How to prepare for a deepfake attack
In February 2024, an employee at Arup, the British engineering firm behind the Sydney Opera House, joined what looked like a routine video call with his CFO and several senior team members. By the end of it, he had authorised fifteen transfers totalling £20 million.
Every person on that call, except him, was fake. Not photoshopped. Not badly lip-synced. Fully AI-generated, moving and speaking in real time, indistinguishable from people he had worked alongside for years.
I think about that call often. Not because of the figure, though the figure is staggering. I think about it because it proves that the thing standing between a company and a catastrophic loss used to be a person’s judgment. That’s no longer enough. When the fraud looks exactly like your colleague, sounds exactly like your colleague, responds in real time like your colleague would, judgment has nothing left to hold onto.
Now, I know what you’re thinking. Arup is a huge company with a huge budget. Surely scam artists are going after companies like that, not wasting their time on startups. I run a deepfake detection company, and am here to tell you that’s not the case. Here’s why …
The numbers are not vibing
According to a recent report, 76% of UK businesses have already faced some kind of deepfake attack. That same report found that only 40% feel even slightly prepared for the next one.
Most of these stories never make the news. They happen to smaller, quieter companies, get written off internally as ‘that weird thing that happened in March,’ and filed under one-off embarrassment instead of what it actually is: a pattern.
But we’re just a startup, why would anyone bother?
A lot of time you’re actually not the primary target. What you are, though, is the side door. A way to easily access those bigger companies who have made the headlines.
Going after large companies directly is genuinely hard as they tend to have security teams, legal departments, the works.
But, what I’ve seen increasingly more over the last year is criminals looking sideways at the smaller startups these large companies work with. For example, you may be the agency producing their marketing videos, or the dev shop with admin access to their systems.
Because startups have less money to spend and focus on security, they’re unfortunately a lot easier to compromise. Once attackers are inside your systems, they can use that trust and access as a stepping stone into much larger organisations.
Here’s an uncomfortable truth. You may actually already be helping these criminals without knowing it. If you’ve signed up for a new and unvetted AI app recently, or recorded a voice note so some chatbot would sound more natural, you’re putting yourself at risk by handing over the raw material these criminals need. You’re giving them your voice. Your face. The little verbal tics that make you sound like you on a call.
None of it felt sensitive at the moment. It felt like a regular Tuesday. But it’s exactly the type of material needed to build a version of you convincing enough to fool your own team.
Why this can’t wait
Most businesses think of security as locks on doors. You set up a firewall, you encrypt sensitive files, maybe you’ve got rules and regulations about who on your team can access what. All genuinely good cybersecurity moves to have.
None of it helps with this, however, because a deepfake isn’t trying to break into your systems at all. Instead, it’s trying to get a real person to trust it. Specifically, it’s betting that your finance team will believe the face and voice on their screen really is your boss, and just do what they’re asked at the time.
Voice and video used to be the standard for proving someone was who they said they were, precisely because faking them took serious time and money. But that just isn’t the case anymore. Anyone can do a decent job of it for next to nothing, and most workplace security training still hasn’t caught up.
What you can actually do about it
Glad you asked. Because, lots of preparation is absolutely the best way to remedy this.
Audit your AI tool stack: find out which tools your team is actually using. Not the official ones – the unvetted tools people picked up because they made life easier last Tuesday. Every voice note or video uploaded to these apps can be at risk of becoming training data for someone’s future scam.
Build a verbal codeword protocol: seriously, just agree on one. A phrase that confirms a request for money or access is genuinely coming from the person it claims to be from, and stick to it even if the face and voice on the call look completely real. Nobody should be authorising a transfer purely because a video call looked convincing.
Treat your digital likeness like a business asset: your founder’s conference talk, your CEO’s LinkedIn videos are no longer just marketing content that brings in new clients. It can also be attack material. Decide who’s actually responsible for keeping an eye on how it gets used.
Brief your investors and partners: they’re connected to you, which means they’re exposed through you, too. Getting everyone on the same page about what a real request looks like (and what one never looks like) closes a door that a lot of companies don’t even realise is wide open.
Build real-time detection into your verification stack: codewords and protocols are only as strong as the person remembering to use them under pressure, mid-call, with someone who sounds exactly like their boss. That’s the gap AI detection tools are built for: flagging a synthetic voice or face in real time, so your team has a second check that doesn’t rely on instinct alone, holding up against a convincing fake.
Final thoughts
Deepfakes aren’t coming. They’re already here, and three out of four UK businesses have already had a run-in with one. The real question is whether you’re the open door, or the one that’s already locked. So lock it.
