Hello, X. We're SecureAi, an Irish team building Argus.
Argus continuously checks your Microsoft Copilot and Copilot Studio agents against your own written policy and gives each one a verdict: Ship, Ship with conditions, or Fix first.
Ignite made it clear that agents are now part of the workforce. Agent 365 is the control plane. Argus adds a daily, read-only check of every agent in your M365 tenant against your own policy, with zero changes. #MicrosoftIgnite#AISecurity
@xxxaarr@Microsoft This is the bit I'd want every CISO to read before budgeting for 2027. Microsoft-built agents get insights included, but third-party agents and advanced governance sit in a different tier. Map which of your agents fall on which side first, or you'll be governing half your estate.
@XavierRiveraX Useful change. Worth saying out loud, though: active users, sessions and error rates tell you an agent is busy, not that it's behaving. The next question for most security teams is whether each agent still matches policy: who owns it, what it can reach, who it's shared with.
BCG benchmarked 26 systemically important banks: 73% let AI agents into production. Only 8% have one standard framework to monitor and control them.
Policy is written. Operations aren't.
Your agent inventory: built from production, or from approvals?
#Banking#NexusOne
@amvnkr From production, every time. An approvals list tells you what was signed off. It doesn't tell you what's running today, who owns it, or whether it still meets the policy it was approved under. Agents drift: new connectors, wider sharing, owners who leave.
Big news for MSPs from Ignite: multi-tenant agent management in Agent 365 is in public preview. Managing agents across client tenants is now the job, and the evidence you keep for it matters. #MicrosoftIgnite#mspaint
Autopilot is rolling out to Frontier customers: an always-on agent with its own identity, memory and tools, working in Teams and Outlook without a prompt. Worth asking now what your policy says it can reach. #MicrosoftIgnite#MicrosoftCopilot
50 million agents registered in Agent 365 since May. That's the number from Ignite yesterday. The question for every CISO: how many are in your tenant, who built them and do they still meet your policy? #MicrosoftIgnite#AISecurity
I've written a one-page Frontier readiness checklist for Microsoft partners. It covers the specialisation requirements and the govern and operate evidence. Reply or DM me and I'll send it over. #MicrosoftPartner#MSP
A simple question for Copilot partners: can you name every AI agent in each client's Microsoft 365 tenant, who built it and what it can reach? If not, that's the first gap to close before Frontier. #MicrosoftCopilot#MSP
Microsoft partners: the Frontier badge retires at the end of June 2027. Can you show how every AI agent in your clients' tenants is governed and operated? That evidence takes months to build. Start now. #MicrosoftCopilot#MSP
@HeikoBrenn Good list. My one to watch: what Microsoft says about ownership when the maker leaves, and whether the sponsor/owner in Entra Agent ID and Agent 365 matches who actually answers for the agent in the business. The tooling is catching up fast. The RACI usually isn't.
@ragnarh Agree. The gap I see most: every team owns a piece and nobody owns the picture. Makers own the build, IT owns environments and DLP, security owns identity, compliance owns the policy. What ties it together is a named business owner per agent plus one shared verdict per agent
@donasarkar The 'who gave them access?' line is the one every admin will feel next week. The fix I keep seeing work: every agent gets a named owner, a list of what it can read and touch, and someone checking that it still matches policy a month after launch, not just on day one.
@AbidKhanHQ@krishnan Agreed. What auditors want is a dated record of each agent checked against the written policy (sponsor, permissions, knowledge sources, sharing), kept outside the team that builds them, and the trend over time. Disclosure: I work at SecureAi; Argus produces that record read-only.
@salvestad Useful map, thanks. One gap I keep seeing: approval is a point in time, but an agent's tools, knowledge sources and sharing keep changing afterwards. How are you thinking about re-approval when an approved agent drifts?
@ruadams Good summary. The flip side this week is Proofpoint's BluePhishing PoC: consent to a third-party blueprint and it can create agent identities in your tenant. Worth checking "Users can create agent blueprints" and reviewing consented blueprint principals now.
An AI agent inventory should tell you more than what each agent is called. For every agent in Microsoft 365, a security team should be able to answer five questions: Who built it? Who can use it? What does it connect to? Whose credentials does it run on? Who owns the risk today?
When someone leaves, IT disables their account. Their Copilot agents often keep running.
Add three lines to your offboarding checklist:
1. Reassign every agent they built or shared the same day
2. Check what each one can still reach
3. Retire anything nobody will own
@sharepointfocus This is the bit that slips. You update and re-share the skill, and the agents still on the old copy quietly keep doing the old work. Add "open every agent that uses it and take the update" to the change checklist, with a named owner on each.