Mindgard

Mindgard is an AI security platform that automates red teaming of AI systems. It works as a dynamic application security testing tool for AI, testing LLMs, AI agents, computer vision, audio and multimodal models, and its command-line tool and GitHub Action need only an inference or API endpoint rather than access to model internals, pulling the latest attack techniques on each run. Image-model red teaming was announced in September 2026. The company raised a $30 million Series A led by Album VC in August 2026, bringing total funding near $42 million, while an earlier source puts it at about $11.6 million including an $8 million round led by .406 Ventures in December 2024, which predates the Series A. Vendor-reported claims differ on its research team's track record, more than 100 or more than 150 vulnerabilities disclosed in production AI products, and sources report SOC 2 Type II and GDPR compliance.

Mindgard's own pricing page, read directly, shows no prices, only Let's Talk, Book a Demo and Contact Us buttons, with no plan names or free trial, so treat it as contact sales. One comparison lists pricing as custom. Because testing is run against your endpoints, ask how a quote counts models, endpoints and test runs, and whether continuous testing in CI is included. The figures on vulnerabilities and attack library size are company-reported, not independently audited.

Mindgard suits security and ML teams that need repeatable adversarial testing of models and agents before and after release. Compare HiddenLayer, Lakera and XBOW. Browse the Cybersecurity tools directory for more.

Automated red teaming of LLMs and agents

DAST for AI.

Testing image and multimodal models

September 2026.

Pricing?

Custom, book a demo.

Recent funding?

$30M Series A in August 2026.

What does it test?

LLMs, agents, vision, audio, multimodal.

Need model access?

No, just an endpoint.