OpenAI debuts Aardvark, a GPT-5-powered agent for cybersecurity research

OpenAI just dropped its advanced AI assistant Aardvark, a new agentic security researcher powered by GPT-5. Released on Thursday, the powerful agent can assist security teams by identifying and helping patch vulnerabilities.

According to OpenAI’s blog post, the agent’s purpose is to address current issues in software security, given that tens of thousands of new vulnerabilities are identified annually in both enterprise and open-source code. OpenAI indicated that Aardvark originated as an internal tool to assist its own developers.

Matt Knight, VP at OpenAI, told ZD Net that the company’s developers found real value in how clearly it explained issues and guided them to fixes. That signal clears that OpenAI is on the path to something meaningful.

What Aardvark is?

Aardvark is an agent designed to identify, explain, and help resolve security vulnerabilities within a connected repository. It accomplishes this through LLM-powered reasoning and tool use, following a unique, multi-stage approach.

According to Knight, Aardvark examines the repository to understand the codebase’s purpose and its security implications, including objectives and design. Subsequently, it searches for vulnerabilities by analyzing past actions and recently committed code. As it scans, it clarifies the vulnerabilities discovered by annotating the code, making them accessible for human review and remediation.

Aardvark will then attempt to verify the existence of a susceptibility by placing it in a sandboxed environment, where it will attempt to trigger it. The results are then marked with metadata that can be used to filter and explore deeper.

Finally, Aardvark assists defenders in fixing identified vulnerabilities by utilizing OpenAI’s agentic coding assistant, Codex. It offers users a patch, generated by Codex and scanned by Aardvark, for human review and implementation.

How to access

Aardvark is only available in private beta to select partners who are invited by OpenAI to participate. As the tool is in the beginner phases, OpenAI said it will use participants’ feedback to filter the entire experience, working with the team to enhance detection accuracy, improve validation workflows, and provide additional benefits.