KYA is governance and trust infrastructure for autonomous systems — vulnerabilities here have outsized impact on downstream deployments. We take security reports seriously and respond within one business day.

We support the latest minor version. Critical security fixes are backported to the previous minor version when feasible.

Do not open a public GitHub issue for security vulnerabilities.

Please report privately via one of these channels (in order of preference):

1. GitHub Security Advisories — open a private advisory on this repo. Preferred — handles disclosure, CVE assignment, and patch coordination cleanly.
2. Email — kola@veldtlabs.ai. PGP key on request.

• A description of the vulnerability and the impacted KYA component (e.g., kya/inbound.py, evidence.py, delegation_policy.py)
• Steps to reproduce, including minimal code and KYA version
• The threat model context (what the attacker controls; what they gain)
• A proposed fix or mitigation, if you have one
• Your name and affiliation as you want them credited (or "anonymous")

If the issue is being actively exploited, we'll fast-track disclosure and ship a patch immediately.

In scope:

• Authentication / authorization flaws in any KYA component
• Cryptographic weaknesses in the evidence chain (evidence.py), the four-gate apply pipeline (inbound.py, _inbound_signing.py), or the Ed25519-signed federated recommendation flow
• Tenant-isolation bypasses (KYP, principals, weights)
• Composition algebra violations (only-tighten bypass)
• Delegation-attribution evasion in multi-agent fan-out
• Red-team harness escape (kya_redteam)
• Supply-chain integrity of the veldt-kya PyPI package

Out of scope:

• Issues in third-party agent frameworks (LangChain, CrewAI, etc.) — please report to those projects
• Issues in deployed applications using KYA — please report to those operators
• Findings from automated scanners with no demonstrated exploit
• Social engineering against Veldt Labs personnel

We will not pursue legal action against good-faith security researchers who:

• Report vulnerabilities privately and give us reasonable time to respond
• Avoid privacy violations, data destruction, or service disruption during research
• Do not exploit vulnerabilities beyond the minimum necessary to demonstrate impact

We publicly credit reporters in the CHANGELOG.md and the GitHub Security Advisory unless they request anonymity.

Some intentional design boundaries are documented in the paper (arXiv:2605.25376, §11). These are not vulnerabilities. Please read §11 before reporting findings related to those boundaries.

Thank you for helping keep KYA and its downstream users safe.