are when AI involvement is presented for assessment within the normal filing review. The review windows, the EMA Day 120 list of questions and Day 180 list of outstanding issues, and the FDA Information Requests across the cycle toward the PDUFA goal date, are where AI either has been pre-empted by good engagement or surfaces as a regulator-raised issue, and the entire strategy is built to ensure the former.

The discipline is powerful because it makes AI engagement legible and plannable rather than reactive and anxious. A strategist who has mapped each AI engagement question to the window where it belongs knows, for any program, when the AI conversation happens and through which channel, and can sequence the validation, governance, and disclosure work to be ready at each window. The opposite, raising AI for the first time when a regulator asks at Day 120 or in a late-cycle IR, cedes the initiative to the agency and forces the function to respond under the clock with whatever it has assembled, which is precisely the position the named-window discipline exists to prevent. The strategist's mastery shows in a program where AI was raised in the Pre-IND meeting, refined in scientific advice, disclosed cleanly in the cover letter, and never appeared as a Day 120 question, because by then it had been settled in every window where it could have been raised. That sequencing is the difference between leading the regulatory relationship and being led by it.

Building the Engagement Playbook for Your Function

The deliverable that institutionalizes this strategy is an AI regulatory engagement playbook, a standing document the function maintains so that engagement is a repeatable process rather than a per-program improvisation. The playbook maps each agency to its AI engagement channels and the timing of each, names the materiality test that decides whether and what to disclose, holds the approved disclosure language templates calibrated to use-materiality levels, and links each engagement point to the validation and audit-trail evidence that must be ready to support it. With the playbook in hand, a program team preparing a Pre-IND meeting knows to consider raising AI use, knows what to say, and knows which controls to have ready, and the function presents a coherent AI story across FDA, EMA, and every other agency it files with. The playbook is the engagement counterpart to the inspection-readiness binder: one prepares the function to be asked about AI, the other prepares it to raise AI first.

The playbook must also evolve, because the regulatory landscape on AI is moving faster than any single program, and a static playbook becomes wrong quietly. The FDA's May 2025 draft considerations on AI to support regulatory decision-making remain in the comment-response window and will finalize, the EMA AI Workplan will produce new positions, and the MHRA AI Airlock and PMDA AI Working Group will publish learnings, each of which can shift what the agencies expect and therefore what the playbook should advise. The strategist assigns ownership for tracking these developments and updating the playbook, so the function's engagement strategy reflects the current regulatory direction rather than last year's, and so the disclosure language the function uses this quarter is the language that will hold up at the next meeting rather than the language that would have held up at the last one. A function whose engagement playbook tracks the regulators is a function that engages from a position of currency, which is the only position from which proactive engagement is credible.

Key Takeaways

• Disclosure is one event inside the larger relationship of engagement, and engagement begins long before the cover letter. Raise AI use in the pre-submission meetings where the agency's expectations are still being shaped, so that when disclosure comes it lands on prepared ground rather than triggering suspicion.
• Map each AI engagement question to the right channel and the right named window. FDA Pre-IND Written Response Only within 30 days, Type B within 60 days, Type C within 75 days; EMA scientific advice and the Innovation Task Force; the MHRA AI Airlock and PMDA AI Working Group, each is a distinct door for a distinct question.
• Run engagement against the review-cycle windows so AI is settled before it can surface as a regulator-raised issue. The EMA Day 120 list of questions and Day 180 list of outstanding issues, and the FDA Day 74 communication and Information Requests toward the PDUFA goal date, are where good engagement pays off or absence of it costs.
• Disclose on materiality, not by blanket rule: if a reasonable reviewer would want to know how the AI was used to assess the submission, disclose proactively. Effective wording names what the AI did, names the human accountability, and references the validation and audit-trail controls, in a few precise sentences, never an exhaustive appendix of prompts.
• Hold a single coherent AI account across all agencies and institutionalize it in an engagement playbook that tracks the moving landscape. The FDA-EMA transparency principle makes inconsistency a credibility problem, and the FDA May 2025 draft, the EMA AI Workplan, and the MHRA and PMDA channels keep shifting expectations the playbook must follow.