Skip to content

Commit 32133b7

Browse files
emmanuelgjrclaude
emmanuelgjr
and
claude
committed
Add security policy and acknowledgements
Establish coordinated-disclosure process for code, datasets, and published guidance. Scope explicitly covers AI-specific risks (prompt injection in shipped tooling, dataset poisoning) alongside conventional vulnerabilities. Reports route through GitHub Private Vulnerability Reporting or emmanuelgjr@owasp.org. SECURITY-THANKS.md seeded for future reporter credit. Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
1 parent 4dc2ee4 commit 32133b7

2 files changed

Lines changed: 185 additions & 0 deletions

File tree

‎SECURITY-THANKS.md‎

Lines changed: 36 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,36 @@
1+
# Security Acknowledgements
2+
3+
The **OWASP GenAI Data Security Initiative** thanks the following researchers
4+
for reporting security issues in good faith under our
5+
[Security Policy](./SECURITY.md).
6+
7+
Inclusion here is at the reporter's request and reflects coordinated, valid
8+
disclosure of an in-scope issue.
9+
10+
| Date (YYYY-MM-DD) | Reporter | Affected artifact | Severity | Advisory |
11+
|-------------------|---------------------------|-------------------|----------|----------|
12+
| _No reports yet — be the first._ | | | | |
13+
14+
## How to be credited
15+
16+
When you submit a report, let us know:
17+
18+
- The name or handle you'd like used
19+
- Optional: an affiliation, profile URL, or social handle
20+
- Whether you'd prefer to remain anonymous (your report is still valued)
21+
22+
We will not publish your identity without consent, and you can ask to be
23+
removed from this list at any time.
24+
25+
## What qualifies
26+
27+
A finding is added here once **all** of the following are true:
28+
29+
1. It is in scope under [`SECURITY.md`](./SECURITY.md).
30+
2. It has been triaged and validated by the maintainers.
31+
3. A fix, mitigation, or formal acceptance has shipped.
32+
4. The reporter has consented to public credit.
33+
34+
---
35+
36+
*Last updated: 2026-05-11*

‎SECURITY.md‎

Lines changed: 149 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,149 @@
1+
# Security Policy
2+
3+
The **OWASP GenAI Data Security Initiative** publishes security research,
4+
guidance, datasets, and tooling for the GenAI/LLM ecosystem. As a security
5+
project, we hold our own artifacts to the standards we ask others to meet —
6+
this policy describes how to report vulnerabilities or material defects in
7+
anything we publish.
8+
9+
## Scope
10+
11+
This policy applies to content in this repository:
12+
13+
- **Code** — Python validators, the `dsgai_scanner_tool`, the JavaScript
14+
crosswalk explorer, HTML resources, and CI/build scripts.
15+
- **Datasets** — `dsgai-bench`, validation datasets, and community-contributed
16+
data hosted here.
17+
- **Frameworks & documentation** — risk taxonomies, best-practices guides, and
18+
the framework crosswalk content, where errors could materially mislead a
19+
reader in a way that weakens their security posture.
20+
21+
### In scope
22+
23+
- Code execution, injection, deserialization, path traversal, SSRF, XSS in
24+
tools we ship
25+
- **Prompt injection, jailbreaks, or unsafe LLM-output handling in
26+
`dsgai_scanner_tool`** or any other AI-enabled component we ship
27+
- **Dataset poisoning, contamination, or adversarial-example susceptibility**
28+
in `dsgai-bench` or validation datasets
29+
- Sensitive data accidentally committed (keys, tokens, PII, proprietary
30+
third-party content)
31+
- Datasets containing live secrets, PII, copyrighted material, or content
32+
that could enable harm at scale
33+
- Supply-chain issues (typosquatted/compromised dependencies, malicious build
34+
steps, tampered releases)
35+
- Authoritative errors in published guidance that would lead a reasonable
36+
implementer to a less-secure outcome
37+
- Unverifiable or modified release artifacts
38+
39+
### Out of scope
40+
41+
- Misuse of our published frameworks or taxonomies by third parties
42+
- Vulnerabilities in upstream dependencies — please report those upstream;
43+
we will pin/patch once an advisory exists
44+
- Issues in third-party frameworks the crosswalk maps to (NIST AI RMF,
45+
ISO/IEC 42001, MITRE ATLAS, etc.) — those belong with their maintainers
46+
- Theoretical risks already documented in our published taxonomy
47+
- Social-engineering, physical, or denial-of-service testing against
48+
contributors or infrastructure
49+
- Findings produced solely by automated scanners without demonstrated impact
50+
51+
## Reporting a Vulnerability
52+
53+
**Do not open a public GitHub issue for security reports.**
54+
55+
Use one of the following private channels:
56+
57+
1. **Preferred — GitHub Private Vulnerability Reporting:**
58+
<https://github.com/GenAI-Security-Project/GenAI-Data-Security-Initiative/security/advisories/new>
59+
2. **Email:** `emmanuelgjr@owasp.org`
60+
Subject: `[SECURITY] GenAI Data Security Initiative — <short summary>`
61+
62+
For escalation involving OWASP infrastructure (not this repository's content),
63+
contact the OWASP Foundation directly: <https://owasp.org/contact/>.
64+
65+
Please include, where possible:
66+
67+
- Affected file(s), commit SHA, dataset record(s), or release artifact
68+
- Description of the issue and realistic impact
69+
- Reproduction steps or a minimal proof-of-concept
70+
- Suggested remediation, if you have one
71+
- Whether you wish to be credited, and the name/handle to use
72+
73+
If you need to send sensitive material, request our PGP key in your initial
74+
email (fingerprint published once generated) and we will respond before you
75+
transmit details. Reports are accepted in **English**.
76+
77+
## Our Commitments
78+
79+
| Stage | Target |
80+
|--------------------------------|------------------------------|
81+
| Initial acknowledgement | within 3 business days |
82+
| Triage and severity assessment | within 10 business days |
83+
| Status updates during triage | at least every 14 days |
84+
| Fix or mitigation plan | within 90 days of validation |
85+
86+
We use **CVSS v3.1** for severity scoring and will request a CVE for
87+
qualifying issues in shipped code or releases. AI-specific findings are
88+
additionally classified using the **OWASP Top 10 for LLM Applications** and,
89+
where applicable, the categories in our own *GenAI Data Security Risks and
90+
Mitigations 2026*.
91+
92+
## Coordinated Disclosure
93+
94+
We follow a **coordinated disclosure** model:
95+
96+
- We will agree a public-disclosure date with you, default 90 days from
97+
validated triage, or sooner if a fix ships earlier.
98+
- We will not disclose your identity without consent.
99+
- We will credit reporters in the published advisory and in
100+
[`SECURITY-THANKS.md`](./SECURITY-THANKS.md), unless you request anonymity.
101+
- If a vulnerability is being actively exploited in the wild, we may publish
102+
guidance ahead of the agreed date and will notify you first.
103+
104+
## Safe Harbor
105+
106+
We consider security research conducted in good faith under this policy to be:
107+
108+
- Authorized under applicable anti-hacking and anti-circumvention laws, and
109+
- Exempt from restrictions in our terms of use that would interfere with
110+
good-faith security research.
111+
112+
You are expected to comply with all applicable laws, avoid privacy violations,
113+
data destruction, and service disruption, and to use only the access necessary
114+
to demonstrate the issue. If in doubt about whether a planned activity is
115+
covered, contact us before testing.
116+
117+
## Supported Versions
118+
119+
We maintain security fixes for the **current published edition** of each
120+
artifact and the `main` branch. Prior editions receive fixes only for
121+
**critical** issues, for 12 months after a successor edition ships.
122+
123+
| Artifact | Currently supported |
124+
|--------------------------------------------------|----------------------|
125+
| GenAI Data Security Risks and Mitigations | 2026 v1.0 |
126+
| LLM and GenAI Data Security Best Practices | 2025 v1.0 |
127+
| `dsgai-bench` | latest tag on `main` |
128+
| `dsgai_scanner_tool` | latest tag on `main` |
129+
| Crosswalk Explorer | `main` |
130+
131+
## Dependencies & Supply Chain
132+
133+
- Direct dependencies are pinned (`requirements.txt`, `package-lock.json`).
134+
- We monitor advisories via GitHub Dependabot and act on `high`/`critical`
135+
alerts within the SLA above.
136+
- Releases are produced from signed, tagged commits on `main`. Verify
137+
integrity using the SHA-256 hashes (and, once available, signatures)
138+
published with each release.
139+
140+
## Acknowledgements
141+
142+
We thank the researchers who help keep this project — and the wider GenAI
143+
security community — safer. Credited reporters are listed in
144+
[`SECURITY-THANKS.md`](./SECURITY-THANKS.md).
145+
146+
---
147+
148+
*This policy is published under [CC BY-SA 4.0](./LICENSE) and is itself open
149+
to community review — open a PR to suggest improvements.*

0 commit comments

Comments
 (0)