Airship
WorkServicesHorizonsSectorsCase studiesAboutContact
AI, governed

We can tell you how every asset was made, and show you the record.

Don't take the policy on trust. Ask us for the evidence: the tool list, the vendor terms, and a chain-of-custody record for an asset you pick.

Why this is the question now

Outright bans on AI are giving way to disclosure terms in the contracts we're asked to sign. Your question has changed with them, from whether we use AI to whether we can say what's in the file.

Several studios have shipped titles carrying AI artefacts nobody decided to put there, then had to patch and explain. None of it was strategy; each was a gap in review, and a policy doesn't close one. A process does.

What that means in practice
01
A reviewed tool list, with the vendor terms behind it

Not a list of what we like. A list of what's been read, negotiated and approved, each under an agreement that meets your confidentiality, security, IP and no-training terms. Most tools don't pass.

02
Chain of custody on every deliverable

Who made it, which tools touched it, what reference was used, what a human authored. Ask how any single asset was made and the answer comes with the record attached.

03
A clean room around your material

Your IP never enters a third-party model, directly or indirectly, aggregated or not. Deletion certified in writing when you ask.

04
Records that survive an audit

Documented diligence per deliverable. A named human reviewer accountable for each pass. The file available to you on reasonable notice.

The five rules
01
Written approval first, and revocable
Before any AI touches your project we ask, in writing, and record the answer. You can say no, yes, or yes with conditions such as human review, limited data access or watermarking. You can withdraw the yes at any point. We re-ask when anything material changes.
02
Tools under terms at least as protective as yours
Every AI tool we use sits under a written agreement that protects your material at least as well as your contract with us does. If a tool can't give us that, we don't use it on your work, whatever it would save.
03
Nothing of yours trains a model
Not your reference, not your briefs, not what we make for you. No model of ours or anyone else's learns from it, even anonymised or aggregated. It's the rule with no exceptions.
04
Marked deliverables and a kept record
Any deliverable that used AI is marked, with the human and AI parts identified. We keep the prompts, references and process on file, and hand them to you on request.
05
If ownership is ever at risk, we redo it
If an AI output ever threatens your ability to own the work outright, we tell you, and we remake it by hand at our cost.
Ask any vendor for these four

Including us. It's the fastest way to find out whether a studio has done this work or written a paragraph about it.

01Your tool allowlist, and the vendor terms behind each one.
02A chain-of-custody record for one specific asset that you choose.
03Your no-training clause, written into the contract.
04What happens, in writing, if an output threatens our ownership.
Where we already work to this

One of our largest clients writes a full AI addendum into its master agreement: prior approval, tool-vendor terms, an absolute training prohibition, marking and explainability, documented diligence, indemnity and audit rights. We've delivered under it for 2 years. The standard is the publisher's, and it's a higher bar than we'd have set ourselves.

The compliance spine underneath it already existed: Cyber Essentials certified, TPN in renewal, ISO 27001 in progress. AI governance was bolted onto that spine, and this page describes it.

Where we use it, and where we do not
With approval, never on the final asset unmarked

Reference gathering, blockout and layout exploration, texture variation through WispR, our own tool, and the parts of a pipeline that check work without producing any.

Where a studio says no

None of it happens on your work, and every artist on the project is told so in writing. A prohibition that lives only in a contract isn't a prohibition.

Why we publish it

Because the honest answer to your legal team is a process they can inspect, not a promise they have to take. Several clients have written their own AI terms into our contracts. This is the policy that lets us sign them, and it's the same for everyone. If yours is stricter, send it and we'll work to it.

Ask for the policy, or send us yours to sign.

Request the AI policySend us a brief
Book a 30-minute callOr write: hello@goairship.ukYou'll hear back within a working day.
Airship

An art and tech-art studio of 100+ artists, credited on 82 shipped titles since 2017.

Work
The indexCase studiesSectors
Two ways in
Core: send a briefHorizons: ask for a seatBook a call
Studio
AboutPeopleCareersAI, governed
Contact
hello@goairship.ukThe Barn, Blackmore Park Road
Malvern WR14 6PH
Airmergent Limited, trading as Airship. Registered in England and Wales, company number 15983872. © 2026.
PrivacyTerms
AirshipMENU
AI, governed

We can tell you how every asset was made, and show you the record.

Don't take the policy on trust. Ask us for the evidence: the tool list, the vendor terms, and a chain-of-custody record for an asset you pick.

Why this is the question now

Outright bans on AI are giving way to disclosure terms in the contracts we're asked to sign. Your question has changed with them, from whether we use AI to whether we can say what's in the file.

Several studios have shipped titles carrying AI artefacts nobody decided to put there, then had to patch and explain. None of it was strategy; each was a gap in review, and a policy doesn't close one. A process does.

What that means in practice
01
A reviewed tool list, with the vendor terms behind it

Not a list of what we like. A list of what's been read, negotiated and approved, each under an agreement that meets your confidentiality, security, IP and no-training terms. Most tools don't pass.

02
Chain of custody on every deliverable

Who made it, which tools touched it, what reference was used, what a human authored. Ask how any single asset was made and the answer comes with the record attached.

03
A clean room around your material

Your IP never enters a third-party model, directly or indirectly, aggregated or not. Deletion certified in writing when you ask.

04
Records that survive an audit

Documented diligence per deliverable. A named human reviewer accountable for each pass. The file available to you on reasonable notice.

The five rules
01
Written approval first, and revocable
Before any AI touches your project we ask, in writing, and record the answer. You can say no, yes, or yes with conditions such as human review, limited data access or watermarking. You can withdraw the yes at any point. We re-ask when anything material changes.
02
Tools under terms at least as protective as yours
Every AI tool we use sits under a written agreement that protects your material at least as well as your contract with us does. If a tool can't give us that, we don't use it on your work, whatever it would save.
03
Nothing of yours trains a model
Not your reference, not your briefs, not what we make for you. No model of ours or anyone else's learns from it, even anonymised or aggregated. It's the rule with no exceptions.
04
Marked deliverables and a kept record
Any deliverable that used AI is marked, with the human and AI parts identified. We keep the prompts, references and process on file, and hand them to you on request.
05
If ownership is ever at risk, we redo it
If an AI output ever threatens your ability to own the work outright, we tell you, and we remake it by hand at our cost.
Ask any vendor for these four

Including us. It's the fastest way to find out whether a studio has done this work or written a paragraph about it.

01Your tool allowlist, and the vendor terms behind each one.
02A chain-of-custody record for one specific asset that you choose.
03Your no-training clause, written into the contract.
04What happens, in writing, if an output threatens our ownership.
Where we already work to this

One of our largest clients writes a full AI addendum into its master agreement: prior approval, tool-vendor terms, an absolute training prohibition, marking and explainability, documented diligence, indemnity and audit rights. We've delivered under it for 2 years. The standard is the publisher's, and it's a higher bar than we'd have set ourselves.

The compliance spine underneath it already existed: Cyber Essentials certified, TPN in renewal, ISO 27001 in progress. AI governance was bolted onto that spine, and this page describes it.

Where we use it, and where we do not
With approval, never on the final asset unmarked

Reference gathering, blockout and layout exploration, texture variation through WispR, our own tool, and the parts of a pipeline that check work without producing any.

Where a studio says no

None of it happens on your work, and every artist on the project is told so in writing. A prohibition that lives only in a contract isn't a prohibition.

Why we publish it

Because the honest answer to your legal team is a process they can inspect, not a promise they have to take. Several clients have written their own AI terms into our contracts. This is the policy that lets us sign them, and it's the same for everyone. If yours is stricter, send it and we'll work to it.

Ask for the policy, or send us yours to sign.

Request the AI policySend us a briefBook a 30-minute call
hello@goairship.uk
You'll hear back within a working day.
Airship
Work
The indexCase studiesSectors
Two ways in
Core: send a briefHorizons: ask for a seatBook a call
Studio
AboutPeopleCareersAI, governed
Contact
hello@goairship.uk
The Barn, Blackmore Park Road
Malvern WR14 6PH
Airmergent Limited, trading as Airship. Registered in England and Wales, company number 15983872. © 2026.
Privacy · Terms