Zero-Click Attack

A zero-click attack compromises a device, account, or application without any action from the victim: no link clicked, no attachment opened, no prompt approved. The exploit rides inside data the target's software processes automatically — a message preview, an image parser, a missed call, a calendar invite — so the first and only "user interaction" is the device receiving the payload.

How it works

Modern software parses enormous amounts of untrusted input before a human ever sees it. Messaging apps render previews, mail clients parse attachments for thumbnails, phones process incoming call signaling. A zero-click exploit targets a vulnerability — frequently a zero-day — in one of those automatic parsers, achieving code execution the moment the malicious input arrives. Well-documented examples include mobile spyware delivered through messaging-app exploits against journalists and executives, where the infection left no visible message at all.

The term has also widened beyond exploit chains. Researchers now demonstrate zero-click paths in AI-enabled workplaces: a poisoned email or shared document carrying a prompt injection can trigger an AI assistant that automatically processes inboxes or files, exfiltrating data without the user touching anything. The common thread is the same — the attack targets what software does on your behalf, not what you click.

Zero-click attacks are expensive and comparatively rare, typically reserved for high-value targets: executives, admins, journalists, and diplomats. For most organizations they sit at the far end of a spectrum whose near end — one-click phishing — remains the overwhelmingly more common entry.

How to defend against it

• Patch aggressively. Zero-click chains die when the parser is fixed; rapid OS and app updates, especially on mobile, are the single highest-value control.
• Reduce the automatic attack surface. Mobile lockdown modes, disabling link previews, and stripping or detonating attachments at the gateway all shrink what gets processed unprompted — the same logic applies to constraining what AI assistants may open and act on autonomously.
• Protect likely targets differently. Executives and privileged users warrant hardened devices and monitoring — the same population logic a human risk score applies to phishing exposure.
• Keep the human layer strong anyway: attackers use zero-click exploits only when cheaper, click-dependent routes fail.