Mindgard
London AI red teaming platform that attacks models and agents the way an adversary would, then retests the fixes
Quick Overview
| Company | Mindgard Ltd |
|---|---|
| Category | AI Security Platforms |
| Headquarters | London, United Kingdom |
| Founded | 2022 |
| EU Presence | Yes - United Kingdom |
| Data Location | Not stated on the pages reviewed |
| Open Source | No |
| GDPR Compliant | Yes |
| Self-Hosting | Not stated |
| Main Features | AI recon, automated AI red teaming, remediation guidance, continuous assurance, CI/CD and Burp Suite integration, runtime protection, AI governance and compliance |
| Pricing | Pricing on request (demo-led) |
| Best For | Security teams that want evidence of what an attacker could do to a model or agent before release |
| Replaces | HiddenLayer AI attack simulation, manual AI penetration testing |
Detailed Review
Mindgard starts from the attacker side. The platform maps what an AI system or agent can reach, runs attacks against it and reports which weaknesses are actually exploitable, instead of listing theoretical risks.
The company states it was founded in 2022 on research by Dr Peter Garraghan at Lancaster University. Its terms name Mindgard Ltd with a registered office in London and are governed by the laws of England, Northern Ireland, Scotland and Wales.
Recon first, then attack
The workflow has four named steps: Recon, Attack, Remediate and Assure. Recon discovers what the AI can do, Attack finds the risks that matter, and Remediate takes action on the gaps and validates the improvement.
Because the tests are repeated whenever the AI system changes, the same run can serve as a release gate and as an audit trail.
Research published in the open
Mindgard lists more than 150 publicly disclosed AI vulnerabilities, with write-ups on its disclosures page covering products such as Google Antigravity, OpenAI Sora and xAI Grok. That output feeds the attack library inside the platform.
A buyer can read those disclosures before talking to sales, which is a rare way to judge a security vendor in this category.
Where it plugs in
Deployment is described as APIs, CI/CD pipelines, Burp Suite or a one-click workflow. The FAQ also positions the product against the open source tools Garak, PyRIT and Promptfoo, arguing that a managed platform saves the effort of assembling and maintaining several scripts.
Limitations to Consider
Mindgard publishes no prices and sells through demos. Hosting region, self-hosting and certifications were not stated on the pages reviewed. Runtime protection is mentioned as something deployed "where needed", so teams that mainly want an inline prompt filter should compare NeuralTrust and Lakera, which centre on runtime enforcement.
Who Should Use Mindgard
Application security and red team leads in Europe who need repeatable, evidence-backed testing of LLM apps and agents, from a UK-incorporated vendor.
Pros and Cons
Pros
- Attack-led workflow with a retest step
- More than 150 public disclosures to inspect
- CI/CD, API and Burp Suite entry points
- Research origin at Lancaster University
- UK company with English-law terms
Cons
- No public pricing
- Data location and self-hosting not stated
- Runtime defence is secondary to testing
- Demo required to see the product
Alternatives to Mindgard
Other European ai security platforms worth comparing before you decide:
NeuralTrust
Spain · Barcelona agent gateway and runtime guard with free tier
Advai
United Kingdom · UK AI assurance: choose, test and monitor AI systems
Giskard
France · Paris open-core red teaming and testing for AI agents
QuantPi
Germany · Saarbrücken validation engine with statistical confidence
See all European ai security platforms or compare them directly against HiddenLayer AI attack simulation on our alternatives page.
Frequently Asked Questions
Mindgard says it was founded in 2022 on AI security research by Dr Peter Garraghan at Lancaster University, and its terms name Mindgard Ltd in London.
Not primarily. Its centre is discovery and red teaming, with runtime protection offered as a follow-on. A gateway such as NeuralTrust covers inline enforcement more directly.
Mindgard argues that open source frameworks give technical teams building blocks, while its platform adds managed attack intelligence, remediation tracking and reporting without assembling the tooling yourself.