How we operate, ethically and legally.
Labs engagements are designed to reduce breach exposure without creating new exposure. This page names the operating boundaries, data-handling rules, sub-processors, and legal limits we expect a serious buyer to inspect before signing.
Trust starts with constraints.
01
Authorization required.
We never run discovery against a surface we have not been explicitly authorized to investigate. Engagement contracts include a written authorization clause that names the root domains, GitHub organizations, Docker namespaces, NPM scopes, and adjacent public artifacts in scope.
02
OSINT-only methodology.
We use public sources: public Docker registries, NPM, GitHub public artifacts, CT logs, NVD, Wayback Machine, Sourcegraph public, archived dotenv aggregators, public pastebins, and similar. We do not probe, scan, brute-force, exploit, or test credentials against infrastructure.
03
Responsible disclosure.
For findings outside paid engagements, we follow coordinated disclosure: private notification, reasonable validation support, patch window, then public writeup only when publication is safe and approved by context. Active exploitation shortens the timeline.
04
Evidence minimization.
Discovered evidence is minimized at collection time. Plaintext secrets are never persisted. Reports use hashes, truncated previews, paths, timestamps, registry manifests, and screenshots that prove exposure without widening it.
Evidence without unnecessary retention.
The engagement produces an evidence-grade artifact. It does not require us to keep the dangerous material itself. We capture enough to prove exposure, business impact, remediation, and closure.
Artifacts are encrypted at rest using AES-256-GCM on dedicated EU infrastructure.
Plaintext credentials are not copied into reports, prompts, tickets, email, or chat.
Evidence packs use hashes such as sha256:abc12345... and truncated previews such as sk_live_abcd...
Raw evidence is deleted 30 days after engagement closure unless the contract requires shorter retention.
The final report and attestation are transferred to the client through an agreed secure channel.
LLM challenge prompts are sanitized before submission and logged as sanitized challenge records.
The current processing chain.
We keep the list public so prospects can review it before legal review. LLM providers receive sanitized artifacts only: no plaintext secrets, no live credentials, no unnecessary client labels.
| Provider | Purpose | Controls |
|---|---|---|
| Anthropic | LLM compute | Sanitized prompts only; secrets, URLs, and IPs replaced or hashed. |
| Google AI | LLM compute | Sanitized prompts only; used for independent challenge passes. |
| OpenAI | LLM compute | Sanitized prompts only; business account controls where available. |
| Resend | Transactional email | Booking, contact, and operational email routing. |
| Cal.com | Booking | Discovery-call scheduling and pre-call questions. |
| Stripe | Invoicing | Stripe Invoice payments when wire transfer is not used. |
| Self-hosted Umami (analytics.bleedwatch.com) | Privacy-first analytics, run by BleedWatch on its own EU infrastructure | Cookieless, no PII, no cross-site tracking, no third-party processor. Aggregate page views and event analytics only. |
GDPR / RGPD posture.
BleedWatch SASU is the data controller for personal data processed during engagement qualification and delivery. Client-controlled engagement data is handled under the contract and the data-processing terms agreed during signature.
DPO contact: [email protected]. Data subject rights are honored within 30 days. Transfers to non-EU recipients require appropriate contractual safeguards.
Rights we support.
Work we will refuse.
We refuse work that requires violating CFAA, Computer Misuse Act, RGPD, or equivalent jurisdictional law.
We do not accept engagements aimed at competitors, former employers, journalists, activists, or private individuals.
We do not run intrusive vulnerability scanning unless a future written scope says so. Labs v1 is OSINT-only.
We do not retain discovered secrets as proof. The proof is the evidence chain, not the secret value.
We do not publish client-linked findings without explicit written permission after remediation.
Controls that keep the engagement quiet.
Labs does not need broad internal access. The risk is evidence handling, not production access. The controls below are designed around that fact.
Least-privilege access to engagement workspaces.
Separate client folders and encrypted evidence storage.
No client evidence in personal chat tools.
No plaintext secret values in model prompts.
No public screenshots of engagement material.
Secure transfer channel agreed before report delivery.
Professional indemnity.
Professional indemnity insurance held; details available on request. Founder must confirm provider and coverage amount before final public launch.
Reciprocal NDA.
Prior engagement references and sensitive research identifiers are available after the discovery call under reciprocal NDA. We do not publish client logos without permission.
Report issues in our systems.
If you believe you found a vulnerability in labs.bleedwatch.com or bleedwatch.com, email [email protected]. PGP fingerprint remains a launch placeholder.
What legal should expect to review.
01
Statement of work with surface boundaries.
02
Written authorization clause for OSINT-only discovery.
03
Mutual NDA or reciprocal confidentiality language.
04
Sub-processor and data-handling appendix.
05
Retest and attestation language for closure.
Common pre-signature questions.
Can we review prompts before model use?
Yes. For sensitive buyers, we can share the sanitized prompt template and agree on redaction rules before evidence reaches any model.
Can we require shorter retention?
Yes. The 30-day deletion window is the default. A stricter contract window can override it when the engagement requires that.
Can legal approve the authorization clause?
Yes. The engagement does not start until both sides agree on authorization, surface boundaries, confidentiality, and disclosure handling.
Can references be verified without public logos?
Yes. References are handled under reciprocal NDA and direct introduction. We do not turn them into a public testimonial wall.
Need the legal packet before a call?
Send the review requirement and the jurisdictional concerns. We will answer directly, and if a concern changes scope, we will say so before you spend procurement time.