About Services Pricing Case Studies Blog Contact Book Free Audit →

What we find when we test AI agents

Anonymized findings from real GaleOps engagements. Names, industries, and identifying details removed.

Critical

Indirect Prompt Injection via RAG Pipeline

Fintech SaaS · MCP Security Assessment

A customer support agent retrieved documents from a shared knowledge base. An attacker planted a system-override instruction in a public-facing document. The agent exfiltrated the user's session token via the send_email tool.

OWASP LLM01 Indirect Injection Tool Misuse
Read the full teardown →
Critical

MCP Server Default Permissions

AI Agent Platform · MCP Security Assessment

An MCP server shipped with all tools registered as publicly callable — no authentication, no rate limiting, no confirmation boundaries. An attacker could chain read_file + send_email to exfiltrate any file on the host.

OWASP ASI06 MCP Protocol Privilege Escalation
Read the full teardown →
High

System Prompt Extraction via Verbose Error

B2B SaaS · AI Security Audit

An AI agent returned its full system prompt in a verbose error message when given a malformed input. The prompt contained API keys, database schema details, and internal tool naming conventions — enough for a targeted attack.

OWASP LLM02 Data Leakage Credential Exposure
High

Tool Permission Escalation Chain

Healthcare AI · AI Security Audit

An agent with read-only document access could chain search_docs → summarize → send_email to exfiltrate PHI. No confirmation boundary existed between reading data and sending it externally.

OWASP ASI07 Tool Chaining Data Exfiltration
Medium

Jailbreak via Persona Override

E-commerce Chatbot · Founding Risk Review

A DAN-style persona override bypassed the agent's content safety guardrails. The agent generated harmful product recommendations and exposed internal pricing logic that was never meant to reach customers.

OWASP LLM05 Jailbreak Persona Bypass
Medium

Unsafe Output Handling in Code Generation

Developer Tooling · AI Security Audit

An AI coding assistant generated SQL queries and shell commands that were executed without sanitization. An attacker could inject malicious code through a crafted prompt that persisted in the generated output.

OWASP LLM05 Code Injection Output Safety

Published findings & methodology

Detailed writeups from bug bounty research and public vulnerability disclosures.

HIGH

Khan Academy VDP — Information Disclosure

HackerOne · June 2026

My first HackerOne report landed as a HIGH-severity information disclosure on a major education platform. The methodology applies to almost any AI-enabled site.

Bug Bounty Information Disclosure AI Recon
Read the full report →
Medium

Zest Protocol V2 — Smart Contract Methodology Review

Immunefi · June 2026

Downloaded 13 Clarity 4 contracts from a $100K bug bounty program, generated 8 hypotheses, tested the 3 highest-priority ones, and refuted all at the code level. A high-quality negative result.

Smart Contract Methodology Clarity
Read the methodology →
Critical

MCP Server — CVE-2025-6515

CVE Disclosure · August 2026

Found a vulnerability in a production MCP server in under 10 minutes. Tool poisoning, session hijacking, and the 3-line fix that blocks it all.

CVE MCP Prompt Injection
Read the teardown →

Want to know what we'd find in your AI agent?

Book a free 15-minute fit call. We'll scope your agent and recommend the right starting point.

Book a Free Fit Call →