Mindgard

Automated AI red-teaming and security testing platform — a Lancaster University spinout that attacks your models, agents, and AI apps to find AI-specific vulnerabilities before attackers do.

One-liner — Offensive AI security: continuously red-teams your LLMs and AI applications for jailbreaks, prompt injection, data leakage, and model-level vulnerabilities.

Categories — ai-red-teaming

What it does

Mindgard runs automated attacks against AI systems — large language models, multimodal/image models, agents, and the apps wrapping them — to surface AI-specific weaknesses (jailbreaks, prompt injection, evasion, extraction, data leakage) that traditional application security scanners miss. It positions as a testing/assurance layer: you point it at a model or deployed app, it generates and runs adversarial test campaigns, and reports exploitable findings. The thesis from its founding research is that “traditional AppSec could not address AI-specific risks.”

Where it sits in the stack

ai-red-teaming in the model/prompt layer. Its primary focus is untrusted input — it probes how the model handles adversarial/malicious input. It is a pre-production and continuous-assurance tool rather than an inline runtime control, so it complements (not replaces) an ai-runtime-security firewall that blocks attacks live.

Deployment & architecture

SaaS platform with API/automation hooks for continuous testing in CI/CD. Targets both models and live applications. Exact integration surface (SIEM, gateways) not fully documented on public pages — see open questions.

Positioning & differentiators

Known as an academically-rooted, red-teaming-first vendor (spun out of Dr. Peter Garraghan’s research at Lancaster University). Nearest neighbors are other red-teaming/guardrail testers: splxai, haize-labs, promptfoo, patronus-ai, and the testing side of hiddenlayer and enkrypt-ai. Unlike the eval-platform players (patronus-ai, maxim-ai) that center on quality/ hallucination metrics, Mindgard leads with offensive security testing.

Ownership, funding & M&A

Independent and VC-backed. Founded 2022. Raised ~$11.6M total, including an $8M round in December 2024 led by .406 Ventures (with Atlantic Bridge, WillowTree Investments; earlier IQ Capital, Lakestar). No seed M&A flag and no acquisition found — ownership confirmed independent (confidence high). James Brear joined as CEO in 2025; Garraghan is Chief Science Officer/founder.

CTO / hedge-fund lens

Day-2 and niche for a hedge fund. You only need a dedicated AI red-teaming vendor if you are building and shipping your own LLM applications/agents and want adversarial assurance before release. A fund that mostly consumes enterprise AI assistants (ChatGPT Enterprise, Copilot) gets little from this; the controls that matter there are access governance and DLP. Some SR 11-7 / model-risk relevance as evidence of adversarial testing, but lighter-weight eval tools or a pen-test engagement may suffice.

Competitors / alternatives

splxai, haize-labs, promptfoo, patronus-ai, hiddenlayer, enkrypt-ai, mindgard’s testing overlaps with cisco-ai-defense validation.

Open questions / to verify

• Exact current funding total and any 2025–2026 round.
• Deployment specifics: agent/CLI vs API-only; SIEM/gateway integrations.
• Whether it offers any runtime/inline protection or is testing-only.

Sources

• About Mindgard — fetched 2026-06-28 — supports: what it does, founding 2022, HQ, founders, CEO; confidence: med (vendor marketing).
• Mindgard — Crunchbase — fetched 2026-06-28 — supports: $8M 2024 round, .406 Ventures, ~$11.6M total; confidence: med (aggregator).

History

• [2026-06-28] Stub created from seed registry.
• [2026-06-28] Researched; established independent VC-backed (Lancaster spinout, 2022, London/Boston, ~$11.6M, $8M Dec-2024 .406 Ventures); no acquisition. Set ownership_confidence high, hedge_fund_fit low.