Part-145 AI Prepares Work, It Does Not Release

Part-145 maintenance is not short of paperwork. It is short of time and easy proof that work stayed inside the approved system.

The aircraft is released only when authorised people have completed, recorded and certified the right work. The MOE sets the route. Certifying staff work within their authorisation limits. Tooling has to be calibrated. Training and competence evidence must be current. Findings need containment, corrective action and closure evidence that can stand up in an audit.

AI has a useful place in that system, but it is a working place, not an authority place. It does not certify maintenance, issue a certificate of release to service, approve MOE changes or procedures, grant staff authorisations, close findings, replace certifying staff or become the official record. Everything below sits on that side of the line. Where a draft stops and a person decides is in assistance versus decision-making.

Reduce friction, don't build a second system

A Part-145 organisation already has enough moving parts: aircraft availability, defect rectification, shift handover, parts shortages, hangar space, tooling, customer requirements, subcontractors, stores and records.

The wrong AI workflow adds another uncontrolled layer. People paste work packs into a private tool, ask it to interpret maintenance data, then copy the answer into the real system. That is not support. That is a bypass.

The useful model is smaller and role-shaped. A quality manager uses AI to shape an audit preparation pack. A supervisor turns rough handover notes into a cleaner draft for review. A training coordinator prepares a competence checklist. A procedure owner converts a change request into an MOE amendment draft with open questions clearly marked.

None of that moves accountability away from the exposition, approved procedures, nominated roles, certifying staff authorisations or official records. It gives people cleaner material before judgement.

MOE and procedure change packages

Part-145 procedures are control documents. They describe how the organisation manages maintenance, records, inspections, shift handover, subcontracted work, tooling, defects, stores, authorisations and release to service. (The MOE guide covers what the exposition has to contain.)

Procedure changes rarely start tidy. A supervisor reports that night-shift handover is inconsistent. A quality audit finds tool quarantine wording unclear. A stores process changes after a customer requirement. The first note is never ready for the exposition.

AI can turn that rough input into a reviewer pack that shows:

• the reason for the change;
• the affected MOE section or local procedure;
• roles affected in maintenance, quality, planning, stores or tooling;
• forms, registers or records that need update;
• training or familiarisation needed before the effective date;
• handover or escalation points;
• conflicts with existing wording;
• questions the human owner must answer.

This is where AI earns its keep. It can spot weak wording such as "engineering will check" when no role is named. It can flag a missing record reference, a procedure with no owner, or a change that needs a briefing before it takes effect.

The controlled review and approval process still decides what goes into the MOE. A draft in an AI workspace is not a manual amendment. It is preparation for the person who owns the document.

Certifying staff and the CRS boundary

Certifying staff carry real legal and professional responsibility. They need accurate information, clear records and enough time to make a sound judgement within their authorisation. They do not need software telling them to sign a certificate of release to service.

Before a work pack review, AI can help build a non-decision checklist:

• task reference identified;
• required signatures visible;
• independent inspection evidence present where required;
• deferred items recorded and controlled;
• parts, batch or serial details traceable;
• tooling or test equipment evidence referenced;
• open defects separated from closed work;
• CRS or release statement left only for authorised staff.

The checklist certifies nothing. It helps the certifying staff member or reviewer see what still needs attention.

The release boundary is the hard line. AI must not release an aircraft, tell certifying staff to release one, or stand in for required inspections, signatures, maintenance data, work pack review or defect control. The wording has to show it: "check whether evidence is present," not "this aircraft is ready." "Reviewer to confirm," not "approved." Open questions stay visible, not buried. That language protects accountability, and it makes the output more useful because it fits the real maintenance system. The official work pack and maintenance record remain in the approved system.

Tooling and calibration control

Tooling is a classic audit tripwire because the evidence is plain. A tool is in date or it is not. Calibration evidence is available or it is not. An unserviceable tool is quarantined or it is not.

The texture is familiar to anyone who has worked near a hangar: a torque wrench loaned between bays, a local test box with a calibration sticker nobody can read, a special tool returned without status, a certificate stored somewhere only one person knows.

AI can prepare a tooling review by tool ID, due date, location, owner, status, task linkage, quarantine action and evidence reference. It can draft questions for the tooling coordinator, compare a local checklist against procedure wording and highlight unclear responsibilities: who quarantines a suspect tool, who assesses out-of-tolerance impact, where the calibration certificates are stored.

The serviceability status belongs in the approved tooling or asset system. AI should not become the register, decide that a tool is acceptable for use, or clear the impact of an expired or out-of-tolerance tool. The organisation records the result. What 145.A.40 and its AMC require, with a register template and what to do when a tool is overdue, is in the tool control guide.

Training, competence and authorisation

Part-145 compliance depends on competence. Human factors, EWIS, FTS, continuation training, type training, procedure briefings and authorisation evidence all need control. Training records are easy to ignore until an audit asks who completed continuation training, who was familiarised with a procedure change, and whether the briefing happened before the effective date.

AI can support the preparation side. It can draft continuation training summaries, build role-based checklist templates, and turn an MOE change into a short familiarisation note for the training owner to review before the change takes effect. It can help a manager prepare questions for an authorisation review, such as whether recent experience, training evidence and scope limits have been checked. The certifying staff guide covers the personnel side, and AI for training compliance goes further into matrices and familiarisation.

The boundary is firm. Company training records, staff personal data, authorisation files and official competence evidence stay in company systems. A personal AI layer should not become a hidden staff-record database or a second version of the competence file. AI does not grant privileges, decide that a person is competent or replace the authorisation process. It helps people arrive better prepared for it.

Hangar audits and objective evidence

Part-145 audits should test whether control works in the hangar, stores, planning office and records process, not stop at manual text. An auditor may walk the hangar, sample work orders, check staff authorisations, look at tool calibration labels, review stores quarantine, ask about shift handover and compare the MOE with what happens on the floor.

AI can turn the scope into a preparation list. For a hangar audit, that might cover work orders and CRS records, component release documents, calibration due dates, certifying staff authorisation samples, continuation training records, shift handover notes, subcontractor controls, stores control, defect rectification and previous finding closure evidence.

It can also prepare interview prompts for supervisors, planners, stores staff and certifying staff. Good prompts are specific. "Show how an expired tool is quarantined" beats "explain tooling control." "Show the handover record for this aircraft" beats "describe shift handover."

For findings, it keeps the wording fair and traceable:

• requirement or procedure reference;
• sample checked;
• objective evidence;
• observed gap;
• impact question;
• owner for response;
• CAP evidence expected.

It should not decide the root cause, classify the finding, accept the CAP or verify closure. The auditor decides whether a finding exists and how it is classified, and the compliance monitoring process controls issue, response, review and closure. What 145.A.95 asks for after a finding, with a worked root cause analysis, is in Root Cause Analysis and CAPs for EASA Audit Findings.

Keep the labels honest

The main risk is not only that AI can be wrong. It can make unfinished work look finished.

Part-145 teams need blunt labels: draft, reviewer note, evidence request, interview prompt, official record, approved procedure, CRS. Each one means something different.

AI output should show what source was used, what was not checked, what assumption was made, what evidence is missing, who must review it and which process controls final approval. A transparent draft is easier to challenge than a polished paragraph with no trail.

Confidentiality and system boundaries

Maintenance records are not training examples. Work packs, customer aircraft data, defect details, component histories, safety reports, staff files, tooling records, controlled procedures, commercial terms, proprietary maintenance data and release evidence belong in approved company systems. They should not be copied into a personal AI workspace unless the organisation has approved and controlled that use. If the organisation wants AI inside official workflows, it needs approval, access control, data governance and integration with the systems that already hold the record.

A personal layer can still be valuable on safe ground: public regulatory notes, generic procedure templates, personal study material, certificates, non-confidential checklists and reusable structures.

The human review gate

The best Part-145 AI workflow ends with a human review gate. Before an AI-supported output is used, someone competent checks source, revision, applicability, evidence location, confidentiality, assumptions, procedure alignment and the accountable decision owner.

That gate is not bureaucracy. It is the control that lets AI help without weakening the system. AI can prepare better drafts, cleaner checklists and stronger audit material, and it can make gaps visible earlier.

It cannot carry the certificate, the authorisation or the release decision. That is the right balance.