It drafts. A person decides.

The AI assistant is an optional part of AviIS. It drafts the Part-IS work that takes an information security manager hours — threat scenarios for an element or an interface, a triage suggestion for a reported event, the narrative of a report to the competent authority, a section of the ISMM, the statement of how a requirement is met — and it finds the section and page of your own manuals that apply to a finding or a change. Every draft is labelled as unverified and written nowhere until a named person accepts it. It decides nothing, and it sends nothing to an authority.

Optional add-on. €49 a month, €29 for training organisations. Included in every 60-day trial.

Suggested threat scenarios · AST-012 Electronic flight bagsSchematic

Medium × High · not acceptable · G-04 score 6
Altered performance data reaches the EFBs through a compromised software update channel
Threat source: a criminal group targeting the EFB software supplier. Consequence: wrong take-off performance figures — an unsafe condition. Annex A 5.21, 8.19.
Suggested by AI, not verifiedDetailsKeep
Low × Moderate · acceptable · G-04 score 4
A lost EFB is used to reach the company document portal
Threat source: an opportunistic finder. Consequence: disclosure of operational documents, no direct effect on flight. Annex A 5.15, 8.1.
Suggested by AI, not verifiedDetailsKeep

Schematic of the threat scenarios suggested for one element, drawn for this page rather than captured from the application: each with its rating on the organisation's matrix, the G-04 likelihood score, the reasoning, and the label it carries until someone keeps it. A kept scenario becomes a draft risk.

What it does.

Five drafting features for Part-IS work and one for your manuals — each where the work is done, none of them a chat window.

Risk register · IS.I.OR.205(c)

Threat scenarios

For an element or an interface, three to six threat scenarios in the structure GM1 IS.I.OR.205(c) describes — threat source, attack path, vulnerabilities, consequence and the safety aspect it affects — rated on your own risk matrix with the G-04 likelihood score and its reasoning, with the ISO/IEC 27001:2022 Annex A controls that would treat them. You edit them and keep the ones that apply; each becomes a draft risk associated with the element or interface, to assess and decide like any other.

Events · IS.I.OR.215(b)(1), 230(b)

A triage suggestion

For a reported event: a classification, the safety impact, whether it is reportable with the AMC1 IS.I.OR.230(a)&(b) trigger that applies and the unsafe-condition reasoning, the other recipients to consider — the design approval holder, the system designer, the Reg. (EU) No 376/2014 channel — and the questions still to answer. Apply it to the triage form, review every field, save. The event does not change until you do.

External reporting · IS.I.OR.230(c)

The narrative of a report

For the notification, the report or the follow-up to one recipient: the narrative drafted from the event, the filings already made and the actions raised — factual, without speculation — with what is still missing and, when the report goes out after 72 hours, the exceptional circumstances to state. Once you use it, it is the text of the PDF and the email draft; you edit it and you send it.

Documents · IS.I.OR.250(a)

An ISMM section

A draft of a section of the manual from your own registers — appointments by post, the policy, staffing, the scope, the reporting settings, the controlled documents — following GM1 IS.I.OR.250(a), with prompts left where the registers say nothing. The accountable manager's statement is never drafted: its words are the organisation's commitment.

Compliance · IS.I.OR.200(a)(12)

A compliance statement

For a requirement of the compliance matrix: how you comply, where your manuals address it — section and page, from the indexes of your manuals (IS.I.OR.250(d)) — the evidence an inspector would expect, and the gaps. It opens the assessment form prefilled; the status stays the assessor's to set.

Documents · findings and changes

References in your manuals

Index a manual or procedure in the document register — a PDF, or a Word, OpenDocument or RTF file converted to PDF first, up to 24 MB — and the assistant builds a section index of that revision. Writing a finding or a change, ask for references: up to three come back with the document, the section, the page and the reason, and Open page shows that page beside the record before you accept it.

What it does not do, and cannot.

The limits below are properties of how the features are built, not promises about how they will be used.

Nothing is written until a person accepts it

Every answer carries the label “Suggested by AI, not verified”. It becomes data only when someone keeps it, applies it or uses it and saves — and then it is that person's record, in its history and the audit log. A risk, a triage or an ISMM section drafted with the assistant, like an accepted manual reference, records who accepted it and when.

It decides nothing

It suggests a classification, a reportability, a rating, a status; people decide each one. The triage of an event, the acceptance of a risk, the approval of a change and the level of a finding stay where Part-IS puts them.

It sends nothing to an authority

AviIS does not transmit to competent authorities at all, with or without the assistant. The reports are yours to send.

Identifiers are replaced before anything is sent

The names of the people AviIS holds — members, appointment holders, supplier contacts, a confidential reporter — email addresses, phone numbers and aircraft registration marks are replaced before text leaves the server. A reporter's identity, the witnesses and indicators of compromise are never sent, and only the fields the feature needs are sent at all.

TLP:AMBER+STRICT and TLP:RED stay inside

A record marked TLP:AMBER+STRICT or TLP:RED is refused before any call to the model, and restricted interfaces and risks are left out of the context of other records — the confidentiality IS.I.OR.200(a)(13) asks for does not depend on the assistant.

Your manuals stay yours

Indexing reads a revision once, as a PDF, to build its index. A reference question or a compliance statement sends the index, not the manual, and the manual is held under the same terms as the rest of your data.

Usage is counted

Every call is recorded per organisation and month — actions and tokens, by feature — against a fair-use cap of 100 AI actions a month.

How it runs

The parts of the answer an inspector or an IT department asks for.

How it runs
Question	Answer
Where does it run?	In server-side functions in Google Cloud europe-west1 (Belgium), the same region as the rest of your data. They read the record, de-identify it and call the model; the browser never holds a key.
Which model?	Claude, from Anthropic, called by AviIS on your behalf through Anthropic's API. You need no account with anyone.
What is sent?	The record the feature works on and the context it needs — the element and its interfaces, the event and its filings, the requirement and the section indexes of your manuals — de-identified. To build an index, the revision being indexed, as a PDF. Never a record marked TLP:AMBER+STRICT or TLP:RED.
Who can use it?	Administrators and information security managers, for every feature; compliance monitoring (auditors) also for compliance statements. It adds no permission and removes none.
Can we turn it off?	It is an add-on: not bought, not present. Bought and unwanted later, it comes off the subscription and the buttons disappear.
What does it cost?	€49 a month for an air operator or any other Part-IS organisation, €29 for an approved training organisation. Included in every 60-day trial.

Try it on your own registers.

Every 60-day trial has the assistant switched on from the first day. Register an element and ask for threat scenarios, report a test event and ask for a triage suggestion, or add your ISMM to the document register and ask for references.

Create your organisation See the price list