MahaGuardian gives every AI agent in your practice a credential that opens exactly one engagement — one client, project or deal — and nothing else. The barrier is in the credential, not in the prompt.
Your AI agent can reach every client file your staff member can reach. An agent working on one engagement can read — and incorporate — data from any other engagement the same person has access to.
The firm finds out when the other client does. MahaGuardian closes this gap at the credential layer, before the content can reach the draft.
Every agent carries a credential that opens exactly one engagement's information. There is no key on that credential for any other engagement. The agent is not asked to be careful — it holds a key that does not open other matters through the Guardian-controlled boundary.
Every retrieval the agent performs through MahaGuardian is recorded in a hash-chained log on your firm's infrastructure. Each agent operates under its own identity, distinct from the staff member's, so the agent's actions never blur into the staff member's own activity.
For actions your firm has flagged as sensitive — external communications, sharing privileged material, accessing information above an agent's normal scope — the agent must request approval from a named human before proceeding. The request and the approval are both recorded.
An agent becomes dangerous when it combines private data, untrusted content and a way to send data out. MahaGuardian does not stop untrusted content reaching the agent — it limits what that content can achieve. The credential opens one engagement, so only one client's data is within reach; external sends can require a named approver; every attempt is recorded.
Signing keys and audit records stay with your firm, and client data never passes to MahaGuardian. Per-engagement access supports GDPR data protection by design and appropriate security (Articles 25 and 32). The audit trail and human-approval step are designed to support the EU AI Act's record-keeping and human-oversight requirements (Articles 12 and 14).
MahaGuardian sits alongside Microsoft, Google Workspace, Okta, iManage, and the rest of your security infrastructure — not instead of them. It decides whether an agent credential comes into existence and can destroy it; your platform decides what that credential can do once it exists.
Guardian authority records and your platform's own activity logs can be reconciled, so a credential issued without firm authority is detectable by the firm itself. The signing keys never leave your infrastructure.
MahaGuardian works with most AI agents that interact with your systems through standard APIs — OpenAI, Anthropic, Google, self-hosted models. If your firm switches vendors next year, you do not re-architect your information barriers.
MahaGuardian can sit underneath agent platforms, so engagement-scoped barriers travel with your agents whichever model they use.
MahaGuardian was built by Alexander Landia, an investor and director of several companies, including Lambert Energy Advisory and The Mobility House, with 30 years of M&A and strategy work. It began with a case: on an energy-transition mandate, one adviser legitimately held access to engagements that had to stay apart, and an AI agent working for him could reach all of them. He developed it at the MIT Media Lab, where it was selected as a top project in the Agentic Web course.
Whether you're evaluating, building, or just thinking about agent governance — a direct conversation is the best way to start.
Get in touch