Vulnerability Disclosure
Found a security issue? We want to hear from you. Here is how we handle vulnerability reports, and what to expect from us in return.
Last updated 2026-08-29 · Machine-readable summary at /security.txt
Email us
Send a clear write-up to security@pentos.ai, or use our Security Issue form if you prefer.
We acknowledge
You will hear back within 3 business days, with an initial triage and status update within 7 business days.
We coordinate
We work with you on a fix and a disclosure timeline. Default coordinated-disclosure window is 90 days.
We credit you
Valid reports are eligible for public credit on the acknowledgments section, with your permission.
Core Infrastructure
Pentos is primarily powered by Cloudflare and Supabase. All sensitive data is encrypted in transit and at rest with AES-256-GCM under a three-level HKDF key hierarchy.
Learn moreWorkers, R2, Workflows, AI Gateway, Browser Run (PDF downloads), WAF (OWASP Core Ruleset), DDoS protection, Turnstile.
Postgres with Row Level Security, Auth (with TOTP MFA), Edge Functions on Deno, Vault for root keys.
What to include
- • A clear description of the issue and its security impact.
- • Step-by-step reproduction instructions, including affected URLs, accounts, or payloads.
- • Proof-of-concept code, screenshots, or HTTP traces where helpful.
- • Your name or handle (for credit, if you want it) and a reply-to address.
In scope
-
pentos.aiMarketing and Documentation -
app.pentos.aiPentos Web Application -
data.pentos.aiDatabase and Edge Functions -
private.pentos.aiPrivate User Files Directory -
public.pentos.aiPublic User Files Directory -
search.pentos.aiPublic Search Directory -
collab.pentos.aiReal-time Collaboration Backend
Out of scope
- • Cloudflare, Supabase, or other third-party platforms and infrastructure we utilize.
- • Vulnerabilities in the AI providers we route to that do not involve Pentos integration code.
- • Reports based solely on automated scanner output without a demonstrated security impact.
- • Missing security headers or cookie attributes on pages that do not handle sensitive data.
AI behavior reports
Pentos routes your prompts to models built by other companies. We cannot fix how those models behave, but we are fully responsible for what happens at the boundary between them and your data. That line decides whether a report is in scope.
- • Getting a model to produce offensive, biased, or objectionable output.
- • Jailbreaks or prompt tricks that only bypass a provider’s own model guidelines.
- • Hallucinations, factual errors, or low-quality answers.
- • A model refusing a prompt, or failing to refuse one.
- • Prompt injection that reads or modifies another user’s data.
- • Injected content that triggers a connector action the user never authorized.
- • Extraction of system prompts, API keys, or infrastructure details.
- • Any path where model output escalates privileges or crosses an account boundary.
The short version: if your finding ends with “and then I reached something that was not mine,” send it.
Test in good faith, and we have your back.
We will not pursue legal action against researchers who, in good faith, follow this policy. Uncertain whether your research is permitted? Contact security@pentos.ai first.
This safe harbor covers claims from Pentos only. It does not bind Cloudflare, Supabase, our AI providers, or any other company whose terms you may be subject to while testing. Keep your testing on the hosts listed in scope above.
Rewards
Pentos does not currently operate a paid bug bounty program. Valid reports are eligible for public credit, with the researcher's permission, and our gratitude.
Acknowledgments
This list will be populated as reports are validated and researchers consent to public credit.