Can an organisation be ISO 27001 certified and still struggle to respond effectively to AI-driven security risks?
The answer is yes! ISO 27001 gives organisations a structured way to manage information security risks. However, having the framework in place does not automatically mean that risks are being identified, decisions are being made, and controls are being applied effectively.
AI is making that gap harder to ignore. It can make familiar threats faster, more scalable, and more convincing. It can also increase the amount of information security teams need to assess when deciding what requires action. That raises a more important question…..
Is the real challenge protecting AI, or making sure the organisation can make and act on the right security decisions?
This was the focus of a recent Grow Skills Store webinar hosted by Punit Bhatia and presented by ISO expert Quentin Sorneau. The session explored what AI is actually changing about cyber risk, why ISO 27001 can lose its impact in practice, and how organisations can move from governance to execution.
Quentin summarised the shift simply:
“ISO 27001 does not make you secure. It gives you the system to become secure.”
In this article, we explore what that means for organisations using AI and the questions they should ask to determine whether their security framework is actually working in practice.
What Does AI Actually Change About Cyber Risk?
Contrary to common belief, AI is not making phishing, impersonation, or social engineering completely new problems. What it actually changes is how quickly attackers can create convincing content, how easily they can personalise it, and how many people they can target at once.
This idea was captured by Quentin Sorneau during our webinar, where he stated that “AI does not create cyber threats. It INDUSTRIALISES them.”
Quentin further explained that this industrialisation changes three things in particular.
- Speed, because threats can be created and adapted much faster.
- Scale, because attackers can reach more targets with less effort.
- Accessibility, because sophisticated techniques become easier for less-skilled attackers to use.
We can already see this shift in the evidence, as discussed below:
AI Makes Familiar Threats Faster and Easier to Scale
In fact, the latest Microsoft Digital Defense Report 2025 provides further evidence of how AI is making familiar cyber threats faster, more scalable, and easier to execute. The report examines the changing cyber threat landscape and how attackers are using AI to strengthen existing techniques.
Microsoft found that AI-automated phishing emails achieved a 54% click-through rate, compared with 12% for standard phishing attempts. It also reported that AI automation could increase the profitability of highly targeted phishing by up to 50 times.
These findings clarify Quentin Sorneau’s point about AI industrialising existing threats.
Phishing itself is not new, but AI can significantly reduce the effort required to create convincing, targeted attacks and deploy them at scale. The threat therefore becomes more efficient without necessarily becoming fundamentally different.
AI Can Make Social Engineering More Convincing
When an attack depends on trust, making a message feel familiar can make it much harder to question. This is where AI can make social engineering more convincing, because attackers can use it to create messages that sound more like the people or organisations their targets already know.
The Europol report on the criminal use of ChatGPT provides further evidence of this shift. The report examined how large language models could be misused for criminal activity. It further noted that they can generate realistic phishing messages and reproduce the language patterns of particular people or groups. This helps clarify why AI can make familiar social engineering techniques more difficult to recognise.
An attacker is no longer limited to sending a generic message and hoping someone responds. AI can help make the communication feel more relevant, more familiar, and more consistent with the person or organisation being impersonated.
From our perspective, this is where the risk becomes particularly significant. That’s because social engineering has always relied on people making decisions based on what appears trustworthy.
When AI can help an attacker imitate someone’s tone and personalise a request using publicly available information, that trust becomes easier to manipulate. We therefore see AI as strengthening the human element of social engineering rather than replacing it. The technology makes it easier for attackers to exploit the trust that these attacks already depend on.
AI Can Make Trust Easier to Manipulate
During the seminar, we discussed how AI can take the manipulation of trust beyond emails and messages. Quentin Sorneau mentioned a 2024 incident in Hong Kong to show how this can happen when deepfakes are used to influence an actual business decision.
According to the Hong Kong Government’s account of the incident, an employee was deceived during a video conference involving deepfake representations of company executives. The employee then authorised transfers totalling approximately HK$200 million. The attackers had used publicly available video clips and an impersonated voice to create the deception.
Quentin explained that the attack exploited trust rather than the systems themselves. He described it as a trust failure in which AI influenced people’s decisions while the security system could only detect an anomaly afterwards.
He also pointed to the human element in cyber breaches, citing Verizon’s breach research and noting that around three in four breaches involve people.
We agree with this perspective because it shows why AI-related security cannot be treated as a purely technical challenge. When technology is being used to influence human behaviour, organisations also need to consider how their people recognise risks, verify requests, and make decisions under pressure.
This is where ISO 27001 AI security becomes more than a question of technical controls. It also needs to support the people and decision-making processes that determine what happens when a threat is detected.
Why Is Cybersecurity Becoming a Decision-Making Problem?
AI does not just make cyber threats faster and more convincing! It also increases the amount of information organisations need to assess when deciding how to respond.
In fact, the Microsoft Digital Defense Report 2025 states that Microsoft processes more than 100 trillion security signals every day. While this figure reflects Microsoft’s own security operations, it illustrates the scale of information modern security teams can encounter.
At this scale, detecting a potential threat is only the beginning. Security teams still need to determine which signals matter, understand the risk involved, and decide what action should follow. This was the thinking behind another of Quentin Sorneau’s key observations during the webinar:
“Security is no longer a detection problem. It is a decision problem.”
Detection Does Not Tell You What to Do Next
AI does not just make cyber threats faster and more convincing. It also increases the amount of information organisations need to assess when deciding how to respond.
Sorneau pointed to current industry data during the webinar to illustrate how large that challenge has become. Referring to Microsoft’s Digital Defense Report 2025, he highlighted that more than 100 trillion security signals are analysed every day. At that scale, detecting suspicious activity is only the beginning. Security teams still need to determine which signals matter, how serious they are, and what action they require.
Quentin also drew attention to the pressure this creates for security teams by mentioning the Research on alert fatigue from IBM. The research stated that important issues can be missed, and responses can be delayed when teams face excessive or repetitive alerts.
The wider threat environment adds another layer to this decision-making challenge. ENISA’s Threat Landscape 2025 analysed 4,875 incidents between July 2024 and June 2025, highlighting the range of threats organisations may need to assess. This reinforces Quentin’s broader point that security teams are making decisions across an increasingly complex threat environment.
This is the point we want to emphasise. More detection does not automatically mean better security decisions. A security team may identify suspicious activity quickly, but the organisation still needs to determine how serious the threat is, what could be affected, and who has the authority to respond. Detection gives you information. Decision-making determines what happens next.
Decision-Making Needs Clear Ownership
Quentin also highlighted decision latency during the webinar. He mentioned that a threat can be detected quickly. However, the response may still be delayed when people are unclear about
- Who should assess the risk
- Who should make the decision
- When an issue needs to be escalated
Quentin described the potential result as a “circle of procrastination”, where different stakeholders become involved but responsibility for the final decision remains unclear.
This is why we believe the decision-making problem is ultimately an organisational one. Security teams need defined responsibilities and escalation processes so that identified risks lead to decisions and those decisions lead to action.
Where Does ISO 27001 Fit Into This Decision-Making Problem?
If cybersecurity is about making the right decisions, organisations need a structured way to manage those decisions. This is where ISO 27001 comes into the discussion.
During the webinar, Quentin Sorneau explained that ISO 27001 is not simply a technical security framework. ISO 27001 is a management system designed to help organisations structure how they identify, assess, and manage information security risks.
At its core, ISO 27001 AI governance follows a continuous Plan-Do-Check-Act (PDCA) cycle where:
- Organisations identify what needs to be protected
- Implement appropriate controls
- Check whether those controls are working
- Improve them as risks change
Quentin’s point was that this structure already exists in many organisations. However, the real question is whether organisations are using it to support active security decisions or simply maintaining it for certification and audits.
We believe that distinction is critical in an AI-driven environment because AI can change the speed and complexity of risks. However, the ISO 27001 AI security framework can provide the structure organisations need to assess those risks and decide how to respond.
But Having the ISO 27001 Framework Does Not Guarantee Security
An ISO 27001 AI governance framework gives organisations a structured way to identify information security risks, put appropriate controls in place, and continually improve how those risks are managed. It can therefore give organisations a stronger foundation for making consistent security decisions.
However, it is easy to assume that simply having the framework in place means the organisation is secure. That is where the distinction between having an Information Security Management System and actually using it becomes important.
Sorneau explained during the webinar that certification shows that an organisation has established a management system that meets the requirements of the standard. It does not, by itself, guarantee that the organisation is effectively managing every security risk in practice.
We believe this is where organisations need to look beyond the certificate itself. An ISMS has value only when it becomes part of
- How security decisions are made
- Risks are reviewed
- Controls are followed in day-to-day operations
Consider what happens when an organisation prepares for an audit by reviewing its policies and updating its risk register, but does not maintain the same discipline afterwards. The organisation may still have a certified ISMS, yet the way security risks are actually managed can change significantly between audits. That is the distinction Quentin’s argument brings into focus.
Where Does ISO 27001 Lose Its Impact?
As discussed above, having an ISMS does not mean that every security decision will be made effectively. What matters is what an organisation does when a risk needs to be resolved.
A solid approach should make it clear who takes the decision, how the risk is weighed against other priorities, and what happens once the decision has been made. When those steps are unclear, security risks can remain unresolved even when the right processes are documented.
That is the gap Quentin Sorneau pointed to during the webinar. He identified three areas where organisations can struggle to turn security governance into action:
- No ownership means no real decisions
- No risk discipline means no arbitration
- No execution means controls are not applied
No Ownership Means No Real Decisions
When a risk needs to be resolved, someone needs to have the authority to make that decision.
An AI-related risk may involve security, IT, compliance, legal, and business teams. Each may contribute to the assessment. However, the risk can move between teams without being resolvedif ownership of the final decision is unclear. That’s exactly where ISO 27001 for AI loses its impact.
No Risk Discipline Means No Arbitration
Once ownership is clear, decision-makers still need a way to determine which risks deserve attention first.
Security teams often have competing risks and limited resources. Without consistent criteria for assessing severity and business impact, priorities can become driven by urgency rather than risk.
So, simply having the ISO 27001 framework for AI is not enough. Your organisation also needs to have risk discipline, as it can give your organisation a basis for making those trade-offs. It helps determine what needs action, what can be accepted, and what should be escalated.
No Execution Means Controls Are Not Applied
A resolved risk still requires action.
If the organisation decides to restrict access, change an AI-related process, or introduce a new control, that decision needs to reach the teams responsible for implementing it. Otherwise, the risk assessment and the decision remain separate from what happens operationally.
Quentin captured the exact underlying problem during the webinar by stating that “Responsibilities exist in documents, not in decisions.” That line connects the three gaps where ISO 27001 for AI loses impact.
Someone must own the decision, the organisation must have a basis for making it, and the resulting action must reach the operation.
Five Questions to Test Your AI Risk Readiness
The gaps Quentin identified do not mean that the ISO 27001 framework has limited value. They show what organisations need to get right if they want the framework to have a real impact on how security risks are managed.
ISO 27001 can become far more than a certification requirement when:
- Ownership is clear
- Risks are properly assessed
- Decisions reach the people responsible for acting on them
It can give the organisation a structured basis for making and managing security decisions as risks evolve. So how can you tell whether your organisation is actually using that framework effectively?
Quentin proposed five questions that can help organisations assess whether their approach to AI risk is working in practice:
1. Is AI risk formally owned at leadership level?
AI-related risks need clear ownership at the leadership level. Someone needs to be accountable for deciding how significant those risks are and what the organisation should do about them.
If that responsibility is unclear, even a well-defined ISO 27001 for AI approach can struggle when a decision needs to be made.
2. Does your ISMS explicitly cover AI usage?
AI may already be part of everyday work across different teams. The question is whether the organisation’s ISMS recognises that usage and the risks it creates.
An effective approach needs to reflect how AI is actually being used, rather than treating it as a separate technology issue.
3. Are your sensitive data classified and controlled?
AI also raises a question about:
- What information employees can use with AI tools
- What information must remain protected
Clear data classification and access controls help establish those boundaries before sensitive information is exposed through inappropriate AI use.
4. Are your suppliers assessed for AI-related risks?
AI-related risks can extend beyond an organisation’s own systems. Suppliers may use AI within the services they provide or handle information through AI-enabled processes.
Understanding those dependencies helps organisations account for risks that may otherwise sit outside their immediate security controls.
5. Have you tested an AI-related crisis scenario?
An organisation may have policies, assigned responsibilities, and documented response procedures. That does not tell you whether those arrangements will work when a real incident puts people under pressure.
Testing an AI-related crisis can expose gaps that are difficult to see on paper. It can show whether people know who should decide, when they should escalate, and what action they need to take.
The findings then give the organisation something practical to improve before a real incident puts those arrangements to the test.
What Does This Mean for Organisations?
The webinar’s broader message was not that organisations need another framework simply because AI has emerged. They need to make the frameworks they already have work together and influence real decisions. That means securing the organisation before trying to secure AI in isolation.
Strong governance, clear access and data controls, and defined responsibilities provide the foundation for effective AI governance.
Quentin also highlighted how different frameworks can complement one another. ISO 27001 provides the broader information security management structure, while approaches such as NIST Zero Trust can support more specific technical and access-control decisions.
The same principle applies to AI governance. Organisations should not treat AI security as a separate exercise disconnected from their existing security and governance practices. The goal is not to collect more frameworks. It is to create a system that helps people make better security decisions and act on them.
Training also has an important role to play. A framework can provide the structure, but people need the knowledge to interpret its requirements, assess risks, and apply them to situations they actually face.
That is why practical understanding matters as much as knowing the requirements of ISO 27001. At Grow Skills Store, learners can build that understanding through self-paced courses or live group sessions.
Those who want more personalised guidance can also access 1-to-1 coaching with ISO expert Quentin Sorneau. Certification support and exam preparation are also available as part of the learning journey.
The goal is not to collect more frameworks. It is to build the capability to use them when security decisions actually matter.
