Methodology & independence

Most "best AI security tools" lists are written by vendors who rank themselves first. Agent Security Index was founded by Carlos Paredes, an independent practitioner, and has no ownership ties to any vendor listed.

How a tool gets listed

• It has to address a real AI agent or LLM security problem in one of our six categories.
• It needs an official website or public repository that we can open.
• Open source and commercial tools are listed side by side. Basic listings are always free.

How we verify facts

• Every field comes from the vendor's own site, its public GitHub repo, or a linked public source such as an SEC filing or an acquisition announcement.
• If we cannot confirm a fact, we leave it blank rather than guess.
• Each listing shows its last-verified date, and we re-check the whole directory on a rolling schedule.
• Acquisitions are recorded with a source link, because ownership changes fast in this market.

Lab testing

Lab results come from a repeatable harness of attack batteries, starting with:

• Direct prompt injection and jailbreaks
• Indirect prompt injection through documents, web pages and tool output
• MCP tool poisoning and tool-description attacks
• Secret and data exfiltration attempts
• Excessive-agency and unsafe tool-call scenarios

For each test we will publish the version tested, the configuration, the batteries run, pass and fail counts, and what we did not test. Vendors see results before publication so they can flag a setup error. They cannot block or edit results.

How we make money

Labeled featured listings, one newsletter sponsor per issue, paid independent tests, and possibly marked affiliate links. None of these affect rankings or results. See the disclosure & independence policy.