Platform
AI data provenance that shows where your data came from and where it goes
AI data provenance in Aicura records the data your AI systems and agents run on, where it came from, who owns it and how it is classified. When someone asks whether customer records were used to train a model, you can show which systems used which data, and at which stage, from training to logging.
Included in Pro and Enterprise. Compare plans
Assets
A register for the data your AI runs on
The questions about an AI system often turn out to be questions about its data, such as where it came from, whether it holds personal information, who is accountable for it and whether it was used for training. Aicura answers them for each piece of data and each way it is used.
Every dataset with its owner, classification and sensitivity
Aicura calls each piece of data an asset. Each asset carries its type, classification, sensitivity, whether it holds personal or special-category information, its owner and whether a connected platform brought it in or someone added it by hand.
- Types from datasets, tables and buckets to streams, corpora, feeds, model artefacts and document sets.
- Governance fields for steward, jurisdiction, legal basis, licence, retention, data categories and quality, owned in Aicura.
- Reviews that record who looked and set the next review date. An asset in declared use that has not been reviewed in time opens a finding.
Which system uses which data, and at what stage
A use links an AI system or agent in your register to an asset at one stage, with a purpose. The same asset can be used for training by one system and as inference input by another, and each use is recorded separately. An approved AI use request declares the data it named on the entry, and declared uses are kept as dated snapshots in the Evidence Vault.
- Collection
- Training
- Fine-tuning
- Evaluation
- Grounding and retrieval
- Inference input
- Generated output
- Logging and retention
See where data came from and where it goes
Lineage records where an asset came from and where it goes, upstream and downstream, with each change to the data described in plain words. Lineage arrives from a connected platform or is added by hand.
The map draws a picture around any AI system or asset, with its uses labelled by stage and its lineage labelled by what changed, expanding as you explore.
Rules for your data, and findings when one is not met
Rules are what your organisation holds itself to. A rule names a condition over the map, such as a stage, personal or special-category data, a sensitivity or a border crossing, and what it requires. A rule not met is a finding for a person to work.
- Personal information observed but not declared, when the AI gateway (the component that checks AI traffic in your network) removes personal information from a system's traffic and no declared asset carries it.
- Register entry and data assets disagree on whether personal information is involved.
- Asset changed at the source, when a sync finds the platform changed a classification.
Platforms
Connect the catalogue your data already lives in
Connect a data catalogue, lineage or privacy platform and Aicura brings in its datasets, classifications, owners and lineage, suggests which AI system uses a dataset when the names match, and keeps a history of every change the platform reports. A person confirms each suggestion. Credentials are write-only and never shown back.
AWS DataZone, Google Dataplex, Informatica and Varonis are coming soon.
Questions
Do we have to connect a data catalogue to use data provenance?
No. Datasets and other data assets can always be added by hand, beside any that a connected platform brings in. Connecting a data catalogue, lineage or privacy platform brings its assets, classifications, owners and lineage into Aicura and keeps them current on every sync.
Will a sync overwrite what our team has recorded?
No. The platform that reports an asset owns its definitional facts, such as name, type, classification and source identifier. Aicura owns the governance fields, such as owner, steward, jurisdiction, legal basis, licence and retention. A sync refreshes the first set and never touches the second.
How are data findings raised and closed?
Findings come from the rules your organisation sets and from what the gateway observes in live traffic. Rules are re-evaluated after every write and every sync, and a finding resolves on the next evaluation once the rule or the data no longer applies. Nothing is closed by hand, so a resolved finding means the facts changed.
Does a proposed data use count as declared?
No. A use suggested by a sync, for example where a dataset's name matches an AI system in your register, is not your organisation's position until someone confirms it. Suggested uses are left out of what Aicura reads when it drafts documents for you, and out of the Evidence Vault.
Know what data your AI runs on
Data provenance and its connectors are part of Pro. Contact us about the platforms your data already lives in.