IBM released its 2026 Cost of a Data Breach Report on July 29. Every year the number goes up. Every year the security industry writes about it for two days and moves on.
This year has one finding worth actually sitting with.
One in four malicious breaches are now AI-enabled — a 56% jump in a single year. These attacks cost an average of $6 million to recover from, about $1 million more than the global breach average. But the dollar figure isn't the thing that caught our attention.
It's what kind of AI attack is doing most of the damage.
Not AI-generated malware. Not automated phishing at scale. The single most common form of AI-enabled attack — 45% of them — is deepfake impersonation. AI-generated voice or video of someone the target knows and trusts, asking them to do something they shouldn't.
What This Actually Looks Like
A caller sounds exactly like your executive director. They're calling from the right area code. They reference an internal project you're working on. They need you to wire $50,000 to a vendor to close a contract before end of day — the board approved it, they'll send the paperwork later, but the funds need to move now. The CFO who answered authorized a $243,000 transfer. Another finance director wired $499,000.
The audio wasn't constructed by a sophisticated criminal operation over weeks. Voice cloning now requires as little as three seconds of sample audio. The service to do it costs under $20 on dark web marketplaces. The raw material — recordings of your executive director on a donor webinar, a conference panel, a podcast interview, a LinkedIn video — is already public.
The FBI logged over 22,000 reports of AI voice and video scams in 2025, with losses approaching $893 million. The IBM data suggests 2026 is tracking significantly higher.
Why Small Orgs Are Squarely in the Crosshairs
There's a misconception that deepfake fraud is an enterprise problem. It isn't.
The attacks scale to the target. A 15-person NGO doesn't need to lose $6 million — it needs to lose $50,000 to face a genuinely existential crisis. The attackers know this. Wire fraud against small and mid-sized organizations is faster, lower-risk, and often more profitable per unit of effort than targeting large enterprises with hardened security teams.
And small orgs have exactly the profile that makes deepfake attacks easy to execute. Executive directors of nonprofits give talks, record donor videos, appear on community panels. Public sector managers present at conferences and record training sessions. SMB founders are all over LinkedIn and YouTube. Every one of those recordings is usable training data.
Roughly 40% of business email compromise attacks now incorporate AI-generated deepfakes — up from a negligible percentage just two years ago. Average losses from AI-augmented BEC now exceed $4.1 million. The attacks targeting smaller organizations hit proportionally smaller numbers, but the organizations they hit have no cushion.
The Defense Costs Nothing — But You Have to Actually Do It
Here's the frustrating thing: the strongest defense against deepfake voice fraud is procedural. It doesn't require a vendor contract or a security tool. It just requires changing one habit.
Never authorize financial transactions or sensitive access over voice or video alone.
That's it. If someone calls asking you to wire money, reset credentials, or approve access — regardless of how convincingly they sound — verify through a completely separate channel before doing anything. Call them back on a number you already have. Text them on a thread with history. Email them on an address you know. If they're really your ED or your CFO, they'll understand why you're calling them back. If they push back or claim there's no time, that's your signal.
Some organizations go further and establish a shared verbal "code word" — a rotating phrase that callers must provide to authenticate a high-stakes request. Feels a little cloak-and-dagger until you realize it's exactly what banks and intelligence services have used for decades.
A few practical steps for this week:
- Audit who on your team has financial authorization and make sure they know this protocol exists
- Add a line to your vendor payment process: no wire transfers authorized by phone call alone, regardless of who's calling
- Do a quick Google of your executive team — what's public? What could be cloned? Knowing your exposure is the first step to communicating about it
- If your team uses Microsoft 365 or Google Workspace, check whether your email security settings flag unusual wire transfer request patterns — most platforms have some version of this built in, most orgs never turn it on
The AI-for-Defense Angle
The IBM report has an underreported flip side: organizations that have deployed AI in their security operations cut breach costs by an average of nearly $2 million compared to those that haven't.
That's not theoretical. Tools like Abnormal Security use AI to detect anomalous communication patterns — including BEC attacks — that rule-based filters miss. AI-powered identity verification tools are getting cheap enough that small orgs can use them at the front door of their systems. Even something as basic as an AI-assisted email monitoring setup (we've deployed these for NGOs with no dedicated IT staff) catches a meaningful percentage of the attacks that used to get through.
The arms race is real. Attackers are using AI to run cheap, effective deepfake impersonation attacks. Defenders who use AI to monitor for those patterns have a measurable advantage. The IBM data just quantified what that advantage is worth.
What to Do Right Now
The threat is real and it's growing. But the response doesn't have to be complicated.
Start with the free stuff: establish a verification protocol for financial requests, brief your team on how deepfake calls work, and scan your external-facing content to understand your voice and video exposure.
Then, if you want to add an actual detection layer — tools that flag AI-generated phishing emails, suspicious wire transfer request patterns, and anomalous account behavior — that's a sprint. We've set it up for 12-person nonprofits with Microsoft 365 and for 40-person public sector teams on Workspace. The setup is not the hard part. The hard part is knowing which tools are worth the cost for an org your size, and which are enterprise products dressed up as SMB solutions.
The IBM report is a useful forcing function. Use it.