Open data and artifacts
What the Institute intends to publish, under which licence and in what form. The table below is split honestly in two: what does not exist yet, and what will appear once there are results.
Licence: CC BY 4.0
Under CC BY 4.0
Research texts, preprints and reports, the glossary and methods, research protocols, diagrams and charts. One condition: attribution and a link to the source.
All rights reserved
Pricing and commercial terms, paid courses and training programmes, client project reports, software code, and the Institute's name and logo. The CC BY 4.0 licence grants no rights to the mark or the name — the licence says so itself.
Datasets — deferred openness. A dataset is published under CC BY 4.0 twelve months after the article it belongs to is published, or earlier by decision of the Institute. Until then access is on request, and commercial use is under a separate agreement. A dataset does become open over time, but not at once: the article it supports comes first.
Why this way. An open licence makes the material lawfully usable and lets others check our conclusions; grant applications and journals expect exactly this. For datasets the opening is deferred: publishing the article first protects both scientific priority and annotation quality. The commercial part stays closed so that openness does not become giving up revenue. A non-commercial licence (CC BY-NC) is not used: it forbids use even where we want to allow it and sits badly with funder requirements.
Rights holder: ANO «ISAI», 2026.
What does not exist yet
The Institute is not registered, research has not started, there are no publications. Below is what is to appear and in what form, so that it can be checked.
| Artifact | Format | Status | Access |
|---|---|---|---|
| AI Behaviours Corpus — annotated corpus of agent behaviour | JSONL + annotation schema | planned, 2027 | open after 12 months |
| Ecosystem Graph — graph of agent interactions in a test environment | GraphML + field description | planned, 2027 | open after 12 months |
| AI Incident Chronicle — chronicle of incidents involving AI agents | CSV + case cards | planned, 2027 | open after 12 months |
| Benchmark of resilience to indirect injection | scenario set + run script | planned | on request |
| Research protocol on content injection | PDF, version 0.1 | draft prepared | open |
| Unpatched vulnerability details | — | — | not released |
Three access tiers
Open
Research texts, the glossary, methods and protocols are published immediately, without registration or request, under CC BY 4.0. Datasets open under the same licence twelve months after the article they belong to (or earlier by decision of the Institute).
On request
Until a dataset opens it is released on request, after the intended use and organisation are stated; commercial use before opening is under a separate agreement. The reason for the restriction is given plainly: the set contains working injection scenarios.
Restricted
Not released: unpatched vulnerability details, personal data, and client material without their written consent.
We do not publish datasets "for the future" and do not open an empty repository. While there are no artifacts, this page states the rules: so a reader has something to hold us to when results appear.