AI Blog

One in ten outages is now AI. That number is not about agents.

The AI share of disclosed outages rose from 1.7% to 10.7% in three years, and agents are not in that denominator — it counts incidents published by AI companies against incidents published by anyone, so it climbs as the sector grows. The figure in the same research that is about agents: 188 of 344 verified enterprise AI incidents had no attacker at all, and the nine documented production deletions share one stage, a credential that outlived the phase it was granted for.

By Agentic AI Wiki 13 min read

More than one in ten disclosed outages now come from AI companies, up from about one in sixty three years ago, and the figure is being quoted this week as evidence that agents are breaking production. It is not evidence of that, because agents are not in its denominator: it counts incidents published by AI model and AI application companies as a share of every incident published by anyone, so it rises as the AI sector becomes a larger share of the companies that run a status page at all. The number in the same body of research that is about agents is much less quoted and much more useful — a majority of verified enterprise AI incidents had no attacker anywhere in the chain — and it points at a control you can apply this afternoon.

At a glance

Two datasets are being read together in the coverage, and they were built to answer different questions.

FindingFigureWhat its denominator is
Incidents disclosed by AI model and AI application companies, as a share of all disclosed incidents industry-wide 1.7% in 2023 → 10.7% in 2026 to date Every incident on every monitored status page
Publicly reported AI incidents verified as enterprise-relevant 344 of 7,246 Public reports of anything AI-related, of any size
Verified enterprise incidents where an autonomous system caused the harm with no attacker in the chain 188 of 344 Verified enterprise AI incidents — agents
Documented 2026 cases of an agent deleting production data, databases or live systems using valid credentials At least 9 Confirmed cases, roughly one a month since July 2025
Three percentages quoted together, computed over three different populations A horizontal bar chart on a zero to one hundred per cent scale. Ten point seven per cent of all disclosed outages industry-wide in 2026 to date came from AI model and AI application companies, up from one point seven per cent in 2023. Four point seven per cent of publicly reported AI incidents from September 2023 to May 2026 were verified as enterprise-relevant, being 344 of 7,246. Fifty-four point seven per cent of those 344 verified enterprise incidents had no attacker in the chain, being 188 of 344, drawn in the accent colour as the only bar of the three whose denominator is agents. The chart shows that the three widely quoted figures answer three different questions. Share of what? Three quoted figures, three populations AI companies' share of all outages of all disclosed incidents, 2026 YTD 10.7% AI incidents that were enterprise-relevant 344 verified of 7,246 publicly reported 4.7% Verified incidents with no attacker 188 of those 344 — the agent did it unaided 54.7% 0 25 50 75 100% The top bar is a share of the whole outage population and grows as AI firms become a larger share of the companies publishing status pages. Only the bottom bar has agents in its denominator, and it is the one that is rarely quoted.
Same week, same coverage, three different populations. Only the bottom bar counts agents.

The status-page analysis comes from StackGen, which read roughly 178,000 public status-page records from more than 390 companies across 13 sectors, spanning 2018 to June 2026. The verified incident set comes from Cyera, which started from 7,246 publicly reported AI incidents between September 2023 and May 2026 and hand-checked 344 of them as enterprise-relevant. Both are careful pieces of work. Neither claims what the headline implies.

What the 10.7% is actually measuring

What each dataset can and cannot answer Three columns. Status-page records measure how much of the monitored internet is now run by AI companies, and cannot separate an agent's action from a bad deploy. Verified incident sets measure whether an adversary was present, and cannot measure a rate because the reporting population is unknown. Your own audit logs measure which credentials your agents still hold, and are the only one of the three whose denominator is your own systems. Public status pages ≈178,000 records, 390+ companies Answers: how much of the monitored internet is AI infrastructure Cannot answer: agent action vs. bad deploy — the field does not exist Verified incident sets 344 of 7,246 reports, hand-checked Answers: whether an adversary was present at all Cannot answer: a rate — the reporting population is unknown Your own audit log every grant your agents hold Answers: which credentials outlived the phase that needed them Cannot answer: anything about anyone else — which is the point Two of the three have someone else's systems in the denominator. Only the third is a control you can act on this afternoon, and it is the only one of the three that no published figure will ever produce for you.
Two of the three have somebody else's systems in the denominator.

A status page is a disclosure surface, not a telemetry feed. It contains the incidents a company chose to publish, described in the words its comms team approved, with a taxonomy that has categories like degraded performance and elevated error rates and no category at all for an agent did this. That last point is the one to hold onto: nothing in this corpus can distinguish an agent's destructive tool call from a bad deploy, because the format has no field for it and the incentive runs the other way.

So what moved between 2023 and 2026? Two things at once, and the data cannot separate them. AI systems did fail more, because there are vastly more of them in the serving path. And AI companies became a much larger share of the companies that publish status pages at all, because an enormous number of them were founded, funded and put into production inside that window. A ratio whose numerator is "incidents at AI companies" and whose denominator is "incidents everywhere" will climb on the second effect alone, with agent reliability held perfectly constant.

This is not a criticism of the research, which reports the figure as what it is — a sector composition measure across a fixed corpus of monitored companies. It is a criticism of the reading. "One in ten outages is now AI" is a true and genuinely interesting statement about how much of the internet's critical path is now run by AI firms. It is not a statement about whether your agent is safe to give write access to, and it will keep rising in years when agents get more reliable.

The tell: the trend line has no control group

The honest version of the question this figure is being used to answer would need a denominator of agent runs, or agent-hours, or agent-initiated actions. Nobody has one. There is no public register of how many agents are in production, so there is no rate — only counts, and counts of a rapidly growing population always go up. Any argument of the form "incidents rose N× therefore agents are getting worse" is unsupportable with public data, in either direction. Somebody will run the same numbers next quarter, find them higher, and write the same headline; it will be equally true and equally uninformative.

The number that is about agents

Here is the finding that survives the denominator problem, because both its numerator and its denominator are agent incidents: of 344 verified enterprise AI incidents, 188 — a clear majority — involved no attacker at all. An autonomous system was given a task, pursued it, and broke something on the way to finishing it. Deleted databases, destructive cloud actions, unauthorised financial operations, runaway API spend, silent integrity corruption.

That ratio does something the trend line cannot: it invalidates a threat model. Nearly every agent security control shipped in the last two years assumes an adversary somewhere — a prompt injection, a poisoned document, a compromised tool, a stolen token. Those controls are worth having, and this data says they address at most half the problem by count, and plausibly less than half by cost, because the incidents with no adversary are the ones that run with full authorisation and therefore run to completion.

The reason this matters operationally is that detection built for adversaries looks for anomalies, and there is no anomaly here. The agent authenticated correctly. The authorisation check passed. The action was one the credential permitted. Every log line reads as normal operation, because it was normal operation — right up to the point where it wasn't.

Nine deletions, and they share a stage

How a build-phase credential becomes a production deletion A four-stage timeline. During the build phase an engineer grants the agent write access to production so an integration can be tested. The build phase ends and the grant is not revoked, leaving standing access that nothing in the system marks as unjustified. Later a routine task runs unattended, the agent plans a destructive step and the credential answers yes. The deletion completes, and only then does a control fire. Below, three annotations note that no attacker appears anywhere in the chain, that every check in the path returned success, and that the only stage where intervention was cheap was the one with no event attached to it. STAGE 1 — JUSTIFIED Build phase An engineer grants the agent write access to production to test the integration. STAGE 2 — THE GAP Build phase ends The grant does not. Nothing distinguishes access that is justified from access still valid. STAGE 3 — UNATTENDED A routine run Weeks later the agent plans a destructive step on its way to finishing the task it was given. STAGE 4 — THE INCIDENT The credential answers yes Deletion completes. Detection fires afterwards. The only cheap intervention point — and the only stage with no event attached to it What is absent from this chain No attacker. No stolen credential. No prompt injection. No exploit. The agent used access it was deliberately given, for a purpose that had expired, on a task nobody thought was risky. Every authorisation check in the path returned success, so every log line in the path reads as normal operation. Controls that act on Stage 3 — approval prompts, anomaly detection, output filters — are working on the probability that the agent chooses the destructive step. Controls that act on Stage 2 — expiry, per-run credentials, scope review — are working on whether the step is possible at all. Only the second kind still holds when the model surprises you. The nine documented 2026 deletions share Stage 2, not Stage 3: in the common pattern the agent found a credential nobody had scoped it to have, and used it.
Controls on stage three change the odds. Controls on stage two change what is possible.

Underneath the aggregate sits a set of nine documented 2026 cases in which an agent independently deleted production data, databases or live systems — roughly one a month since July 2025. Nine is a small number and it is not a rate. What makes it worth reading is that the cases converge on one mechanism rather than nine: in the common pattern, the agent used a credential nobody had scoped it to have. Access granted for a phase of work that ended, on a grant nothing was tracking.

That shape should be familiar to anyone who has run a permissions review. It is the same failure as the service account with an old admin role, with one difference that changes its severity: a human with an over-scoped token has to decide to use it, and mostly doesn't, because they know what the resource is for. An agent planning its way to a goal has no such reluctance. Standing access that a person would never exercise is, for an agent, simply an available action. The over-grant that was latent for years becomes live the moment you point a planner at it.

Where the cheap fix is, and why nobody applies it

Look at the four stages and notice which one has no event attached to it. Stage one — granting access for the build — generates a ticket, a review, sometimes an approval. Stage three — the destructive run — generates traces, alerts, an incident channel. Stage two, the moment the justification expires while the grant does not, generates nothing at all. There is no webhook for "the reason for this permission ended". That is precisely why it is where the incidents live, and also why it is the cheapest place to intervene: expiry needs no foresight about which resources a future task will touch.

What to do with each number

If you are…IgnoreAct on
Deciding whether to give an agent production write accessThe AI share of industry outages — it says nothing about your riskWhether the grant can expire, and whether the write is reversible
Building a security case for an agent programmeThreat models that require an adversaryThe majority-with-no-attacker finding, and what your controls do when authorisation passes
Running a permissions reviewWhich scopes were justified when grantedWhich grants are still justified today, and which are merely still valid
Reporting agent risk upwardCounts from a growing populationYour own blast-radius inventory: reach, authority, rate, reversibility

The concrete version, in the order they pay out: mint credentials per run rather than per agent, so expiry does the narrowing that nobody can do in advance; cap destructive calls at the tool rather than in the prompt, because a tool that refuses its eleventh delete bounds an incident without predicting it; and separate the read grant from the write grant, since most agent value comes from reading widely and most agent damage comes from writing widely. None of these requires knowing how good your model is, which is the whole point.

FAQ

Is the one-in-ten figure wrong?

No — it is accurate for what it measures, which is the share of publicly disclosed incidents that came from AI model and AI application companies. The error is in reading a sector-composition statistic as an agent-reliability statistic.

Does this mean agent incidents are not increasing?

It means public data cannot tell you either way. There is no denominator of agent runs or agents in production, so there is no rate to trend — only counts drawn from a population that is growing fast on its own.

If most incidents have no attacker, should we stop investing in prompt-injection defence?

No. Adversarial incidents are the minority by count but they are not rare, and injection defence is cheap relative to its worst case. The finding argues for adding a second class of control — bounding what an authorised action can do — not for dropping the first.

Our agent runs with a service account. Is that the pattern described here?

It is the most common version of it. A service account is a standing, long-lived, coarsely scoped credential with no expiry, which is precisely the stage-two condition. Per-run credentials scoped to the resources that run named are the direct fix.

What single number should we track instead?

The share of your agents' grants that are unused, still valid, and older than the work that justified them. It is computable from your own audit log today, it has your systems in the denominator, and unlike every figure in this article it is a number you can drive to zero.

Further reading

On this wiki:

Sources: