Analysis · Third-party access

Your vendors' admin tools are your perimeter: what N-central, ScreenConnect, Qubole and Optimizely have in common

In the space of a few weeks, attackers went after a remote management platform, a remote support tool, a data platform's cross-account roles and an experimentation tag. Different products, one pattern: each is a channel a vendor uses to act inside many customers at once. Most companies have more of these channels than they know, and none of them shows up on a firewall diagram.

AnalysisSumanta Mukhopadhyay8 min read7 sources
#news#security#governance#cloud
ANALYSIS · ANALYSIS · THIRD-PARTY ACCESS18 Nov 2026 · 21:20 GST

Four incidents, one pattern

N-able took its hosted N-central remote monitoring and management environments offline on 9 November after proof-of-concept exploits for more than 30 alleged flaws were published, and told on-premises customers to take their servers down1. A week later, attackers were exploiting a new N-central chain within hours of its proof of concept appearing2.

Meanwhile Huntress reported a campaign against managed service providers' ScreenConnect servers: between 20 October and 16 November, attackers posing as guests made around 688,000 requests across 42 servers to push a malicious file, waiting for a technician to join a session so a flaw in the technician's client would run it3. Wiz reported that an attacker used credentials tied to Qubole's AWS account to assume roles customers had created to trust Qubole, then collected secrets and data4. And attackers used a flaw in Optimizely's sign-up flow to inject malicious JavaScript into customer projects, which then ran on customers' websites5.

These are reported as four separate stories. They are one story. Each product is a channel a vendor uses to act inside many customers at once, and each attack went after the vendor's side of that channel rather than the customer's front door.

Where vendor access actually lives

Most organisations can list their suppliers. Far fewer can list what each supplier can do inside their environment right now, without asking. That second list is the one that matters.

Endpoints and servers
Remote management agentsrun commands, push softwareRemote support toolslive sessions, file transferSecurity agentskernel-level, auto-update
Cloud accounts
Cross-account rolesdata, monitoring, cost toolsMarketplace integrationsOAuth apps, service accountsCI and deploy toolswrite to production
Web and apps
Third-party scriptstesting, analytics, chatSDKs in mobile appsupdate with vendor releasesTag managerschange without a deploy
SaaS and identity
Admin consentsread mail, files, directorySupport accessvendor staff into your tenantAPI keys held by vendorsoften never rotated
Standing vendor access sits in four places. Each of this month's incidents hit one of them.

The common property is that the access is standing: it exists whether or not the vendor is doing anything for you today, and it is protected by the vendor's security, not yours. A remote management agent that can run any command at any time is the same risk on a quiet Sunday as it is during a support call.

Put vendor actions through the same gates as your own

The useful model comes from platforms that already treat every caller the same way. In Kubernetes, every request to the API server, from a person or from a workload, passes authentication, then authorisation, then admission control before it changes anything7. Vendor access usually skips most of that: the agent is authenticated once at install and then trusted for everything.

Diagram of a human user and a pod service account sending requests to the Kubernetes API server, which passes each request through authentication, authorisation and admission control before it reaches storage
Every request, from a person or a workload, passes the same three gates. Vendor access deserves the same treatment. Source: Kubernetes documentation · CC BY 4.0

Applied to suppliers, that means three questions for every channel. Who is acting: the vendor's platform, or a named technician with strong sign-in? What may they do: everything, or a defined set of actions on a defined set of systems? And what checks the action at the moment it happens: nothing, or a policy, an approval or at least a record someone reviews?

Cloud providers have built some of this already. AWS documents the confused deputy problem for exactly the Qubole pattern, and recommends conditions such as an external ID on roles that third parties assume6. That helps against one customer tricking a vendor into acting on another's account. It does not help when the vendor's own credentials are stolen, which is why narrowing what each role can reach matters as much as how it is assumed.

What to do: a short programme, not a project

  1. InventoryList every agent, role, script, app consent and support path a vendor holds
  2. OwnerGive each one an internal owner and a reason it still exists
  3. RemoveDelete access for vendors and features you no longer use
  4. NarrowCut each remaining channel to the systems and actions it needs
  5. WatchLog vendor actions where you can see them, not only where the vendor can
  6. Cut-offRehearse switching each channel off in minutes
Most of the risk reduction comes from the two steps in the middle.
ChannelThis month's exampleFirst thing to change
Remote managementN-central exploits1Limit which machines the agent covers and what scripts it may run
Remote supportScreenConnect technician attacks3Patch technician clients; block file transfer by guests; record sessions
Cross-account cloud rolesQubole roles assumed4Remove unused roles; keep secrets out of what roles can read
Third-party scriptsOptimizely injection5Keep vendor scripts off payment and login pages; restrict with a Content Security Policy

The test is simple: if a vendor told you at 3am that its platform was compromised, could you cut its access to you within the hour, and know what it had done in the last month? If not, start there.

The conversation with your managed service provider

Many mid-sized companies reach remote management and remote support tools only through a managed service provider, so the exposure is real but the controls belong to someone else. That makes the contract and the questions the control. Ask which tools they run on your systems and at what versions, how technician accounts sign in, whether sessions are recorded and whether you can see the logs, and how they would tell you about a compromise of their own tooling. The MSP cyber resilience case study shows what a provider that takes this seriously looks like from the inside.

For cloud and SaaS access, the same thinking runs through the third-party cloud access case study and the machine identity case study. For web scripts, see the payment page script security case study, and for AI vendors specifically, AI inference providers are now supply chain.

The CTO's take

Supplier risk is usually run as paperwork: questionnaires, certificates, annual reviews. This month showed it is an engineering problem. The question is not whether a vendor is reputable, but how much it can do inside your environment without you noticing, and how fast you can stop it. Those are things you can measure, reduce and rehearse.

If you want a second pair of eyes on vendor access across your estate, or as part of technical due diligence, I am a fractional CTO based in India, the United States and the UAE, working globally. You can analyse your challenge and bring the result to a call.

References
  1. 1
  2. 2
  3. 3
  4. 4
  5. 5
  6. 6
    The confused deputy problemAWS Identity and Access Management User Guide
  7. 7
FAQ

Common questions

What do these incidents have in common?

Each abused a channel that a vendor uses to act inside many customer environments: remote management agents, remote support sessions, cross-account cloud roles and a script loaded on customer websites. Compromising the vendor's side gives access to every customer at once.

What is standing vendor access?

Access a supplier holds all the time, whether or not it is doing work for you: an agent that can run commands, a role it can assume, a tag it can change. It is convenient, and it is exposed for as long as it exists.

What should I ask my managed service provider this week?

Which remote management and remote support tools they use on your systems, whether those are patched or offline, how technician accounts are protected, whether sessions are recorded, and how quickly they can show you what was run on your machines in the last month.

Keep reading

Related

Open to advisory, fractional CTO & select builds

Let's talk about what you're building.

The fastest route is a free 30-minute call. Prefer to write? Send a message and I'll reply personally within two business days.

Start a conversation

How would you like to reach me?

Book a 30-min callFastest. Pick any open slot on my calendar.→ Write in GmailOpens a pre-filled draft in your browser.→ Use my email appOutlook, Apple Mail or your default client.→

I reply personally within two business days. Time zones from the Americas and Europe to the Middle East and Asia-Pacific.