KAS acceptable use

AI Verify process checks 9.1.1. This page is the public text for those checks. It is not a pass mark. Certification is in progress. Qypu holds no certificate yet.

The short version

This short version helps you understand the full text. Read the full text for the complete terms.

• Use KAS to declare real effects, sign a narrower mandate, and keep every receipt. (Full text, Permitted use)
• Do not hide an effect, forge a receipt, or point the gate at an illegal act. (Prohibited use)
• KAS does not write model output. If you attach a model, you answer for its words. (Model outputs)
• An allow is not a certification, and it is not a claim that the action is safe. (Prohibited use)
• Report misuse through the concern channel. No review date is set for this policy. (Recourse and review)

Full text

Which checks this page answers

This page is the public text for AI Verify process check 9.1.1. It states acceptable use, unacceptable use, and a path for recourse. It is a draft. It is not in force.

The acceptable use policy for the Qypu social product still applies to that product. This page is about KAS. It does not repeat that policy.

Permitted use

You may use KAS to do these things.

• Declare, on a card, the effects the tool really has.
• Sign a mandate that allows only a subset of those effects. A sub-mandate must be narrower, not wider.
• Run observe mode, and then enforce mode, as the package describes.
• Keep the tape, including refusals and holds.
• Run a witness, on a machine you control, as a party other than the gate operator.

Prohibited use

You must not do these things.

• Hide an effect that the card does not name, or send the real action around the gate.
• Forge a receipt, delete one, or skip the tape.
• Use the gate for an effect that is a crime where you, or the person affected, are.
• Use an agent to harm a child, to enter a system you do not own, or to present software as a human being.
• Tell anyone that an allow means the action is certified, audited, or safe.

The social product's bans on fake accounts, fake reviews, and political campaigning stay on the acceptable use policy and the content policy. KAS is not that publisher.

Model outputs

KAS does not write model outputs. It does not store prompts. The gate does not screen generated text for copyright, abuse, or illegal content. A public sentence can still carry a secret. The safety notes say the project does not claim to block that channel.

If you connect a model, you are responsible for what it writes and for what you do with those words. You must not use a model, with KAS or beside it, to produce sexual material of a minor, a scam, or malware. That sentence is a ban. It is not a guide to doing those things.

Qypu has not selected a model for a KAS product. Eval runs in the repository are experiments. They are not a product decision.

Recourse and review

Anyone can report misuse or harm at the concern channel. A rights claim goes to the licensing page. Product complaints also have the complaints page.

This page does not publish a violation log, a takedown log, or an intellectual-property log.

No end-user consultation has been recorded. The pilot has not been run. This policy has no review date. None is promised.

Change log

• 2026-10-25 · 0.1.1 · This page does not publish a violation log. Not in force.
• 2026-10-23 · 0.1.0 · First public draft of KAS acceptable use. Not in force.

Contact

Qypu, Inc., a Delaware C corporation. Contact address: 51A Marang Road, Singapore 099281. Privacy mail privacy@qypu.ai and security mail security@qypu.ai route to Byron Fuller.

KAS · KAS governance · Trust centre · Privacy · Privacy notice