Analysis
Mindgard closed a $30 million Series A led by Album VC, with Karma Ventures joining a syndicate of existing backers -- Boston's .406 Ventures, Atlantic Bridge, IQ Capital and Lakestar -- according to SecurityWeek. The round lifts the Boston- and London-based company's total funding to nearly $42 million, as enterprises scramble to secure the fast-growing layer of AI agents and coding tools now sitting inside their systems.
What Mindgard actually found
Mindgard's platform automates red-teaming for AI models and applications -- simulating the kind of adversarial probing a human security researcher would do, at a scale no security team can staff for manually. The company's research has already identified more than 150 vulnerabilities across widely used AI products, including a zero-day code-execution flaw in Cursor -- the coding tool SpaceXAI acquired for $60 billion in June -- along with defects in Google's Antigravity agent framework and ChatGPT itself. Founder Peter Garraghan, a professor at Lancaster University, started the company in 2022 on research that goes back more than a decade, which puts Mindgard ahead of most of the AI-security startups that only appeared once agentic coding tools went mainstream this year.
“- Onyx Security -- $113M Series B at a $640M valuation in July, also focused on AI governance rather than red-teaming specifically.”
A crowded, fast-moving category
Mindgard isn't alone chasing this budget line:
- Obsidian Security -- $85M Series D in early August at a valuation over $1.1B, targeting AI governance and access control.
- Onyx Security -- $113M Series B at a $640M valuation in July, also focused on AI governance rather than red-teaming specifically.
The category split is becoming clearer: Onyx and Obsidian sell visibility and control over which AI systems and agents have access to what; Mindgard sells the adversarial testing that tells a security team whether their model can be broken in the first place. Different layer, same budget conversation inside a CISO's office.
Why this round lands differently
The timing is notable. Z.ai's GLM-5.3, released the day before Mindgard's round closed, scored 84.5% on CyberGym -- a benchmark built specifically to test how well a model can find security vulnerabilities in real code. That's the same skill Mindgard sells as a defensive product. Offense and defense in AI security are advancing on the same curve, not with defense lagging years behind the way it did in traditional software security. A well-resourced attacker now has access to frontier-grade vulnerability discovery through a metered API; a security team buying Mindgard's platform is buying the same category of capability pointed at their own systems first.
Mindgard plans to use the new capital to scale product, engineering, sales and marketing, betting that enterprise demand for AI red-teaming keeps outpacing the supply of vendors who can do it credibly. The risk for any single vendor in a category this new is consolidation: enterprise security budgets tend to converge on platform incumbents once a category matures, and it's not yet obvious whether Mindgard, Onyx or a larger platform player ends up owning the AI-security suite by 2027.
