Mindgard is an AI security company that attacks its customers’ AI systems before real attackers do. Its offensive security platform finds the models, agents and connected tools an organisation runs, red teams them automatically, and helps security and engineering teams fix what it uncovers.
The Mindgard AI Security Platform is used across development and production, so AI risk is tested before launch and watched after it.
At A Glance
| Vendor | Mindgard |
|---|---|
| Category | AI security, automated AI red teaming |
| Identities covered | AI models, agents, MCP and A2A servers, connected tools |
| Agentic AI | Agentic red teaming, agent native reconnaissance, runtime protection |
| Certifications | SOC 2 Type 2 |
| Funding | Series A led by Album VC (August 2026) |
| Market classification | AI Model Security; also AI Safety & Guardrails |
What Mindgard Does
AI agents bring a new attack surface, because everything an agent is configured with or allowed to touch can be turned against the organisation running it. Mindgard’s mission, “Secure the World’s AI”, translates into testing those systems the way an attacker would. The platform looks for ways someone could manipulate an agent, reach sensitive information, misuse a tool or slip past a control. Red teaming sits at the centre, and it is itself agentic: attacks run continuously and adapt, following an agent’s permissions and tools to see how far a compromise would reach. Because systems are retested when they change, the result is an ongoing view of AI risk rather than a one-off audit.
Products
Recon
Recon maps the AI attack surface. It discovers models, agents, MCP and A2A servers, connected tools and shadow AI, and picks out the instructions, tools and behaviours an attacker could exploit. Security teams that do not know how much AI is running across the business start here.
Attack
Attack is the automated red teaming engine. It chains weaknesses into realistic attack paths and tests for prompt injection, jailbreaks, system prompt leakage, unsafe tool use and excessive agent permissions. Red teamers can add their own techniques alongside the built-in ones.
Remediate And Assure
Remediate helps teams close defensive gaps and decide which fixes matter most. Assure then checks continuously that those fixes hold and supports compliance, producing evidence-backed reports that engineering, security and audit teams can each use.
AI Threat Response
AI Threat Response protects AI systems once they are live, detecting and responding to threats against models, agents and applications in production. AI Governance and Compliance features sit alongside it, so policy and runtime protection are handled in the same platform. For organisations without in-house AI security skills, Mindgard also offers AI red teaming as a service.
How It Works
Customers connect their AI systems to the platform through APIs and integrations, including CI/CD pipelines, Burp Suite or a one-click workflow. From there the platform runs a repeating loop across development and production environments.
| Component | What it does |
|---|---|
| Assess | Automated discovery, followed by reconnaissance, safety and adversarial evaluations. |
| Analyse | Measures risk exposure and the likely impact of each finding. |
| Route and fix | Sends findings into ticketing and other security and engineering workflows, ranks fixes and deploys protections. |
| Reassess | Retests systems whenever they change. |
Identities And Systems Covered
| AI models | Open source and managed models |
|---|---|
| AI agents | The model, instructions, memory, permissions, data, APIs and tools that make up an agentic system |
| Agent protocols | MCP and A2A connected systems and servers |
| Applications | RAG applications and tool-using agents |
| Shadow AI | Unmanaged AI usage across the organisation |
Company Background
| Founded | 2022, on research by Dr. Peter Garraghan at Lancaster University |
|---|---|
| Headquarters | Boston and London |
| Leadership | James Brear, Chief Executive Officer; Dr. Peter Garraghan, Founder and Chief Science Officer |
| Funding | A seed round in 2024, then a $30m Series A announced in August 2026 led by Album VC with Karma Ventures, .406 Ventures, Atlantic Bridge, IQ Capital and Lakestar, for a reported total of close to $42 million |
| Research | More than 150 publicly disclosed vulnerabilities in AI products, including a zero-day code execution flaw in Cursor IDE, with research covering Grok, ChatGPT, Google Antigravity, OpenAI Sora and the Zed IDE |
Note: This overview is compiled from Mindgard’s publicly available information as of 3 November 2026. It is provided for information only. NHI Mgmt Group does not endorse any vendor, and nothing on this page represents the opinions of NHI Mgmt Group.