Here is an uncomfortable observation. Walk your line tomorrow morning and count the third parties who can reach into it without badging at the gate. The machine-builder's remote diagnostics desk, three time zones away. The CMM vendor who dials in for calibration and somehow never logs out. The integrator whose commissioning VPN is still standing because the project ended on a Friday and nobody revoked anything. Two decades in automotive and aerospace quality taught me that the number in the maintenance log is always higher than the number IT holds. When CISA and the FBI told critical-infrastructure operators to lock down third-party access to their control systems – least privilege, controlled remote sessions – the advisory was filed by the CISO. The doors it describes were opened by procurement. The warning went to the wrong department.

The back door has a purchase order

Nobody hacks their way into most of these paths. They buy them. The €1.4 million machining centre arrives with a service contract bundling "24/7 remote diagnostics" for €50,000 a year, and the builder's engineer connects through a tunnel the plant never sees. The test-rig supplier keeps a support login because ramp-up ended and the account didn't. The MES and SPC vendors hold patch accounts on the servers carrying your traceability data. The integrator's laptop on the line holds credentials to half the PLCs in the hall. Every one of these is bought, contract-bound, permanent by default. Nobody malicious installed it; it came through the front door with an invoice. The most expensive door in the plant came free with the warranty.

Nor is the exposure abstract. Security researchers have crashed industrial servers with a single malformed packet, and a Florida water utility learned that lesson in public in 2021. What turns a crash into a catastrophe is who else is already standing inside the control network when it happens.

The audit scored the QMS and missed the connection

I hold up my own hand. At the greenfield plant where I built a QA/QC department from bare concrete for 900 people – VDA 6.3 as the backbone, multi-site supplier audits – I wrote the supplier-control system myself. Our checklists verified PPAP files, capability indices, reaction plans, calibration records, traceability. They were good. They covered the torque wrench down to its calibration sticker. In years of audits, they never once asked about credentials. Not a single question about remote access.

That is not a personal blind spot; it is the standard. VDA 6.3 and IATF 16949 supplier audits certify parts, process, reaction plan. The supplier's pathway into your systems is invisible to the checklist. Five questions that never get asked:

  • Who holds accounts on our systems?
  • When did they last connect?
  • Who approved it?
  • What did they change?
  • When does the access expire?

Read them again. Not one of them is a security question. Every one is a supplier question.

My other trade explains why this persists. As a certified ethical hacker – I sit on T-Mobile's bug-bounty Hall of Fame for a clickjacking disclosure – I once reported a flaw that had sat untouched for years because no team owned that surface. Not maliciously neglected. Simply unowned. That is the state of third-party access on most lines today. Auditing and penetration testing are the same discipline: finding what a system trusts that it shouldn't. A supplier audit is a penetration test with a scoring sheet, and we have been scoring the wrong pages.

Least privilege is a control plan

Security teams say "least privilege" and quality directors stop listening. Translate it. Least privilege is concession logic, and you already run concession logic daily: temporary, dated, signed, revoked. Standing access is a concession with no expiry date – the one kind no quality director would ever sign.

A standing VPN is a permanent concession granted to a supplier who never submitted a part.

Treat every remote session as an engineering change. A ticket before the connection opens. A defined scope – which controller, which parameter class, which window – and an expiry that is actually enforced. Broker the session through a jump host that records it. When the vendor logs off, re-verify the process: first-off part, capability check. A firmware tweak applied at 02:00 from a service desk abroad is a process change whether or not anyone issued an ECN. We demand a full PPAP when coating thickness moves three microns, then wave through a PLC firmware bump on the station that torques the safety-critical joint. If a defect spike follows a remote session, you want that session log in your 8D the same way you want the lot numbers.

Key takeaways

  • Add five questions to the next supplier audit: who holds accounts on our systems, when did they last connect, who approved it, what did they change, when does the access expire.
  • Treat every remote session as an engineering change – ticket before the connection, defined scope, hard expiry, recorded through a brokered jump host.
  • Apply concession logic to access: temporary, dated, signed, revoked. Renew on evidence, never on habit.
  • After any remote change, re-verify the process – first-off part and capability check. A firmware tweak is a process change with or without an ECN.

Security owns the threat model. Quality owns the supplier relationship. Nobody owns the overlap, and the overlap is where the line actually lives. Two practical starts this month: put the five questions into the next supplier audit, and walk the maintenance log once, counting live remote accounts against the list IT holds. The gap between those two numbers is your real exposure, and every account in it has a purchase order against it. Third-party remote access is a supplier like any other. The next advisory should not be news to the quality department; it should be an audit finding we closed first.