Insights

AI isn’t creating the threat. It’s removing the barriers.

Bryan Poggo
Bryan Poggo
Chief Technology Officer
Intelligence
Security
Technology

A few weeks ago, I wanted to understand how accessible voice-cloning technology had become.

I took publicly available video of a known person speaking, extracted the audio and fed it into a widely available AI tool. Within minutes, I had a synthetic version of their voice convincing enough to pass as that person on a phone call.

I didn’t need specialist equipment. I didn’t need to be a deepfake expert. I needed a laptop, publicly available source material and the knowledge that it could be done.

The most concerning part wasn’t the quality of the result. It was how little time, expertise and effort were required to produce it.

That experiment reinforced something I believe the protective-security industry needs to confront.

AI is not creating an entirely new threat environment. Impersonation isn’t new. Social engineering isn’t new. Open-source intelligence, surveillance and targeted reconnaissance aren’t new either.

What AI is doing is removing many of the barriers that previously limited those activities and enabling malicious actors to move with greater speed and scale.

That changes the security equation for defenders, adversaries and the people we protect.

AI has lowered the barrier to capability

Until recently, building a detailed intelligence picture around an executive, public figure or high-net-worth individual could require considerable time and capability.

Relevant information might be scattered across company records, professional profiles, property information, media coverage, photographs, travel references, social networks and the online activities of family members and associates.

Finding the information was only the beginning. Someone still had to validate it, connect it and understand what it revealed.

AI compresses that process.

It allows large volumes of information to be searched, collected, and organised far more quickly and accurately.

For protective-security teams, this creates a genuine opportunity.

AI can help analysts cast a wider net across more sources, collect large sets of data, and process information at a scale that would be extremely difficult to manage manually. It can surface anomalies, changes and potential relationships that warrant investigation.

But we should be precise about what that means.

This is not a science-fiction version of predictive security in which an algorithm reliably tells us who will become a threat or when an incident will occur.

It is earlier warning.

If technology helps us identify a relevant change sooner, analysts and protective teams have more time to establish context, assess credibility and determine whether intervention is required.

The value isn’t removing analysts from the process. It is allowing them to spend less time searching for information and more time determining what it means so decisions can be made with more confidence.

The same advantage is available to the adversary

The same technology that enables a security team to collect and use information faster can help a low-skilled adversary turn scattered information into a profile of an executive, identify their relationships and routines, and potentially use that information to cause harm. This is particularly concerning because the technology can shorten the path from grievance or criminal intent to action.

Among the most troubling uses of the technology is AI-enabled fraud. Government agencies are already warning about the operational use of synthetic media. The FBI has documented campaigns involving AI-generated voice messages impersonating senior officials, while New Zealand’s National Cyber Security Centre has warned that deepfake audio, video and live calls can be used to induce people to transfer money or disclose information. The speed at which these crimes can occur can also significantly reduce the time available for banks, law enforcement and victims to intervene.

This matters because what we once trusted can no longer be relied upon.

The protective response can no longer depend solely on teaching people to recognise a fake.

Hearing a familiar voice is no longer proof of identity.

Seeing someone on a video call is no longer proof either.

A well-written message does not establish that it came from the person whose name appears at the top.

The protective response can no longer depend solely on teaching people to recognise a fake. Synthetic content will continue to improve, and detection methods will not be equally effective against every model, format or attack.

The more durable response is creating assurance.

Sensitive or unusual requests involving money, travel, access, personal information or security arrangements should be independently verified through a trusted channel. That might mean calling a known number, using an agreed verification phrase or escalating the request to a designated security contact.

The objective is not to make people experts in identifying deepfakes. It is to ensure that a convincing deepfake alone cannot authorise a consequential action.

Your digital footprint is now a physical-security issue

This is where protective security needs to evolve most quickly.

Most people understand the individual elements of their digital footprint. They know that their LinkedIn profile is public. They know that photographs exist online. They know that company announcements may identify where they work, what they are involved in and where they will appear publicly.

What is less obvious is the intelligence picture created when these fragments are combined.

One photograph may reveal very little.

A company announcement may reveal very little.

A family member’s social media post, sporting membership, property record, geotag or travel reference may reveal very little in isolation.

Together, they can expose relationships, routines, likely locations and periods of vulnerability.

In addition, Australian intelligence authorities have also warned that social and professional networking platforms are used to identify, cultivate and target people with access to sensitive information.

Protective security therefore cannot treat digital exposure and physical safety as separate disciplines.

A threat may begin with online research but ultimately manifest through unwanted contact, impersonation, coercion, surveillance, fraud or physical approach.

The front gate is no longer the beginning of the security perimeter.

The principal is not the only target

An executive or high-profile individual may be well protected personally and still remain exposed through the people around them.

The effective attack surface can include partners, children, assistants, household employees, business colleagues and other trusted contacts.

An adversary does not necessarily need to compromise the principal directly.

An assistant may hold access to calendars and travel arrangements. A family member may unintentionally disclose a location. A social connection may respond to an apparently legitimate request. Public information associated with a school, sporting club, property or business relationship may reveal links that were never intended to form part of a security picture.

This is not a reason to treat everybody around a client as a vulnerability. It is a reason to include them appropriately in the protective model.

Protecting an individual while ignoring the exposure of their wider ecosystem creates a significant blind spot.

Physical protection cannot compensate for every weakness in digital identity, personal information, communications and trusted relationships. Modern protective security must understand how those elements intersect.

Clients are also creating their own AI attack surface

The third dimension is the one I believe receives the least attention.

It isn’t only about how security teams use AI or how adversaries use it. It is also about how executives, families and organisations use AI themselves.

AI assistants are being connected to email, documents, meeting recordings, calendars, business systems, home technology and highly personal information.

The productivity benefits are real, and I support responsible adoption. But greater convenience should not come at the expense of scrutiny.

Not all AI services provide the same controls. Consumer and enterprise platforms may differ substantially in how they handle organisational identity, access, retention, administrative control and customer data.

The useful questions are therefore not simply, “Do we use AI?” or “Is AI safe?”

The questions are:

• What information are we giving it?

• What systems can it access?

• What permissions have been granted?

• What happens to the information after it is submitted?

• Who controls that information?

• What would the consequence be if the account, integration or underlying service were compromised?

Nobody would give a third party a master key to every room in their home simply because it made one task more convenient. They would consider what access was required, who held the key and what could happen if it were copied or misused.

We should apply the same logic to AI.

As these systems become more deeply integrated into personal and professional life, their access forms part of the wider protective-security assessment.

AI can identify signals. It cannot be treated as the decision-maker.

There is a temptation to overstate AI at both ends of this debate.

AI can identify patterns, anomalies and correlations across large volumes of information. But a signal is not the same as a threat. Apparent relationships or unusual behaviour still require context before they can meaningfully inform a protective response.

That distinction matters because protective decisions have consequences.

Experienced practitioners still need to assess intent, capability, credibility, context and consequence before determining what a piece of intelligence means and what response is proportionate.

AI itself also requires ongoing monitoring. NIST notes that deployed AI systems can exhibit variability and unpredictable behaviour, and identifies functionality, security, human factors and operational performance as areas requiring post-deployment monitoring.

AI should therefore support professional judgement, not become an unchallengeable authority.

The future is not AI replacing the protective analyst.

It is protective analysts using AI to see more, sooner, while remaining accountable for the decisions that follow.

What executives and families should do now

1. Understand what can be inferred, not only what can be found

A basic search of someone’s name is no longer enough.

A proper digital-footprint assessment should examine the information available about the individual, their family, residences, businesses, vehicles, routines and key relationships.

The assessment should then ask a more important question:

What could a motivated person infer by combining it?

That difference, between available information and actionable intelligence, is now central to understanding exposure.

2. Stop treating voice and video as identity

Recognising someone’s voice or seeing their face should not be sufficient to approve an unusual or sensitive request.

Executives, families, assistants and finance teams should have independent verification procedures for requests involving payments, account changes, travel, access or urgent security matters.

The agreed procedure matters more than any particular deepfake detector.

3. Protect the ecosystem, not only the principal

Protective assessments should consider the people around the individual and the access, information and influence each relationship carries.

This does not mean applying the same controls to everyone. It means identifying where indirect exposure could undermine otherwise strong personal protection.

4. Connect cyber, intelligence and physical security

The boundaries between these functions are increasingly artificial.

An online threat may require a physical protective response. A physical incident may reveal information that should alter cyber monitoring or protective-intelligence priorities.

If those functions operate in isolation, important context can be lost at the point where it matters most.

5. Treat exposure as dynamic

A protective assessment shouldn’t become a document that is updated once a year and then forgotten.

A new board appointment, property, relationship, school, transaction, public appearance or travel pattern can materially alter an individual’s exposure.

Protective arrangements need to evolve as the individual, the technology and the threat environment change.

Protective security has moved beyond the physical perimeter

AI isn’t replacing protective security.

It is changing the environment in which protection takes place. Executives and families now face a far wider range of threats than physical violence alone, from digital surveillance and reputational attacks to intrusions into their accounts and devices. Traditional protective measures remain essential, but they are no longer sufficient on their own.

Protective security no longer begins at the front gate, inside the vehicle or with the person standing beside an executive.

It begins with understanding what information exists about an individual, what can be inferred from it, who can access it and how it could be used. That now includes understanding how AI changes the attack surface, changes what we can trust and accelerates the speed at which information can become a security problem.

The executives, families and organisations that recognise that shift early will be better positioned to protect themselves from what comes next.

‍

Prepared to navigate today’s complex threat landscape? Contact SPG now to discover how our expertise can safeguard your people, assets and operations.

Get in touch