Every operating model of the last decade has pushed work outward. Contract manufacturers. Managed service providers. Logistics partners. Implementation vendors. Agencies, resellers, distributors, the specialist firm that runs one critical system nobody in-house understands anymore.

You didn't build a company. You built an ecosystem.

Meanwhile the security model underneath it still quietly assumes a perimeter with employees inside it. That assumption is now the single largest gap between how businesses actually operate and how they're actually defended.

The Door You Didn't Build

Verizon's 2026 Data Breach Investigations Report puts third-party involvement in 48% of all breaches — a 60% jump in a single year. Look at the trajectory rather than the number: it was 15% two reports ago, 30% last year, 48% now. Nothing about your own security posture explains that curve. What changed is how much of your business runs through somebody else.

The cost follows. Third-party breaches average $4.91 million, against a global all-breach average of $4.44 million. Incidents that arrive through a partner are more expensive than the ones that arrive through your front door — they take longer to detect, and the evidence lives in someone else's systems.

For a COO this isn't a security statistic. It's an operating statistic. Roughly half your breach exposure now sits in relationships your procurement team signed.

You Can't Govern What You Can't List

Here is the part that should be uncomfortable, from Imprivata and Ponemon's 2025 research into third-party access in manufacturing:

  • 43% have no comprehensive inventory of the third parties with access to their network.
  • 54% do not evaluate a third party's security practices before granting access.
  • 35% of third-party incidents trace to excessive vendor privileges — not a clever attack, just more access than the job required.

Sit with the first one. Nearly half can't produce a list of who's inside. Every downstream control — certification, monitoring, revocation, incident response — assumes that list exists. Without it, third-party risk management is a set of policies performed on a population nobody has counted.

And the third finding is the one I would put in front of a board. The dominant failure mode isn't a sophisticated adversary. It's a vendor who was given standing access to more than they needed, for longer than they needed it, because scoping it properly would have taken three weeks and the project started Monday.

That's not a vendor problem. It's an identity problem wearing a vendor costume.

The Tax Nobody Books

The risk framing is only half the story, and honestly it's the half that gets IAM funded and then quietly resented.

Consider what onboarding a partner actually costs you today. Someone raises a ticket. Someone provisions accounts in three systems. Someone emails credentials. Someone else discovers in month four that the pilot ended in month two and nobody switched anything off.

In that same manufacturing research, security teams reported spending 47 or more hours a week analyzing third-party access risk — and 31% spend over a hundred. That's a full-time job, in many cases more than one, spent reconstructing a picture that better infrastructure would simply have.

Meanwhile the partnership waits. The joint venture waits. The acquired subsidiary runs on its own directory for eighteen months because integrating it's "a project." Every one of those delays is a real business outcome deferred, and none of them appear on a balance sheet as the cost of an identity decision.

This is time-to-yes at the organizational boundary. Internally it determines how fast your developers ship. Externally it determines how fast you can partner at all — and partnering is the operating model.

Federation Is an Operating Decision

The instinctive answer to third-party risk is more gates: longer questionnaires, tighter reviews, more approvals. It reduces risk the way a speed limiter reduces accidents. It also guarantees you're the slowest partner in every ecosystem you join.

The better answer is architectural. Let partners bring their own identity.

In a federated model the partner authenticates against their own identity provider. You never hold their credentials, never provision a shadow account, never run a password reset for someone else's employee. You grant scoped, time-bound access to a role — and when that person leaves the partner organization, their access to you ends at their employer's offboarding, not at whenever someone remembers to tell you.

That last point is the one executives underrate. In the account-provisioning model, your revocation depends on a partner's HR process reaching your ticket queue. In the federated model, it's automatic, because you were never the system of record for that human in the first place.

This is what security that says yes looks like at the organizational boundary. Not fewer controls — controls that survive contact with how partnerships actually work. Scoped access granted in hours. Automatic expiry. A real inventory, because federation produces one as a byproduct rather than as a quarterly exercise.

Access stops being a gate and becomes an accelerator: the thing that lets you say yes to a partnership on Tuesday and have people working on Wednesday.

What Agents Make Urgent

The ecosystem is about to stop being made only of people.

Organizations are already deploying AI agents that read from partner systems, submit orders, reconcile invoices, and file tickets across organizational lines. Those agents are non-human identities operating outside your walls, at machine speed and machine volume.

This is my read rather than a cited finding, but the logic is hard to escape: if you can't currently produce a list of the humans with third-party access, you'll not manage a population of autonomous agents acting on partners' behalf. The failure mode we already have — excessive standing privilege, granted quickly, reviewed rarely — doesn't improve when the thing holding the credential never sleeps and never forgets it has them.

Every organization gets to choose whether it solves this before agents arrive in the supply chain or after.

Map Your Ecosystem Access Requirements

Start with the inventory, because everything else is theater without it.

Count them. Produce a single list of every external organization with access to your systems, and every identity inside those organizations. If you can't assemble it in a day, that's the finding — you've just measured the gap the 43% are living in.

Score by blast radius, not by contract value. Your largest vendor is rarely your largest exposure. The one with standing admin on a production system is, whatever they invoice.

Find the standing privilege. Every permanent grant that should be time-bound is a future incident with a date not yet assigned. This is where the 35% lives.

Measure your partner time-to-yes. From signed agreement to a partner actually working: how many days? Ask what fraction of that's access provisioning. That number is your collaboration tax, and it's almost always larger than anyone senior believes.

Ask who deprovisions. For each partner, name the event that ends their access and the system that enforces it. If the honest answer is "they tell us," you don't have a control. You have an arrangement.

The organizations that will compound advantage over the next decade aren't the ones with the most partners. They're the ones that can safely onboard a partner in days and cleanly remove them in minutes — and can say, at any moment, exactly who's inside.

Half of all breaches now arrive through someone else's door. The same infrastructure that decides how well you close that door decides how quickly you can open it.

Map your ecosystem. Start with the list.