Taking security back in-house: an MSP exit on a 30-day clock
An outsourced IT and SOC provider held admin access across identity, the password vault, device management, the network and monitoring, and its contract was about to end. I led the consulting team that took that access back in one weekend, without an outage and with an evidence trail an auditor can follow.
Facundo Montefalcone, security lead. Client and provider names and open findings are withheld.Context
A mid-sized US company, sister organization of a fintech client, had outsourced IT and security monitoring to one provider for years. The fleet is almost entirely macOS on Apple silicon. New leadership decided to bring security in-house.
The provider contract ended about 30 days after formal notice. Its device-management tenant, monitoring billing and compliance tooling ended with it. I was security lead on a 7-person consulting team in Q4 2026.
The problem
Nobody had the full list
Access had accumulated over years across about 27 business applications, 10 of them core. There was no single record of what the provider could reach.
Order decides the outcome
Rotate credentials in the wrong order and you lock out your own team, or leave a live session on the layer you have not reached yet.
Hidden dependencies
Cutting off a provider also cuts off what it quietly ran: the MDM tenant, log retention and compliance evidence.
How I ran it
- DiscoverAudited the shared password vault entry by entry. Years of history were split by period across the team so it finished in days. I used an LLM to categorize the export, then verified by hand, and built the application inventory alongside it.
- Access matrixOne sheet, one row per application: provider accounts, roles, how they authenticate, status. It became the single source of truth for the client and the team.
- Runbook and GanttA step-by-step rotation runbook with owners, order, rollback and a communication plan. Rule: credential changes are announced to their owners before rotation, never after.
- Cutover weekend7 engineers rotated admin credentials and MFA factors in parallel: vault administrators first, then identity provider, network, MFA recovery channels, then SaaS.
- Revoke and removeProvider accounts suspended and roles stripped under a 30-day soft-delete policy to preserve audit trails. Legacy remote-management agents uninstalled by MDM script, then verified by policy: 0 hosts remaining.
- Monitor and provePost-cutoff anomaly monitoring, a hunt for leftover persistence such as scheduled tasks, and a chain-of-custody log of 19 artifacts with SHA-256 hashes.
Deliverables
Consolidated access matrix, credential-rotation runbook, Gantt, provider deactivation tracker, chain-of-custody evidence log, post-cutover action plan with architecture evaluations.
Why the order matters
| Layer | Action | Reason |
|---|---|---|
| Password vault | Remove provider administrators first | Every other secret lives here |
| Identity provider | Rotate administrators, revoke provider users | It is the control plane for everything federated |
| Network edge and wireless | Rotate device admin passwords, remove provider source addresses and admin users | Closes the path that does not depend on single sign-on |
| MFA recovery channels | Move shared second factors to client-owned numbers and channels | Stops account recovery from undoing the rotation |
| SaaS and collaboration suite | Suspend accounts, strip roles, soft delete | Keeps ownership and audit history intact |
| Endpoints | Uninstall remote-access agents by MDM, verify by policy | Removes the last interactive path |
Decisions after the cutover
Taking the keys back was half the job. Each service the provider had been running needed a new home before the contract ended.
| Area | Before | Decision | Trade-off I flagged |
|---|---|---|---|
| Device management | Open-source MDM hosted in the provider tenant, expiring with the contract | Interim self-hosted instance on a managed PaaS (MySQL, Redis, encrypted object storage), then consolidation into the parent organization's Apple MDM | Vendor export plus re-enrollment of every device |
| Monitoring | Separate observability platform billed through the provider | Consolidate into the EDR vendor's SIEM and let the old contract lapse | Export and back up logs first: SOC 2 needs the retention |
| Enterprise browser | Legacy secure-browser product with unclear licensing | Managed Chrome Enterprise | Lost download-encryption controls, covered with the EDR browser extension |
| Compliance tooling | GRC platform under the provider contract | Export all evidence and move to the parent organization's tooling | Confirm full deletion on the provider side |
| Endpoint detection | Standalone tenant | Sub-tenant under the parent, after assessing exposure management, prevention policies, device control and host firewall | Some prevention settings held back until scripts were tested |
| Identity | Applications signing in with local accounts | Federate about 27 applications through the identity provider, 10 core first | Two to three weeks, estimate padded on purpose |
What I would tell the next team
Start with the vault
If the provider can still read the vault, every rotation after it is theatre.
Announce before you rotate
Owners hear about a credential change before it happens. Surprise lockouts cost more trust than they save time.
Soft delete during notice
Suspend and strip roles. Hard deletion waits until the audit trail has been exported.
Minutes after every meeting
In a handover the written record is a control. I sent minutes after every call.