Security

ChatPlotDB is an invitation-only private beta, and this page is written to match that. It describes the controls that exist in the product today, in enough detail that you can check them, and it is explicit about the promises we are not making yet. Where this page and our Terms of Service disagree, the Terms govern.

The single strongest control is not ours: point ChatPlotDB at a database role that can only read what should be readable. Everything below is defence in depth behind that.

We are building our program against leading industry frameworks and pursuing independent certification. None of the following is complete: we hold no certification today, and we won't imply otherwise.

• SOC 2 Type II (in progress).
• ISO/IEC 27001 (in progress).
• GDPR and CCPA (in progress).

Security documentation is available to customers under NDA on request.

Database credentials are encrypted before they are stored, and there is no endpoint that hands one back. The connection screen shows a redacted summary, so a password you enter cannot be read out of the product again, by you or by anyone else who reaches your account.

You choose the SSL mode your server requires when you register a connection: require, prefer, allow, or disable. The certificate-verifying modes, verify-ca and verify-full, exist in the data model but are not selectable yet, because there is nowhere to supply a CA certificate.

The agent only ever issues read queries. That is enforced in the product, but it is not a substitute for a scoped database role: grant SELECT on what should be queryable and nothing else.

An optional approval step lets you read the generated SQL and approve it before it runs. It does not remove your responsibility to check what comes back, and section 11 of the Terms says so plainly.

To describe and chart an answer, the rows a query returns are sent to the AI model serving that turn. The model provider therefore sees those rows. Our Privacy Policy names every provider we use, so you can choose a model with that in mind, and an enterprise deployment runs the models inside your own AWS account through Amazon Bedrock if they should never leave it.

Uploaded CSV and Excel files are queried in DuckDB against a copy, not against your infrastructure, which is why file questions have no SQL approval step to skip.

Connections, configs, threads, dashboards, and uploads are scoped to the account and organization that created them, and requests are authorized against that scope. Within an organization, roles control who can administer connections and settings.

Two things can be shared publicly: a thread and a dashboard. Both are snapshots or projections with connection details stripped out, both are rate-limited, and both can be revoked.

A share link is unauthenticated. Anyone holding the URL can read it, so revoke links you no longer need rather than relying on the URL being hard to guess.

The hosted Service runs on Amazon Web Services, and traffic to it travels over TLS. On an enterprise deployment the whole stack runs in your own AWS account and VPC, in the region you pick, governed by the IAM roles, KMS keys, and CloudTrail logging you already run.

We collect application error and performance telemetry so failures surface quickly. If we find that a security incident has affected your data, we will tell you, and we will tell you what we know rather than waiting until the picture is complete.

Being straight about the gaps is more useful to you than a page of controls we have not built. During the beta:

• we do not promise the Service will be available, or set an uptime target;
• we do not promise your data is kept or backed up, and it can be lost;
• we do not promise any feature will keep existing;
• we have not had a third-party penetration test, and we hold no security certification;
• we do not promise the AI's answers are accurate, and you must check them.

Sections 5, 10, and 11 of the Terms say the same in the language that binds. If a control on this list is a requirement for you, talk to us before you connect anything.

• When your access ends, we may delete your data after 14 days, so keep your own copies of anything important.
• Uploaded files carry a time limit, so abandoned uploads do not accumulate.
• We will sweep stored artefacts for your account on request.

If you believe you have found a vulnerability, email support@simplifyingai.com with enough detail to reproduce it. We will acknowledge you, and we ask for a reasonable opportunity to fix the issue before you disclose it publicly.