AI-Driven Security Research

Finding Critical Vulnerabilities
Before They're Found

ShellFoundry combines autonomous AI agents with expert-led security research to discover, verify, and responsibly disclose vulnerabilities in critical infrastructure — protecting the platforms you depend on.

shellfoundry — research status

$ shellfoundry --research-status

[OK] Programs Actively Monitored: 5

[OK] Vulnerabilities Discovered & Verified: 6

[OK] Reports Submitted to HackerOne: 4

[WARN] 1 Duplicate (first-to-report claimed — still validated our methodology)

[INFO] AI Agent Uptime: 99.9% · Automated Scans/Hour: 47

$ _

6

Verified Vulns

5

Active Programs

4

HackerOne Reports

47

Scans/Hour

Featured Research

Verified vulnerabilities discovered through autonomous AI-agent analysis and manual expert verification.

HIGH SEVERITY

Stolen Session Replay Enables Unauthorized Financial Transactions

Web Application — Improper Authentication (CWE-287)

Target

Major consumer identity verification platform

Weakness

CWE-287: Improper Authentication — Session credentials replayable from any IP

Method

Browser-assisted proxy capture (mitmproxy) + credential replay from remote infrastructure

Status

Reported via HackerOne Bug Bounty Program

The subscription creation endpoint on a major identity platform accepted state-changing financial operations using session credentials that were captured once from browser traffic and replayed from a completely different IP address. A single HTTP request with an empty body triggered a real financial transaction (confirmed via customer support refund). The vulnerability arose because the client-generated correlation header — containing only browser-produced UUIDs — was the sole mechanism for session binding, with no server-side validation that the correlation values matched the authenticated session. An attacker who captures a user's traffic on public WiFi, via a malicious extension, or through physical device access can create paid subscriptions charged to the victim's stored payment method.

Session Replay CWE-287 CSRF Bypass API Security
HIGH SEVERITY

Hardcoded Cryptographic Signing Keys

FROST Threshold Signature Scheme — Operator Private Keys

Target

Major cryptocurrency infrastructure provider

Weakness

CWE-321: Use of Hard-coded Cryptographic Key

Method

AI-agent automated source code analysis + cryptographic verification

Status

Reported via HackerOne Bug Bounty Program

Five valid secp256k1 private keys were discovered hardcoded across multiple locations in a public source repository. These keys serve as the cryptographic identity of network signing operators in a FROST threshold signature scheme — compromise of any two keys grants majority signing control. Each key was mathematically verified against its expected public key using standard ECDSA derivation.

FROST secp256k1 Threshold Signatures Static Analysis

Additional Findings

LOW

Hardcoded Infrastructure Credentials

Default Bitcoin RPC password discovered in docker-compose configuration and test files. Overrideable via environment variable but exposed in public repository.

LOW

Missing Cryptographic Memory Zeroing

Private key bytes in a cross-platform cryptography library remain in process memory after deallocation. No Drop implementation clears sensitive material — keys could leak via crash dumps.

INFO

Internal Infrastructure Exposure

Staging environment URL and insecure TLS connection factories (NoVerifyTLS / NoTLS) exposed in SDK test configurations across multiple language bindings.

Security Research Services

Autonomous AI agents augment expert-led research to find what manual reviews miss.

Source Code Audit

Deep static analysis of public and private repositories. AI agents scan for hardcoded secrets, cryptographic weaknesses, access control flaws, and insecure defaults across the entire dependency tree.

Bug Bounty Hunting

Continuous monitoring of in-scope programs with automated reconnaissance, attack surface mapping, and targeted exploitation testing. Findings verified and responsibly disclosed via HackerOne.

Red Team Assessments

Full-scope adversarial simulation targeting web applications, API infrastructure, smart contracts, and cloud deployments. AI agents accelerate reconnaissance while human operators drive exploitation.

How We Work

1

Automated Recon

AI agents scan repos, APIs, and infrastructure

2

Cryptographic Proof

Every finding is mathematically verified

3

Expert Review

Human-led verification and impact assessment

4

Responsible Disclosure

Reported through proper bug bounty channels

About ShellFoundry

ShellFoundry is an autonomous security research company built at the intersection of AI agent technology and offensive security. We believe that the best defense is knowing what the adversary knows — and finding it first.

Our approach combines persistent AI agents that continuously analyze source code, API surfaces, and infrastructure for vulnerabilities, with expert human review that validates every finding and assesses real-world impact. This hybrid model lets us scale reconnaissance while maintaining the judgment that separates a true vulnerability from a false positive.

Every vulnerability we discover is responsibly disclosed through established bug bounty programs (HackerOne, Immunefi). We protect the platforms, protocols, and infrastructure that power the modern digital economy.

Secure Your Infrastructure

Have a codebase, protocol, or platform that needs a deep security review? We'd like to hear from you.