Nobody Owns the Blame Anymore: Accountability in the AI Era
Translated from the original Korean post. 한국어 원문 보기 →
The First Question After Any Incident
I've sat through enough security incident response in financial services to notice a pattern. When something big breaks, the first question isn't "which tool missed it?" or "which dashboard failed to alert?"
The real question is always: who owned this?
Who ran the system. Who made the call. Who granted the access. Those questions were always uncomfortable, but in a traditional enterprise you could eventually answer them. There was a team. A manager above the team. A vendor with a contract attached. The trail might be long and ugly, but the chain of accountability never actually snapped.
AI-native infrastructure is quietly snapping it. Nobody stands up and announces "there is no owner anymore." The way the systems work just drifts, steadily, toward making the owner harder to find.
The Assumption Nobody Wrote Down
Traditional enterprise systems rest on one assumption so obvious that nobody put it in a design doc: every meaningful operational action eventually traces back to a stable human owner.
A system, a department, an employee, an outsourced vendor — didn't matter. The ownership chain could be convoluted, but it held. Held well enough to pass an audit, survive legal review, and be followed to the end during a post-incident investigation.
That assumption decided nearly everything downstream. How you design access control. How you structure audit trails. Who your compliance framework assigns responsibility to. It's why SOC 2, ISO 27001, and HIPAA are all built on top of human-centered accountability.
The assumption is starting to fail. Not because the technology changed, but because the thing taking action changed.
What AI Infrastructure Actually Breaks
Non-human identity, AI agents, ephemeral workloads — most of the conversation treats these as a technical complexity problem. Too many identities. Provisioning too fast. Lifecycles too short. Hard to track. All true.
But the technical framing skips the organizational consequence, which is the part that matters in a board meeting or an audit.
Here's the core of it. AI agents, autonomous execution layers, ephemeral environments, machine-to-machine workflows, and dynamic cloud resources all share one dangerous property: they produce operational actions with no stable human accountability behind them.
An ephemeral execution chain spans three clouds. Permissions get inherited from a service account that was decommissioned months ago. AI pipelines mint machine identities automatically during builds. Systems hand actions to each other with no clear ownership lineage.
Taken one at a time, each looks like routine infra hygiene. One permission never cleaned up. One account never deleted. Put them together and you're looking at systematic erosion of accountable ownership inside the enterprise.
The question itself has changed. It used to be "can we see what exists?" Now it's "can we assign accountable ownership for what happened?" The first is visibility. The second is governance. They are not the same problem.
What Storm-0558 Showed
Microsoft's Storm-0558 disclosure in mid-2023 is the proof this isn't abstract. State-sponsored actors stole a signing key and forged authentication tokens for 25 organizations, including U.S. federal agencies.
The technical breach was bad. The organizational response was worse. Nobody could answer, quickly or confidently, what these credentials could actually reach and who owned the decision that granted that access.
Non-human identity sat at the center of it. Permissions had drifted between systems over time, and ownership lineage was scattered across multiple control planes that didn't talk to each other. A chain that broke in one place couldn't be reconnected from another.
Working out the blast radius — how far this reached — took weeks. Congressional oversight followed, and the agencies involved needed extended review before answering questions that should have had immediate answers. The scary part wasn't failing to stop the breach. It was failing to say whose responsibility it had been, even after it was stopped.
The Pressure Is Already Here
This isn't a future risk. It surfaces at a few predictable moments.
Audit prep is the first. Teams hand-reconcile ownership records scattered across different systems and try to retroactively reconstruct accountability chains for resources that now have no clear owner. This usually kicks off about two weeks before the audit, and it always takes longer than anyone planned.
Incident response is the second. Somewhere in the middle of it, you realize the ownership lines are fragmented or missing, and you can't narrow the blast radius of a credential or an access path fast enough. The structure fails hardest exactly when you need it most.
Then there's AI rollout governance, where the board or the risk committee asks who is accountable for this agent's decisions, access, and modifications — and gets an answer that's technically correct and organizationally unsatisfying to everyone in the room. Regulatory review is where it gets ugliest: a compliance framework written on human-centered accountability meets infrastructure that was never designed to expose durable ownership in the first place. So somebody gets temporarily assigned, a spreadsheet gets created, and the deadline slips. Sound familiar?
There's an Opening Here
Most vendors in security, compliance, and governance still compete on visibility, coverage, and detection. Those matter. They're also problems enterprises already know how to budget for — familiar numbers in familiar boxes.
The genuinely new and hard problem is different: how do you maintain unbroken accountability in an environment where human ownership structures are weakening? No product category has cleanly solved it yet.
| Current approach | What's needed |
|---|---|
| Visibility and coverage focus | Accountability continuity focus |
| Technical detection capability | Organizational reliability |
| Security point solution | Cross-functional accountability platform |
The platform that defines this market won't be the one with the best detection logic. It'll be the one that becomes the accountability layer an organization can point to and trust — the thing you can put a finger on during an audit or a board review and say, this is where we maintain the ownership chain.
Where I Think This Goes
A few things I expect.
Governance frameworks will start explicitly requiring accountability lineage for non-human identities, going beyond an inventory record that says "these accounts exist." Audit and legal will shift their requirements too, demanding the same continuous ownership documentation for AI agents and autonomous systems that they've always demanded for human access.
And the win condition changes. The company that wins isn't the one with the most sophisticated asset map. It's the one that becomes the system of record owning the outcome of every decision path, end to end.
What's Left Is Where Accountability Sits
Accountable ownership is eroding at roughly the speed AI infrastructure is spreading. The market still sells tools for this as a visibility problem, while the real pressure builds one layer underneath.
When I was writing code, "does it work?" was the whole job. Moving into architecture, I started seeing "does it hold?" Then operations and consulting taught me that "who is responsible when it breaks?" is the question that should actually scare you. AI is solving the first two faster than ever, and making the last one harder to answer.
The further the technology goes, what's left at the end isn't the technology. It's who's accountable for it, and how.
Was this post helpful?
One click helps me write the next one