Vulnerability Disclosure Policy

Last updated: July 2026.

DRAFT — pending legal/counsel review. The safe-harbor and legal-authorization sections below are a good-faith draft and are not yet a binding legal commitment. They are published for transparency while counsel finalizes the language. Treat the how to report, scope, and expectations sections as operative today; treat the safe-harbor wording as provisional until this notice is removed.

Our commitment

Praesidia is a control plane for authenticating, governing, and monitoring AI agents, applications, and MCP servers. Security is the product, so we hold our own systems to the same standard we ask customers to hold theirs. We welcome reports from the security community and treat every good-faith report seriously. If you believe you have found a vulnerability in a Praesidia system, this page tells you how to tell us.

How to report

Email security@praesidia.ai with enough detail for us to reproduce and validate the issue. A good report includes:

• The affected asset (URL, API endpoint, SDK, or hostname).
• A clear description of the vulnerability and its potential impact.
• Step-by-step reproduction, proof-of-concept, or a short screen capture. Please keep proof-of-concept payloads non-destructive.
• Any accounts, tenants, or identifiers you used — created with your own data only.

Machine-readable contact details are published at /.well-known/security.txt per RFC 9116. If you need to send sensitive material, note that in your first email and we will arrange an encrypted channel.

What to expect (response targets)

These are our current operational targets, not contractual guarantees. We will do our best to meet them and to keep you informed if a report is complex.

• Acknowledgement: within 3 business days of receipt.
• Triage and initial assessment: within 10 business days.
• Progress updates: at least every 15 business days while the issue is open.
• Resolution: timeline depends on severity and complexity; we will share our assessment and target with you.

We practice coordinated disclosure. We ask that you give us a reasonable opportunity to remediate before publishing, and we are happy to coordinate a public write-up and credit you once a fix has shipped.

Scope

In scope

• praesidia.ai and its subdomains (the marketing site and documentation).
• app.praesidia.ai — the Praesidia application and dashboard.
• The Praesidia REST API and the published Python SDK (praesidia-server-sdk).
• The open-source Praesidia core (report code-level issues via our public repositories or by email).

Out of scope

• Denial-of-service, volumetric, or resource-exhaustion testing.
• Social engineering, phishing, or physical attacks against Praesidia staff, users, or facilities.
• Reports from automated scanners without a demonstrated, exploitable impact.
• Third-party services, model providers, or infrastructure we do not operate.
• Findings that require access to another customer's account or data — do not attempt to reach data that is not yours (see the rules below).
• Best-practice suggestions with no security impact (for example, missing hardening headers with no exploit path).

Rules of engagement

To keep testing safe for our customers, we ask that you:

• Only test against accounts and tenants you create and own. Never access, modify, or exfiltrate data belonging to other customers.
• Stop as soon as you have confirmed a vulnerability, and do not pivot further into our systems.
• Do not run tests that degrade service for other users.
• Keep any customer data you inadvertently encounter confidential, stop testing, and report it immediately.
• Give us a reasonable time to remediate before any public disclosure.

Safe harbor (DRAFT — pending counsel)

The following statement is a good-faith draft under legal review and is not yet a binding legal commitment.

We intend to consider security research and vulnerability disclosure activities conducted in good faith and in accordance with this policy to be authorized conduct, and we do not intend to pursue or support legal action against researchers for accidental, good-faith violations of this policy. If a third party initiates legal action against you for activity conducted in accordance with this policy, we intend to make clear that your activity was authorized. This is a statement of intent pending finalization by counsel; it does not waive any rights that cannot be waived by law, and it does not authorize activity that violates applicable law or accesses data belonging to third parties.

Recognition

We do not currently operate a paid bug-bounty program. We are happy to credit researchers publicly (with your consent) for valid, in-scope reports, including in the write-ups we publish in our security research section.

How this connects to our own testing

Disclosure is one half of the picture; the other is testing ourselves before anyone else does. Praesidia runs continuous, opt-in adversarial testing against its own registered entities — scope-guarded to your own agents and MCP servers, using synthetic payloads with no external egress. You can read how that works in Continuous adversarial testing for AI agents, and see the broader control set on our Security & Compliance page.

Contact

Security reports: security@praesidia.ai. For non-security matters, see our FAQ or the contact details on the About page.