Security Disclosure

RiskState welcomes security research and confidential vulnerability reports. This page is the canonical disclosure policy. The machine-readable version lives at /.well-known/security.txt per RFC 9116.

Contact

Channel	Use
`security@likido.xyz`	Confidential security reports. Preferred channel.
`hello@likido.xyz`	General inquiries, contracts, GDPR DSAR

If you need PGP, ask via security@likido.xyz and we will exchange a key out-of-band.

Response SLA

Step	Target
First acknowledgement	Within 72 hours
Triage status update	Within 7 days of acknowledgement
Resolution + public advisory (if confirmed)	Within 30 days for production-impact issues

Response times are measured against business days where local holidays apply. We confirm in writing.

Scope

In scope:

The application at https://riskstate.ai
The dashboard at https://riskstate.netlify.app
All published API endpoints under /v1/* and /v2/*
The published MCP server package: @riskstate/mcp-server

Out of scope:

Third-party services we depend on (CoinGecko, Binance, OKX, Bybit, CryptoCompare, FRED, DefiLlama, Netlify, Resend, Lido, Beaconcha.in, etc.) — please report to them directly.
Issues that require physical access to a user's device.
Reports based on assumed configurations not reproducible against production.
Issues in customer integrations of the API that are not caused by our endpoint behaviour.

What constitutes a vulnerability

Examples of in-scope reports:

Authentication bypass on /v1/* or /v2/* (e.g. accessing audit records without a valid Bearer token).
Privilege escalation: an external API key reading another customer's audit records, policy_config, or usage stats.
Logic bug in policy_config enforcement that loosens the engine's caps (a "tighten only" violation).
Logic bug in policy_hash that produces non-deterministic output for identical inputs and reference_time.
Credential exposure (API key or trading wallet key in logs, error responses, or public artifacts).
Server-side request forgery, command injection, server-side template injection, deserialisation issues.
Account takeover or password / token theft on riskstate.ai.

What is not a vulnerability:

The score_v3 engine's policy output disagreeing with your view of the market. The engine is documented in Methodology; disputes about scoring are not security issues.
Missing security headers on static pages where they are not load-bearing.
Self-XSS that requires a victim to paste attacker-supplied content into the dashboard.
Reports from automated scanners without a working proof-of-concept.

Hard NO

These will not be tolerated and may result in legal action:

Active exploitation against live customer keys. Use only your own API key.
Denial-of-service testing without prior written authorization. Burst load is fine within the 60 req/min rate limit per key; sustained DoS is not.
Social engineering of any RiskState contact, partner, or contracted service operator.
Physical attacks on infrastructure or personnel.
Data exfiltration beyond the minimum needed to demonstrate the issue. If you find a way to read the audit log of another customer, retrieve one record as proof — do not enumerate.

Coordinated disclosure

We follow a coordinated disclosure model. After triage:

We confirm the issue and an estimated remediation timeline.
You agree to hold publication until we have shipped the fix and notified affected customers (if any).
We publish a public advisory at our blog / /docs/research and credit you (unless you request anonymity).
If we cannot reach the 30-day window for a complex fix, we coordinate a longer hold with you in writing.

We do not pay bounties today. We do credit researchers in advisories and, for impactful reports, write a public thank-you. Once we have a paid plan, a bounty programme will follow.

Reporting template

Send to security@likido.xyz:

Subject: [SECURITY] <one-line summary>

Affected component: <e.g. POST /v1/risk-state, GET /v2/audit/decisions, ...>
Vulnerability class: <e.g. authn bypass, IDOR, injection, logic bug>
Severity (your assessment): <CRITICAL | HIGH | MEDIUM | LOW>
Reproduction steps:
  1. ...
  2. ...
  3. ...
Expected behaviour: ...
Observed behaviour: ...
Proof of concept: <minimal payload, screenshots, request/response>
Reporter: <name or pseudonym; whether you want public credit>

A short, reproducible report beats a long speculative one.

Acknowledgements

No external findings to date. We will list confirmed reporters here unless they request anonymity.

Institutional Integration Guide → Incident response
Audit Log — for understanding what is logged on every API call
Privacy Policy
Legal

This document and the /.well-known/security.txt file are reviewed at least annually. Last review: 2026-04-28.