Hackers Abuse Google Sites to Host Fake OpenAI Codex Download Pages

Cybercriminals are using Google Sites to host fake download pages for OpenAI Codex, turning a familiar search into a malware trap.

The campaign targets macOS users and relies on paid search placements to steer them toward a convincing but fraudulent installer page.

The page does not simply deliver a harmful app. It tells visitors to copy a command, open Terminal, and run it themselves.

That approach, known as ClickFix, shifts the final execution step to the victim and can bypass the suspicion normally raised by an unexpected download.

Cato analysts identified the activity after tracking sponsored results for searches including “codex macos download.”

Cato said in a report shared with Cyber Security News (CSN) that the chain has strong overlap with Atomic macOS Stealer, also called AMOS, delivery activity. The risk reaches beyond one spoofed developer tool.

By combining a paid advertisement, a legitimate Google-hosted page, and a recognised software brand, operators build a route that looks trustworthy at every early stage. Similar malicious Google Ads campaigns have shown how paid search can expose macOS users to harmful software.

Hackers Abuse Google Sites

The sponsored result appeared above the genuine result and sent visitors to a Google Sites page designed to resemble a Codex download portal.

It displayed download choices for macOS and Linux, although researchers observed active payload delivery only for macOS users. Google Sites acted as the visible front door rather than the place serving the main malicious content.

The page loaded an attacker-controlled iframe, allowing the operators to retain a trusted-looking Google address while changing the content behind it. This separation also makes takedowns and analysis more difficult.

One newer infrastructure set used a useful evasion trick. Its live ClickFix content was placed at “/codexx/”, while the expected “/codex/” address returned a harmless-looking product page.

Visitors using a non-macOS device also received benign content, frustrating automated scanners and investigators. The campaign shows why search results should not be treated as proof of authenticity.

Users should obtain developer tools through verified vendor channels, inspect the destination before downloading, and never paste an unfamiliar installation command into Terminal. The warning is especially relevant after fake Node.js installer ads used sponsored search results to reach potential victims.

ClickFix Chain Delivers macOS Payload

The fake installer starts with a legitimate-looking Codex npm command, then hides a Base64-encoded address that retrieves a script and pipes it into zsh.

The victim sees what appears to be a normal setup step, but running the command begins a three-stage delivery chain. The first script decodes an embedded component, which produces a second script.

That stage records an “event=pasted” signal, downloads the final Mach-O file to “/tmp/helper”, clears its extended attributes with “xattr -c”, marks it executable, and launches it. Removing those attributes can reduce the warning context associated with downloaded files.

Researchers found overlap with AMOS delivery methods, including encoded curl loaders, hidden zsh stages, telemetry before payload retrieval, and the same temporary staging location.

It does not prove that every final payload is identical, but it gives defenders meaningful evidence for incident hunting. Recent macOS ClickFix attack reports illustrate the broader threat from prompts that persuade users to run commands.

Security teams should review logs for shell commands that decode Base64 data, curl piped to zsh, new executables in “/tmp/helper”, attribute removal, and unusual outbound requests immediately after Terminal activity.

Network controls should also block the listed lure, iframe, telemetry, and payload-delivery infrastructure, while detections should combine these behavioural signs instead of relying on a single domain.

The operators have already reused portions of their infrastructure while rotating domains, payload locations, and brands, including a related Claude Code-themed landing page.

That flexibility means the indicators below are best used with behavioural monitoring. Users can further reduce exposure by treating sponsored results cautiously, checking the publisher’s official download path, and rejecting any page that asks them to paste a command as a “fix.”

Indicators of compromise (IoCs):-

Note: IP addresses and domains are intentionally defanged (e.g., [.]) to prevent accidental resolution or hyperlinking. Re-fang only within controlled threat intelligence platforms such as MISP, VirusTotal, or your SIEM.

Prevent incidents due to slow investigations. Power your Tier 1 with threat intelligence from 15K SOCs: Integrate TI Lookup in your SOC