Vulnerability disclosure policy

Our commitment to you

• We will acknowledge your report within 3 working days.
• We will give you an initial assessment — whether we can reproduce it, how serious we think it is, and what we plan to do — within 10 working days.
• We will keep you informed as we work, and tell you when it is fixed.
• We aim to remediate validated issues within 90 days of your report, sooner for anything serious. If we need longer we will say so and why.
• We will not take legal action against you, or ask anyone else to, for research conducted in good faith within this policy, and we will not report you to law enforcement for it.
• With your agreement we will credit you when we disclose the fix. If you prefer to stay anonymous, we respect that.

What we ask of you

• Report promptly and privately, with enough detail to reproduce: the affected URL or component, steps, and any proof-of-concept. Please do not include real customer data in the report.
• Do not access, modify, delete or exfiltrate data that is not yours. If you reach another tenant's data by accident, stop, record the minimum needed to demonstrate the issue and tell us at once.
• Do not degrade the service — no denial-of-service, no automated scanning at volume, no spam or social engineering of our people or customers, no physical attacks.
• Do not pivot: once you can demonstrate the issue, stop there. Proof is enough; exploitation is not required.
• Give us a reasonable opportunity to fix the issue before you disclose it publicly. We ask for coordinated disclosure: after the fix, or 90 days from your report, whichever is sooner — and we will agree a date with you.
• Test only against accounts you own or are explicitly authorised to use. If you would like a test account on a pilot workspace, ask — we would rather help than have you guess.

In scope

• The product at app.tesserasecurity.ai — authentication and session handling, authorisation and tenant isolation, the model pipeline and the identifier-minimisation control, billing and credits, integrations, exports.
• The agent API and MCP server, and the tessera-mcp client package.
• The public website and any sub-domain we operate.
• Our build and deployment pipeline, where you can show an impact on the above.

We are especially interested in: cross-tenant access or existence disclosure; any path by which an AI agent could validate an element; any path by which un-minimised secrets or identifiers reach the model provider; authentication or MFA bypass; credit-gate bypass.

Out of scope

• Findings from automated scanners with no demonstrated impact; missing best-practice headers with no exploit; clickjacking on pages with no sensitive action; rate-limiting or brute-force reports without a working bypass.
• Published CVEs in third-party operating-system packages that are not reachable in our runtime — we track these in our vulnerability-management process and VEX register; tell us if you can show one is actually reachable.
• Issues in third-party services we use (AWS, Cognito, Anthropic, Stripe) that are not caused by our configuration — report those to the vendor; tell us too and we will act on our side.
• Social engineering, physical security, and denial-of-service.
• The AI model's judgement on security content (a threat you think is mis-scored, a control you disagree with). That is product feedback and we want it — send it to hello@tesserasecurity.ai — but it is not a vulnerability.

Safe harbour

Research carried out in good faith and in line with this policy is authorised. We consider it to be lawful and we will not pursue or support legal action against you for it. If a third party initiates action against you for research consistent with this policy, we will make it known that your actions were authorised. This authorisation does not extend to actions outside this policy, to accessing data that is not yours beyond the minimum needed to demonstrate an issue, or to anything that breaks the law.

How we handle a report

1. Acknowledge and open a tracked case.
2. Reproduce and assess severity, with the founder as incident lead; if customer data may be affected, the incident response and breach-notification runbook starts in parallel.
3. Fix, test and deploy, with independent re-verification for anything serious.
4. Tell you, agree disclosure, credit you if you wish.
5. Record the lesson in our standing review.

Rewards

We do not currently run a paid bounty programme. We say thank you properly, credit you publicly if you wish, and we have been known to send something. We will say so here if that changes.