Security and data handling

Your data enters the platform when you upload a dataset. From there, access to it follows the rules on this page: your team, the experts you chose, and the models you configured. Company-level security and privacy — certifications, retention, deletion — are documented on Toloka’s own pages, linked at the end.

Who sees your items

Everything a project holds — datasets, pipelines, runs, results — belongs to one team. Team members open it under their access level: full or read-only. Through the API, a resource on a team you do not belong to answers “not found”.

During a run, the expert labelers you chose see the items they label — that is the work you are paying for. You control who they are through the node’s audience: language, country, and specialization; labeling tasks default to experts who have passed ID verification, and a general labeling node cannot turn that off. See How the platform works.

Under synthetic labeling, items go to the generation model and the review model instead; the expert audience does not apply on that node.

Before a run, almost nothing reads your dataset. The self-check exercises the pipeline on sample values you type in yourself, with one exception: the Dataset Items lint on the Start node reads the items your filter currently selects, to report an empty selection or missing required values. It reads which items hold a value for each required input field, not the values themselves. One preview also reads one item: where the agent built the labeling interface without choosing a sample, it filled the preview form with the dataset’s first item — see Review the labeling interface.

What stays inside

Payment details never touch the platform. Money lives in the Business Console; the platform reads your balance and links out for anything that moves it. See Manage your team and balance.

Credentials your pipeline needs — an API key for an external service a Code node calls — live as secrets in the project. The value stays out of your code, and the agent can reference a secret by name but cannot create one.

Results leave the platform when you export them. Each export is a file you download, and each download link expires after 30 minutes; the export itself stays in the run.

Deleting data

There is no control that removes your labeled data: nothing deletes a dataset or an individual item. Files you attach to a project are the exception — those you can delete. You can rename or hide a dataset, and hiding does not remove it. See Create and manage datasets.

Deletion is an account-level operation under the Privacy Notice: retention follows per-category criteria, and you can request erasure on the grounds the notice lists — data no longer needed, processed unlawfully, or consent withdrawn — by writing to privacy@toloka.ai.

Certifications and policies

The platform runs under Toloka’s company-wide security program, documented outside this portal:

• Toloka Security — SOC 2 Type II, ISO 27001, ISO 27701, the GDPR posture, geo-redundant infrastructure, backup and business continuity.
• Privacy Notice — how long data is kept per category, your rights — access, rectification, erasure, restriction, objection, portability, and consent withdrawal — and how to exercise them.

For privacy requests, write to privacy@toloka.ai; to report a security issue, security@toloka.ai.

Related

• How the platform works — the full path your data takes, from upload to export
• Manage your team and balance — access levels and what read-only members can do
• Programmatic access (API) — how API keys inherit your team access
• CLI — what the CLI sends when you do not opt out