What has actually changed with AI-powered attacks?
AI has not invented new attack categories. It has made three existing ones far cheaper: convincing voice and video impersonation, high-quality phishing at volume, and the speed at which a newly disclosed vulnerability gets weaponised. The defences that work are still identity, patching and verification — applied faster.
Three shifts the data actually supports
Voice phishing went from niche to second place
Mandiant’s M-Trends 2026 records voice phishing as the second most common initial infection vector at 11%, a sharp rise, while email phishing fell to 6%. Cloned voices make the help desk the soft target: an attacker who sounds like your finance director asking for an MFA reset is a different problem from a badly written email.
The exploitation window collapsed
Verizon’s 2026 Data Breach Investigations Report describes the window between disclosure and exploitation compressing from months to hours, with vulnerability exploitation now the leading breach entry point at 31%. Mandiant puts the mean time to exploit for zero-days at minus seven days — exploitation happening before patches ship.
Your own staff became an attack surface
The same DBIR reports shadow AI use by employees rising from 15% to 45% year on year. Corporate data pasted into unsanctioned tools is not a hacking problem; it is a governance problem that creates a breach surface no firewall sees.
What has not changed
This is the part vendors rarely lead with. AI has not broken cryptography, defeated hardware-backed MFA, or removed the need for an attacker to obtain credentials or exploit a flaw. The entry points are the ones you already know about. What changed is that the attacker’s cost per attempt fell close to zero, so the volume of attempts rose and the quality floor rose with it.
The correct response is therefore rarely to buy an AI security product. It is to remove the assumptions that only held because attacks used to be expensive.
Five defences that hold up
1. Phishing-resistant MFA, not just any MFA
Push notifications and one-time codes can be relayed in real time. FIDO2 security keys and passkeys are bound to the origin, so a convincing replica site cannot harvest anything reusable. Start with administrators and finance.
2. Out-of-band verification for anything that moves money or access
A voice is no longer proof of identity. Payment changes, MFA resets and privilege grants need a second channel and a known-good contact record — never a number supplied during the request itself.
3. Treat the help desk as a security control
Most voice phishing succeeds at the service desk. Define what the desk may do on a phone call alone, and make escalation the default rather than the exception. This costs nothing and closes the most-used door.
4. Shrink the patch window on internet-facing systems
If exploitation arrives in hours, a monthly cycle for edge devices is an accepted breach. Separate the SLA for internet-facing infrastructure from everything else, and automate deployment so the bottleneck is testing rather than ticketing.
5. Give people a sanctioned AI option
Shadow AI grows where there is no approved alternative. An enterprise tool with data controls, plus a clear statement of what may and may not be pasted into it, works better than a ban everyone quietly ignores.
How to tell whether you are actually exposed
Ask three questions. Could an attacker who clones an executive’s voice obtain an MFA reset from your help desk? How long does a critical patch take to reach your VPN concentrator? And where is company data going when your staff use AI tools? The honest answers usually point at process, not technology.
Where Rivo fits
Our cybersecurity services cover identity and access management, Zero-Trust implementation and the GRC work that makes these controls auditable. Where the fix is faster, safer patching, our DevOps services team automates the delivery path. Talk to our engineers about your exposure.
