Every framework promises that risk can be managed into submission. Reality keeps disagreeing.
On 19 July 2024, a routine content update from a security vendor crashed roughly 8.5 million Windows machines. Airports stalled, hospitals fell back on paper, and banks went quiet. Nobody had attacked anyone. Nobody had broken a rule, at least not in the way compliance teams usually count rule-breaking. Thousands of organizations that held current certifications, passed their audits and filed their risk registers found out in a single morning that their controls had never covered this.
I keep coming back to that day when people ask me whether their governance programme is working. The honest answer is usually uncomfortable. A programme can be thorough, well documented and fully audited, and still leave you exposed. That is not a failure of effort. It is a property of the thing we are trying to do. We write controls to feel in charge of systems that are too large, too fast and too tangled for anyone to be fully in charge of.
The pattern has not gone away. On 12 November 2026, a Microsoft 365 outage knocked out Exchange Online and Outlook, then spread to Teams, SharePoint, Copilot and other services that depend on them, and it ran for roughly a day before Microsoft reported service restored. Microsoft said the trouble began with an earlier network capacity upgrade and that it had found no sign of malicious activity. Again nobody had attacked anyone. Hospital services in the UK, airlines in India, government agencies in Australia and Japan, and even the AI connectors of OpenAI and Anthropic felt it, and at least one listed company delayed its quarterly results because of it. It was a different vendor and a different fault, but the same lesson: one provider's internal change can quietly become everyone else's operational incident.
Control, in other words, is the biggest illusion in governance and compliance. Not because controls are useless (they are not), but because we have started to treat them as proof of safety rather than as one input into a much messier picture. Here is how that illusion forms, where it cracks, and what a more honest approach looks like.
1. The Myth of the Perfect Framework
Walk into almost any large organization and you will find a shelf, physical or digital, of frameworks. ISO 27001, NIST, SOC 2, COBIT, PCI DSS, plus whatever the local regulator has added this year. Each one is a serious piece of work, built by smart people who have seen real failures. I have no argument with any of them.My argument is with the quiet belief that arrives alongside them: that if we implement the framework fully, the risk is handled. A framework is a map drawn from past incidents. It tells you what went wrong for others, in conditions that have already changed. It cannot tell you what is about to go wrong for you, in your particular tangle of systems, vendors and habits.
There is also the matter of scope. Every framework needs boundaries to be auditable, so it draws a line around a system, a process or a business unit and asks whether the controls inside that line are working. Real incidents rarely respect the line. They start in the spreadsheet nobody scoped, the integration somebody added on a Friday, the supplier two steps removed from your contract.
Then there is the old trap that economists call Goodhart's law: when a measure becomes a target, it stops being a good measure. Compliance is full of this. We measure patch rates, training completion, policy acknowledgements, and then teams optimize for the number. A dashboard that is green all year can sit on top of a very fragile reality, and everyone in the room half knows it.
None of this means you should abandon frameworks. It means you should read them the way a sailor reads a chart: as a useful, slightly out-of-date sketch of the coast, to be checked constantly against the water in front of you.
2. The Human Variable and the “Tick-Box” Trap
Every control eventually passes through a person. Someone approves the access request, someone reads the alert, someone decides whether the odd email from the finance director deserves a second look. The control is only as good as that moment of attention, and attention is not something you can install.The numbers keep saying so. Verizon’s 2026 Data Breach Investigations Report, which looked at more than 22,000 confirmed breaches across 145 countries, put the human element in 62% of them. Over the last four editions that figure has hovered between 60% and 74%. We have spent a decade on awareness training and the needle has barely moved. One analysis of the same data points out that the median time from opening a phishing email to typing in credentials can be under a minute. That is not a knowledge problem. People know phishing exists. It is a pace-and-pressure problem, and a slide deck on Tuesday morning does not fix it.
This is where the tick-box trap closes. When compliance is measured by completion, the incentive is to complete. Staff click through the annual module with a second monitor open. Managers sign attestations they have not really read, because the deadline is Friday and the sign-off is a formality. Auditors see evidence, and evidence is what they are paid to find. Everyone behaves rationally, and the organization ends up with a perfect paper trail and a thin real defence.
I have sat in rooms where a control was described as “operating effectively” because the log showed it had run, not because anyone had checked that it caught anything. The tick was real. The protection was not.
A control that exists on paper is a claim. A control that has been tested under pressure is evidence.
There is a kinder way to look at this, and I think it is also the more accurate one. People are not the weakest link. They are the only part of the system that can notice something strange, improvise, and say “this feels wrong.” The trick is to design governance that supports that instinct rather than burying it under forms. Make the safe path the easy path. Give people a fast, blame-free way to report mistakes. Measure how quickly problems surface, not how few are admitted to.
Shadow AI is the newest version of the same story. The same Verizon report found that 67% of users who touched AI services on company devices did so through non-corporate accounts, and that regular AI use among employees had jumped from 15% to 45% in a year. Nobody set out to leak source code. They wanted to get a task done faster. A blanket ban that people quietly route around is a control in name only, and it tells leadership nothing about what is really happening.
3. The Speed of Change vs. The Speed of Policy
Policies are written slowly, and for good reason. They need drafting, legal review, stakeholder comments, committee approval and a communications plan. Six months is a respectable turnaround. Meanwhile the thing being governed has been through several release cycles, two vendor changes and probably a reorganization.Attackers are not waiting on committees. The 2026 DBIR found that exploitation of vulnerabilities has overtaken stolen credentials as the most common way in, at 31% of breaches. At the same time, only 26% of known exploited vulnerabilities were fully fixed in 2025, down from 38% the year before, and the median time to remediate stretched from 32 days to 43. The gap between how fast the risk moves and how fast we respond is widening, and a policy that says “critical patches within 30 days” is describing an aspiration, not a fact.
Regulation has the same rhythm problem. Rules arrive years after the technology they address, and by the time they bite, the practice has shifted again. Generative AI is a good example: employees were pasting documents into chatbots long before most organizations had a position on it. The policy, when it arrived, was governing last year.
What helps is treating policy as a living product rather than a monument. Short principles that stay true for years, backed by operating standards that can be updated in weeks. Named owners who review them on a schedule and when something breaks. A habit of asking, after every incident or near miss, “which of our assumptions just turned out to be wrong?” Governance that cannot change its mind quickly is governance that is already out of date.
4. The Borderless Ecosystem (Third-Party Risk)
Not long ago, an organization had a perimeter. You could draw it, defend it and audit it. Today your data sits in a cloud provider, your payroll in one SaaS tool, your customer chat in another, your code in a repository maintained partly by strangers. Your risk is the sum of everyone you depend on, and most of them you have never met.The 2026 DBIR counted a third party in 48% of breaches, a 60% jump on the previous year. The UK’s Financial Conduct Authority said something similar after the CrowdStrike outage: between 2022 and 2023, third-party problems were the leading cause of operational incidents reported to it, and many firms depend on unregulated providers to deliver important services. Some of the firms affected by the outage were themselves suppliers to other regulated firms, which stretched the damage along the chain.
Here is the uncomfortable part. The standard response is the vendor questionnaire, and a questionnaire is a snapshot of what a supplier says about itself on the day it answers. It says little about what the supplier’s own suppliers are doing, and nothing about next month. Data from Tenchi Security in the same DBIR shows how slowly the real work gets done: almost half of third-party cloud authentication and privilege issues were still unresolved after 350 days.
You cannot control a vendor’s engineers, their release process or their sub-contractors. What you can do is know where you are exposed. The firms that coped best in July 2024, according to the FCA, had mapped their important business services and the third and fourth parties behind them, and had existing lines of communication to share information quickly. That is not control. It is visibility and rehearsal, and it turned out to be worth far more.
Regulators have started to formalize this. The EU’s Digital Operational Resilience Act, applying to financial entities since 17 January 2025, requires a register of every ICT third-party arrangement, concentration-risk assessments and oversight of critical technology providers. Notably, it puts ultimate accountability for ICT risk on the management body, not on the IT department. The direction of travel is clear: you may outsource the service, but not the responsibility.
5. The Paradigm Shift: From Control to Resilience
If control is partly an illusion, what replaces it? I would argue the answer is resilience: the ability to absorb a hit, keep the important things running, recover quickly and learn from it. Control asks “how do we prevent this?” Resilience asks that too, but adds the harder follow-up: “when it happens anyway, what then?”The two ideas are not enemies. Good controls reduce the number of bad days. Resilience decides how bad the bad days get. The shift is one of emphasis, and it changes what leaders spend their attention on.
5.1 What resilient governance looks like in practice
Start from the services that matter. Identify the handful of business services whose failure would really hurt customers or the market, then map the people, systems and suppliers each depends on. This is the heart of the UK operational resilience rules and of DORA, and it works for any organization.Set impact tolerances, not just risk appetites. Decide in advance how long each service can be down before the damage becomes unacceptable. It forces honest conversations that a risk heat map never does.
Test against severe but plausible scenarios. Tabletop exercises, failover drills, a supplier going dark for three days. The FCA noted that firms with tested scenarios and rehearsed communications responded faster and with less harm. Aon’s post-incident polling found that although 76% of respondents said their incident response plan performed well, plenty of others had learned where theirs did not, which is exactly why you test.
Remove single points of failure where you can. Phase software rollouts, keep tested backups, maintain manual workarounds for critical processes, and avoid putting every important function on one vendor’s shoulders.
Plan the communication, not only the technology. In a crisis, customers and regulators remember how clearly you spoke to them. Decide who speaks, how quickly, and with what authority before you need it.
5.2 Culture beats paperwork
None of the above works if people are afraid to report a problem. A resilient organization is one where the junior analyst can say “I clicked something I shouldn’t have” within five minutes instead of five days. That is a leadership behaviour, not a policy clause. If the first response to a mistake is blame, the next mistake will be hidden, and hidden mistakes are the ones that become headlines.5.3 Rethink what you report upward
Boards deserve better than a wall of green. Useful reporting shows how long critical issues stay open, how often the plan was exercised and what the exercise revealed, which suppliers carry concentrated risk, and where the organization is knowingly accepting exposure. Reporting that includes uncomfortable truths builds more trust, over time, than reporting that never does.6. A Practical Starting Point
If this all sounds large, begin small. In the next quarter, pick three things. First, choose one critical service and map it end to end, including suppliers. Second, run a two-hour exercise on it with the people who would actually respond, and write down what surprised you. Third, take your most trusted control metric and ask what it would look like if it were green but wrong. Then fix what you find, and repeat. Resilience is built by repetition far more than by programmes.7. Closing Thought
The goal was never to eliminate uncertainty. That was always a fantasy, and a slightly dangerous one, because it makes us stop looking. Frameworks, controls and audits are good tools, and I would not run an organization without them. But they work best in the hands of people who know their limits.So hold your controls a little more loosely. Ask what they cannot see. Practise for the day they fail. The organizations that come through the next big disruption in good shape will not be the ones that felt most in control. They will be the ones that were honest about how little control they had, and prepared accordingly.



