SI Agent News
DevelopingClaudeOpenAIGeminiPolicy

Delinea's Fastpath adds a read-only AI connector for access audits

Delinea's Fastpath unit lets finance and audit staff ask AI assistants plain-language questions about who can access what. It is in private preview, with general release planned for November.

23tools exposed in the first, read-only release

The short version

Why it matters

Access reviews and audit prep are slow, manual work, and AI assistants are now being wired directly into that data. This launch shows one cautious pattern: start read-only, run as the signed-in user, log everything. If your company plans to connect AI to sensitive systems, those are the questions to ask any vendor.

What Delinea Fastpath announced

Delinea's Fastpath business, which makes software for governing who has access to which business applications, has launched a Fastpath MCP Server. As MSDynamicsWorld.com reported, the aim is to let finance, audit and IT staff ask an AI assistant everyday questions about audit readiness and access risk. The article is marked as sponsored by Delinea Fastpath, so the claims below are the company's own.

MCP stands for Model Context Protocol. It is an open standard, a kind of common plug, that lets AI assistants connect to outside software and pull in its data. Here, the outside software is Fastpath's access-governance records.

The product is not yet widely available. It is in private preview with a select group of customers who run it against their own Fastpath accounts. The company says general availability is planned for November.

The problem it is meant to ease

Fastpath says finance teams are struggling to keep up with thousands of segregation-of-duties conflicts raised by the security models in ERP systems. ERP software is the central system a company uses to run finance, supply chains and similar functions. Segregation of duties means that no single person should hold a combination of permissions that would let them, say, both create and approve a payment.

The company describes the old routine as a pile of manual work. An analyst would pull one large report on the risks, then another with details on how each was being handled. After that came exports, a slide deck and a remediation plan assembled by hand.

The pitch is to swap that routine for a conversation. An assistant can answer questions, point out what needs follow-up and draft emails, all inside a chat window the team already uses.

How it works in practice

The first release is deliberately read-only: the assistant can look at data but cannot change anything. It offers 23 tools, meaning specific actions the assistant can call, covering Fastpath's Access Control and Access Review data. It draws on whatever modules and integrations a customer already has.

Fastpath supports more than 100 applications across ERP, CRM and HCM (human capital management) systems. Because the server sees every system a customer has connected, a user can ask about people, roles and access risk across all of them at once. For Microsoft Dynamics customers, that includes Dynamics 365 Finance and Supply Chain, Dynamics 365 Business Central, Dynamics GP and Entra ID. Where Fastpath already gathers deeper signals, such as usage telemetry and license reporting for the first two, the assistant can reach those too.

The server works with Microsoft Copilot, Claude, ChatGPT and Gemini Enterprise, plus other assistants that support MCP. Delinea VP of product management Chris Aramburu said preview customers are using it on live data. In his words: "They are asking real questions on real data."

The safety design

Connecting an AI assistant to access records raises an obvious worry about who can see what. Fastpath says its read-only design keeps things simple. The tool runs as the signed-in user, so it should only show what that person is already allowed to see, and that identity is checked twice.

The company also says it does not store credentials, and that every request is written to an audit log. It adds that its hosting environment is covered by SOC 1 Type 2 and SOC 2 Type 2 reports, which are independent audit standards for how a service provider handles controls and data. It also says no model training is involved.

These are vendor statements from a sponsored piece. Buyers will want to test them against their own security review.

What comes next

Aramburu said the preview includes enablement sessions and regular feedback calls, and that customer input will shape the general launch. Between now and November, the team plans to iterate, finish documentation and train its own staff to support the product. A webinar on November 10 will offer a 30-minute walkthrough.

The roadmap is broader. Fastpath plans to add starter prompts at launch, then extend the assistant to Change Tracking and Access Provisioning data. That would let it answer not only who has access, but also what changed, who approved it and what is still pending.

The company is also exploring actions that keep a person in charge. Examples include starting an access review, drafting an access request, or building a security model for simulation. In Aramburu's description, the assistant proposes and a person reviews or approves in Fastpath. Fastpath says these plans may change as feedback arrives.

What to watch

Try it

If your team is considering connecting an AI assistant to any internal system, ask the vendor three things: is access read-only, does it run as the signed-in user, and is every request logged?

Sources and links

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

Listen to the AI Wire headlines

Listen to the AI Wire

Listen to our daily AI shows every morning: on your commute, in the shower, wherever you are

Listen to the daily shows

First reported by MSDynamicsWorld.com · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

More AI news