How AI deepfakes are upending cyber safety
Cybersecurity used to be a battle of technical barriers, such as firewalls, antivirus software and complex passwords. But today, the attack surface has shifted from systems to humans as the rise of artificial intelligence has made deepfake technology, including synthetic audio, video and imagery, accessible to virtually anyone. As such, at Wirebox, we’ve been educating our clients about how to manage the new risk of highly sophisticated cyber threats. It’s clear businesses can’t rely solely on traditional security awareness training anymore. Here’s why:
Deepfake threat landscape
Deepfakes leverage machine learning algorithms to map and replicate a person’s voice, face or physical mannerisms with precision. While early iterations were easy to spot, modern AI tools require only a few seconds of public audio or video to generate fairly convincing clones of real people. And attackers are deploying these deepfakes for:
Executive impersonation
Cybercriminals clone the voice or video of a senior executive to authorise emergency wire transfers or data releases.
Identity fraud
Malicious actors combine stolen personal data with AI-generated faces to bypass online KYC and identity verification checks to open accounts or access funds.
Social backdoor
Fraudsters impersonate your internal IT staff to trick employees into resetting credentials or granting remote access so they can get into your systems.
Why traditional security awareness falls short
For years, security training taught staff to look for obvious red flags, such as misspellings in emails, unusual sender addresses or pixelated logos. But deepfakes bypass these cues entirely by exploiting your familiarity with someone. And it makes sense…when an employee receives a phone call from what sounds exactly like their manager requesting an urgent transfer, emotional pressure often overrides caution.
3 defences against deepfake attacks
Protecting your organisation against synthetic media requires moving beyond technical filters. You need proper controls like:
- Verification protocols – No single channel should ever be trusted for critical actions. Establish mandatory, multi-person authorisation processes for financial transactions, credential resets or access requests. If a request arrives via voice or video, staff must verify it using a secondary, pre-agreed channel, like calling back on a verified internal number.
- Challenge-response words – For high-risk roles, such as finance or system administration teams, introduce confidential, offline verification protocols. Using a shared, pre-established phrase or codeword during unusual requests ensures that even an exact audio clone cannot complete the exploit.
- Identity controls – Ensure your access control frameworks rely on hardware-backed authentication rather than biometric or SMS verification alone. Deploying passkeys and FIDO2 hardware keys prevents attackers from gaining entry even if they manage to trick a user into approving a session.
Building zero-trust human operations
Tech alone cannot completely block synthetic media, especially when attacks occur over public telecoms channels. So, cyber safety now requires extending Zero Trust (never trust, always verify) from network architecture into everyday human interactions in the workplace. By combining strict verifications, approval processes and modern hardware authentication, you can better safeguard your assets and stay ahead of these AI scams.
