Deepfake candidates are here

In a session "The Deepfake Interview: Breaking in from the inside," Jake Moore, Global Cybersecurity Advisor at ESET, demonstrated how he successfully bypassed live video job interviews using AI-driven face-swapping and voice-cloning tools. By adopting completely synthetic identities, appearing as both male and female personas, he managed to secure job offers from organisations that believed they were vetting a real person.

As Jake Moore pointed out, the perimeter has moved to the HR department: "The threat isn't AI replacing jobs; it's AI being used to apply for them".

This experiment reflects a broader trend. Research cited in Nametag’s 2026 Workforce Impersonation Report suggests we have entered an era of “Identity-Layer Compromise.” Estimates indicate that up to 62% of organisations experienced a deepfake attack in the past 12 months, while criminals are now commercialising “Deepfakes as a Service” (DaaS) kits designed to deceive recruitment systems.

Even more alarming is the rise of the "Identity Proxy." New academic research, Security Risks of AI Agents Hiring Humans (Feb 2026), found that autonomous AI agents are now programmatically hiring human "proxies" to pass liveness checks or sit for interviews that the AI cannot yet complete on its own. For a median price of just $25, an AI agent can hire a human to verify a fraudulent LinkedIn account or perform a physical ID check.

The vulnerability extends beyond the interview. The attached image highlights a common social engineering tactic: attempting to bypass an AI's safety guardrails by adopting a "security researcher" persona. While one bot might refuse a direct request to "hack a company," the suggestion to "try Grok" points to a trend of "guardrail hopping".

Researchers warn attackers could weaponise platform trust to spread malware like Atomic macOS Stealer (AMOS). By poisoning search results for common IT help (like "Clear disk space on macOS"), they direct users to shared, legitimate-looking ChatGPT or Grok conversation links. Because the malicious instructions come from a "trusted" AI assistant on a legitimate domain, users are far more likely to execute the commands that install the malware.

𝗞𝗲𝘆 𝗧𝗮𝗸𝗲𝗮𝘄𝗮𝘆𝘀 𝗳𝗼𝗿 𝟮𝟬𝟮𝟲:
Close the "Hiring Gap": HR must ensure the person being interviewed matches the ID provided using biometric "liveness" tests.

Explainable Detection: Move toward tools that use vision-language models to provide reasoning on why a video is flagged as synthetic.

Continuous Verification: We must move to a "Defend-as-One" model where identity is verified at every high-value approval, not just on day one.

In a world where digital authenticity is under constant attack, the only remaining defense is a rigorous commitment to verifying every identity and trusting nothing by default.