Security Due Diligence for Venture Capital

Know what you're buying before you buy it.

Enter a company. Get a security posture grade by morning — free.

8

Published advisories

6

Startups audited

31

Findings across advisories

The Problem

You're investing in software companies without reading the software.

01

Pitch deck security

Founders say “we take security seriously.” Their API says otherwise. Unauthenticated endpoints, leaked API keys, and zero rate limiting — invisible from a demo.

02

Post-investment surprises

A data breach six months after close doesn't just hurt the portfolio company — it's a headline with your firm's name attached. Due diligence should catch this before term sheets.

03

Technical debt is financial debt

Insecure architecture isn't a bug to fix later. It's a rewrite. The difference between finding it in due diligence and finding it in an incident response is when you look.

04

Gen AI is accelerating the problem

Startups are shipping faster than ever with AI-generated code. More features, more endpoints, more attack surface — with less review. The gap between what's deployed and what's been secured is growing every quarter.

Recent Probes

6 startups. Passive recon only. Most in under fifteen minutes.

No scanning tools. No authentication. No exploitation. Just a browser, curl, and publicly accessible endpoints.

01 Series C Fintech · 8 min

P2 Medium

Cloud function auth keys and admin API schema exposed in production JavaScript bundle

02 Series A Healthtech · 15 min

P2 Medium

API validates parameters before checking authentication — error response leaks internal configuration types

03 Series B EdTech · 15 min

P2 Medium

Third-party marketplace integration leaks internal staff emails and complete SSO provider architecture

04 Series B Proptech · 10 min

P1 High

Full GraphQL schema introspection enabled — complete type system exposed including tenant PII, payments, and smart lock access

05 Series A HR / Payroll · 10 min

P2 Medium

HSTS effectively disabled (max-age=0) — SSL stripping possible on platform processing salaries and bank accounts

06 Accelerator · 15 min

P2 Medium

Unauthenticated email enumeration endpoint with no rate limiting — confirms founder and partner affiliation

Verbatim — one of 6

# Series B Proptech · 10 minutes
$ curl -s -X POST "https://api.[redacted]/graphql" \
    -H "Content-Type: application/json" \
    -d '{"query":"{ __schema { queryType { fields { name } } } }"}'

→ Full schema returned. 675 operations total.
→ Tenant PII models, payment transactions, smart lock access.
→ _billsSkipAuthAndPagination — internal query named to bypass auth.
→ Node.js stack traces with server file paths in error responses.

# Auth required: none.

If this is what ten minutes of passive recon looks like, imagine what we find in forty-eight hours.

Published Advisories

We don't scan. We read.

These are real vulnerabilities we found in production systems — account takeovers, token theft, PII exposure. The same class of bugs that live in the VC-backed startups shipping code faster than they can secure it. Every advisory includes full reproduction steps and coordinated disclosure.

RDXS-2026-001 · 2026-03-07 · Claude Code 2.1.71

Critical

Supply-Chain Code Execution in Claude Code

A cloned repository can execute arbitrary shell commands on your machine the moment you run claude -p.

4 critical · 4 high · 8 total findings

RDXS-2026-002 · 2026-03-09 · Claude Code 2.1.71

Critical

Remote Bridge Trust Boundary Failures in Claude Code

A remote peer can bypass crash guards, trigger local OAuth fetches, and exhaust system memory before the operator approves any action.

2 critical · 1 high · 3 total findings

RDXS-2026-003 · 2026-03-10 · Wallet Extension 3.9.0

High

QR Login Session Hijacking in Crypto.com Wallet Extension

Unauthenticated QR session creation and missing creator/scanner binding allow an attacker to steal the approver's Bearer token and access account data.

0 critical · 2 high · 2 total findings

RDXS-2026-004 · 2026-03-10 · DoorDash Consumer App 15.221.7

Critical

OAuth Account Takeover in DoorDash Android App

Custom scheme redirect hijacking, no PKCE enforcement, and a hardcoded client secret enable full account takeover via authorization code interception.

1 critical · 1 high · 3 total findings

RDXS-2026-005 · 2026-03-11 · Claude Code 2.1.72

Critical

Remote Control Bridge Worker Chain in Claude Code

A malicious repository can steal the live session bearer from a Remote Control bridge worker, forge tool approvals, and persist machine-wide permission corruption that survives into future sessions.

6 critical · 1 high · 7 total findings

RDXS-2026-006 · 2026-03-11 · Claude Code 2.1.72

High

Remote Control Session Isolation Failures in Claude Code

Bridge bearer tokens are not session-scoped, plain OAuth can enumerate and control all active sessions, and organization UUID is not enforced — amplifying the blast radius of any stolen credential.

0 critical · 5 high · 5 total findings

RDXS-2026-007 · 2026-03-16 · gstack 0.4.4

Medium

Security Review of gstack: Y Combinator CEO's Claude Code Toolkit

Two server-level security issues in the headless browser daemon, plus a design review of trust boundaries between human intent and autonomous AI agent action.

0 critical · 2 high · 2 total findings

RDXS-2026-008 · 2026-03-31 · Claude Code Source leak snapshot

High

Live Session Bearer Disclosure to MCP Servers in Claude Code

An attacker-controlled MCP server can receive the live session-ingress bearer in legacy bridge paths because Claude Code promotes product auth into generic MCP transport headers.

0 critical · 1 high · 1 total findings

View all advisories →

What We Look For

The things that don't show up in a pitch deck.

We examine the same attack surface an attacker would — production APIs, client-side code, third-party integrations, cloud infrastructure — using the same tools. No checklist. No automated scanner. Manual, adversarial testing.

Authentication & Access Control

  • Unauthenticated API endpoints
  • Broken authorization on mutations
  • Session management flaws
  • Missing rate limiting

Data Exposure

  • PII leakage (phone, email, addresses)
  • Internal configuration exposure
  • Third-party API key leakage
  • AI training data extraction

Infrastructure Security

  • Cloud storage misconfigurations
  • Message queue injection
  • Presigned URL abuse
  • Service proxy exploitation

AI-Specific Risks

  • Voice cloning / deepfake generation
  • Prompt injection and data exfiltration
  • Model output manipulation
  • Training data poisoning vectors

How It Works

Forty-eight hours from engagement to deliverable.

No six-week timeline. No 200-page compliance document. A focused, adversarial audit of the external attack surface — the same endpoints your users (and attackers) can reach.

01

Scope & Reconnaissance

Hours 0–6

Identify the production API surface, client-side bundles, third-party integrations, and cloud infrastructure. Map every externally reachable endpoint.

02

Adversarial Testing

Hours 6–36

Manual adversarial testing of every discovered endpoint. Authentication bypass, authorization testing, data exposure probing, injection attacks, infrastructure abuse. Every finding verified with reproducible curl commands.

03

Attack Chain Analysis

Hours 36–42

Map how individual vulnerabilities chain together into real-world attack scenarios. A phone number leak alone is one thing — combined with voice cloning and an unauthenticated call API, it's a fraud kit.

04

Executive Report

Hours 42–48

Severity-ranked findings with CVSS scores, OWASP classifications, full reproduction steps, impact analysis, and prioritized remediation guidance. Written for both technical teams and board presentations.