Zero Day Security tests large language model (LLM) applications, AI chatbots, RAG pipelines and AI agents the way real attackers do. Every engagement is led by an HTB Certified Offensive AI Expert (HTB COAE) who also holds CREST CRT and OSCP certifications.
Direct and indirect prompt injection, jailbreaks and system prompt extraction, including malicious instructions hidden in documents, emails and web pages your AI reads.
Whether your model, chatbot or RAG pipeline can be tricked into revealing customer data, internal documents, API keys or other users' conversations.
Access control on retrieved documents, knowledge base poisoning and cross-tenant data exposure in retrieval-augmented generation (RAG) systems.
Abuse of agent tools, plugins and MCP integrations: excessive agency, unsafe actions, SSRF and privilege escalation through the model.
XSS, injection and business logic flaws where LLM output flows into your web app, APIs or downstream systems without validation.
Findings mapped to the OWASP Top 10 for LLM Applications and MITRE ATLAS, with reproducible prompts, risk ratings and clear remediation guidance.
Questions about testing your AI features? Email info@zerodaysecurity.net or book a free scoping call.
It is a security test of applications built on large language models, such as chatbots, copilots, RAG search and AI agents. It covers attacks unique to AI, like prompt injection and data leakage, as well as the web application and API weaknesses around the model.
We follow the OWASP Top 10 for LLM Applications: direct and indirect prompt injection, sensitive data and system prompt leakage, insecure handling of model output, excessive agency and tool abuse, RAG and data poisoning, and unbounded resource consumption.
No. We test the way an attacker would, through your application and APIs. For deeper coverage you can optionally share system prompts, tool definitions and an architecture overview, which helps us find issues faster and confirm how data flows to and from the model.
Yes. Agents that call tools, APIs or MCP servers can be tricked into taking actions or leaking data they should not. We test tool permissions, indirect prompt injection through documents, emails and web content, data exfiltration through tool calls, and whether a human approval step can be bypassed.
Testing is performed by a Hack The Box Certified Offensive AI Expert (HTB COAE) who also holds CREST CRT and OSCP, so both the AI-specific attacks and the underlying web application, API and cloud weaknesses are covered in a single engagement.
AI testing is scoped and quoted at a fixed price after a free scoping call, based on the number of AI features, agents and integrations in scope. Every engagement includes a free retest. See our Pricing page for current packages.
Copyright © 2026 Zero Day Security - All Rights Reserved.
CREST CRT, OSCP, OSWA, OSWP, OSAI, COAE certified