ClickFix: the paste-and-run attack
ClickFix has been around since early 2024, but it has surged in recent months. The attacker gets the victim to paste and run the malware themselves, which sidesteps the controls built for files and links. Here's how it works, why it slips past, and how to defend.
- Old technique, new scale: first seen in early 2024, but Microsoft saw user-pasted malicious commands on more than 1.1 million devices between February and early May 2026, roughly an 8x rise, ESET reports detections more than doubled from H2 2025 to H1 2026, and CrowdStrike cites a 563% rise in fake-CAPTCHA incidents in 2025.
- Mechanism: a fake CAPTCHA or error gives the page a user gesture to write the attacker's command to the clipboard. The victim pastes it into Run (
explorer.exe) or Terminal, and a built-in binary (powershell.exe,mshta.exe,certutil.exe, evenfinger.exe) fetches the next stage, often fileless. - Why controls miss it: no attachment, no exploit and no downloaded file for Mark-of-the-Web to flag. The parent is a user-driven
explorer.exeand the command line is usually obfuscated. - Payloads: infostealers (Lumma, AMOS, Macsync, Shub) and RATs (NetSupport, SectopRAT, ModeloRAT). Stolen sessions and saved passwords get around MFA.
- Detection: behavioural EDR and XDR such as CrowdStrike Falcon should catch this chain, so make sure the sensor is on every endpoint, prevention is set to block and alerts are triaged. Don't make it the only defence: it often fires after the command has started, so pair it with prevention, identity controls and training.
- Control points: disable Run, App Control on script hosts launched from it, PowerShell script block logging, browser-side disruption of the lure, egress filtering (including TCP 79), and RunMRU plus parent-child hunting.
- If hit: contain the device, treat it as an infostealer infection, revoke sessions from a clean device, and hunt persistence (Run keys, scheduled tasks).
What is ClickFix?
ClickFix is a social engineering technique that gets the victim to run the malware for the attacker. A page shows a fake problem, such as a CAPTCHA that won't load, a document that "can't be displayed" or a browser that has "crashed", and offers a quick fix: copy this command, paste it, press Enter.
Proofpoint traces its first criminal use to early March 2024. It spread quickly because it replaces the hardest part of an attack, getting code to execute on a protected machine, with a person doing it willingly. Between October 2024 and April 2025 Proofpoint saw groups tied to North Korea, Iran and Russia swap ClickFix into the install-and-run stages of their existing campaigns.
The name stuck, but the idea is wider than one prompt. Any lure that ends with "paste this into something" is the same attack, and the delivery keeps changing.
Not new, but surging
ClickFix has been around for a while. Proofpoint traces criminal use to early March 2024, and state-sponsored groups were adopting it by late 2024. What has changed in recent months is the volume and the range of lures.
| Source | Period | Finding |
|---|---|---|
| Microsoft Digital Defense Report 2026 (3 Nov 2026) | Feb to early May 2026 | Malicious pasted commands seen on more than 1.1 million unique devices, a roughly 8x increase (baseline not stated). Malicious copy-and-paste was 13% of initial access in Microsoft's Defender Experts data. |
| ESET Threat Report H1 2026 (8 Jul 2026) | H2 2025 to H1 2026 | Detections more than doubled. Lures widened from fake CAPTCHAs to AI-themed help pages, browser extensions and cloud authentication scenarios. |
| Rapid7 Q1 2026 Threat Landscape (21 May 2026) | Q1 2026 | ClickFix was 18.8% of observed threat activity, second only to abused remote monitoring and management tools (22.9%). |
| CrowdStrike (29 Sep 2026, citing its 2026 Global Threat Report) | 2025 | A 563% increase in incidents involving fake CAPTCHA lures. |
| Proofpoint (17 Apr 2025) | Mar 2024 onward | First criminal use in early March 2024, then adoption by North Korea-, Iran- and Russia-linked groups. |
Treat the figures with care: they measure different things (devices, detections, incidents, share of observed activity), so don't add them together. All four point the same way, up. In practical terms ClickFix has moved from a curiosity to a mainstream intrusion path that needs its own controls.
How it works
The flow is short, and almost every step is done by the victim.
- The lure. The victim lands on a page through a search result, a malicious ad, a compromised website or a link in a message. Some campaigns also pose as Microsoft security alerts or document-sharing pages.
- The fake problem. The page shows a CAPTCHA, an error or a "fix" that looks routine. SentinelOne describes fake CAPTCHAs that people click through out of habit.
- The clipboard. A script on the page quietly copies the attacker's command to the victim's clipboard. The person never sees what was copied.
- The instructions. The page tells them to open the Windows Run dialog (Win+R), paste and press Enter. On a Mac the same idea uses Terminal.
- Execution. A built-in tool such as PowerShell,
mshta.exeorcertutil.exeruns the pasted line and downloads the next stage. - The payload. Observed payloads include infostealers such as Lumma and remote access trojans such as NetSupport RAT and SectopRAT.
A recent Windows variant: CrashFix. In February 2026 Microsoft described a fake ad blocker extension in the Chrome Web Store. After a delay it froze the browser, then showed a bogus warning telling the user to paste a command into Run. The command abused the built-in finger.exe utility to fetch code. The final Python-based remote access trojan was only deployed on domain-joined machines, which suggests the attacker was picking out business networks.
On macOS. In May 2026 Microsoft reported fake "free up disk space" guides that give a Base64-encoded command to paste into Terminal. It installs stealers that take Keychain data, browser credentials and wallet keys.
Under the hood
The chain as a process tree:
browser user clicks "I'm not a robot" or a fake fix button
page script writes the command to the clipboard
explorer.exe Win+R, Ctrl+V, Enter (value lands in RunMRU)
powershell.exe | mshta.exe | cmd.exe | finger.exe (renamed copy)
stage 1 obfuscated one-liner fetches stage 2
stage 2 in memory, or dropped under %APPDATA%
payload infostealer or RAT, then persistence
- The click is the trigger. The fake CAPTCHA click gives the page a user gesture, which is typically what a browser requires before a script may write to the clipboard. The victim never sees the command that was copied.
- Living off the land. Microsoft says most chains end in PowerShell or HTA execution, with fileless payloads run through
msbuild.exe,regasm.exeandpowershell.exe. SentinelOne also listscertutil.exe. CrashFix copiedfinger.exeto a temporary folder and renamed it. - Obfuscation. Microsoft observed Base64, string concatenation, escape characters and nested execution, so the logged process command line is rarely readable. Script block logging (event ID 4104) and AMSI see the deobfuscated script; the command line doesn't.
- Anti-analysis and targeting. CrashFix's second stage looked for analysis and debugging tools and checked for domain membership before fetching its Python payload. The macOS campaigns skipped systems with Russian or CIS keyboard layouts, and one variant stopped in virtualised environments.
- Persistence. CrashFix's ModeloRAT ran through
pythonw.exeto avoid a console window, added a Run registry key, and in some infections created a scheduled task named SoftwareProtection that re-ran every five minutes. The macOS variants installed a LaunchAgent or LaunchDaemon disguised as an update component. - macOS specifics. Scripts run in Terminal don't go through the Gatekeeper signing and notarisation checks that apply to apps opened from Finder. The malware showed a fake helper prompt for the user's password, validated it locally, then collected Keychain entries, browser credentials and wallet data and sent a compressed archive out with
curl.
Why it slips past traditional controls
ClickFix isn't clever technology. It works because the controls most organisations rely on are built to stop files and links, and here the dangerous step is a person typing into a trusted Windows box. Microsoft notes that because a human launches the command, automated tools may not catch the first step.
| Control | Why ClickFix gets past it |
|---|---|
| Email gateway | Many lures never arrive by email. Malicious ads, compromised sites and search results reach the user directly, and email-based ones link out through redirectors to a page that looks legitimate. |
| Antivirus and file scanning | There is no attachment to scan. The command runs built-in tools, and payloads are often fileless, running in memory through binaries such as PowerShell. |
| Mark-of-the-Web and SmartScreen | These check files downloaded by a browser. A pasted command isn't a downloaded file, so those checks have nothing to inspect at that point. Commercial ClickFix kits even advertise SmartScreen bypasses. |
| EDR | The first process is started by the user through Run or Terminal, so it can look like an administrator doing something odd rather than malware. Block mode and behavioural rules help, but only if they are on. |
| MFA | MFA protects the sign-in, not a stolen session. Infostealers take saved passwords and session cookies, so the attacker can arrive already signed in. |
| Security awareness training | Most training says don't open attachments or click strange links. "Paste this into Run" doesn't feel like either, and a fake CAPTCHA feels routine. |
Two more points make it worse. Microsoft reports ClickFix builders sold on hacker forums for roughly US$200 to US$1,500 a month, with ready-made lures and anti-analysis checks, so the technique is cheap to run. And CrashFix only deployed its final payload on domain-joined machines, so business networks are a deliberate target, not collateral.
Recent activity
| Date | Source | What was reported | Platform |
|---|---|---|---|
| 29 Sep 2026 | CrowdStrike | A fake video-meeting error led a financial services employee to run a PowerShell and VBScript chain (STARDUST CHOLLIMA, July 2026). Fake CAPTCHAs on compromised Ukrainian sites drove another campaign (VOODOO BEAR, May to June 2026). | Windows |
| 6 May 2026 | Microsoft | Fake macOS utility guides deliver the Macsync, Shub and AMOS stealers. Apple added a Terminal paste warning in macOS 26.4. | macOS |
| 5 Feb 2026 | Microsoft | CrashFix: a fake ad blocker extension crashes the browser, then a fake fix delivers a Python RAT to domain-joined machines. | Windows |
| 21 Aug 2025 | Microsoft | Overview of delivery by phishing, malvertising and compromised sites; Lumma Stealer named the most prolific payload. | Windows, macOS |
| 22 May 2025 | SentinelOne | Fake CAPTCHA chains delivering Lumma, NetSupport RAT and SectopRAT. | Windows |
| 17 Apr 2025 | Proofpoint | North Korea-, Iran- and Russia-linked groups adopt ClickFix in existing campaigns. | Windows |
The pattern across all six: the lure changes (CAPTCHA, crash, disk-space guide, security alert), but the ask never does.
How to defend
You can't patch this, so the defence is layered: remove the easy paths, make the rest noisy, and be ready for the day someone pastes the line.
Behavioural EDR and XDR should catch this, but don't rely on it alone
Modern EDR and XDR with behavioural analysis, such as CrowdStrike Falcon, are your best chance of spotting ClickFix, because the chain produces behaviour that is hard to hide: Explorer starting a script host, an obfuscated download, a new Run key or scheduled task, access to credential stores, a bulk upload. CrowdStrike says Falcon Prevent and Falcon Insight XDR identify and prevent suspicious PowerShell, VBScript, process and command-line behaviour, and reports detecting likely intrusions from a 2026 ClickFix campaign through Falcon Complete MDR. Get the sensor onto every endpoint, including servers and Macs, with prevention policies set to block, and make sure a named person triages the alerts.
It still shouldn't be your only layer, for three reasons:
- It often acts after the command has started. By the time behavioural analytics fire, the pasted command is running and a stealer may already have read browser data. Prevention, such as removing Run and App Control, stops the first step.
- Attackers design around it. ClickFix kits are sold with advertised antivirus and SmartScreen bypasses, and payloads are fileless and obfuscated to look like routine admin activity. In a 2026 espionage campaign, CrowdStrike reported that the operators built checks into their workflow to find hosts without the Falcon sensor, so they could concentrate on unprotected systems.
- Coverage and tuning gaps. Unmanaged and personal devices, Macs without a sensor, policies left in detect-only mode and alerts nobody triages all leave holes.
Treat EDR and XDR as the detection and containment layer, and put the rest around it. CrowdStrike's own guidance has the same shape: disrupt the lure in the browser, stop script execution on the endpoint, detect credential abuse and lateral movement, correlate telemetry across domains, and respond before initial access becomes a wider compromise. Add prevention, identity controls (phishing-resistant MFA, short sessions, Conditional Access) and trained, reporting users.
Prevent
- Remove the Run dialog where it isn't needed. Microsoft recommends disabling Run (Win+R) and removing it from the Start menu, under User Configuration > Administrative Templates > Start Menu and Taskbar. Test first, since some admin and support routines use it. Standard users rarely do.
- Block native tools from launching out of Run. Use an App Control for Business or AppLocker policy so PowerShell and similar binaries can't be started from that dialog.
- Retire tools nobody needs. CrashFix abused
finger.exe, so block outbound TCP 79 and restrict legacy utilities. The same logic applies tomshta.exeandcertutil.exefor most users. - Harden PowerShell. Turn on script block logging so obfuscated commands are recorded, and set the execution policy to AllSigned or RemoteSigned. Treat the policy as a speed bump, not a security boundary.
- Run Falcon in prevent mode on every endpoint. CrowdStrike says Falcon Prevent and Falcon Insight XDR identify and prevent suspicious PowerShell, VBScript, process, command-line and behavioural activity at the point of execution. Check that prevention policies block rather than only detect, that sensors are healthy, and that servers and Macs are covered.
- Disrupt the lure in the browser. CrowdStrike's guidance is to break the chain before the paste, for example with a managed enterprise browser such as Falcon Seraphic Enterprise Browser, which watches web behaviour and interferes with the copy-and-paste step. If you don't run one, use a managed browser with enforced updates and click-time URL protection in your email gateway.
- Lock down browser extensions. CrashFix arrived as a fake extension, so allow-list extensions through Intune policy instead of letting users install their own. This is my suggestion rather than Microsoft's.
- Don't keep work credentials in the browser. Disable browser password sync on managed devices and use phishing-resistant MFA, so a stealer finds less to take. Identity threat detection, such as Falcon Identity Threat Protection, helps catch credential abuse and lateral movement if something does leak.
- On Macs, make sure they run macOS 26.4 or later, where Terminal warns on a risky paste, and that the Falcon sensor is on them too.
Detect
Start with what your sensor already does. CrowdStrike says Falcon Prevent and Falcon Insight XDR flag suspicious PowerShell, VBScript, process, command-line and behavioural activity, and that Falcon Next-Gen SIEM can correlate endpoint, identity, browser and cloud telemetry to rebuild the chain from the first web visit to script execution. Make sure those detections reach a named person, or your MDR provider if you use one. Windows also records what people type into Run in a registry history called RunMRU, which is why the technique leaves a trace. Below are two vendor-neutral Sigma rules as starting points. I haven't tested them in Falcon, so translate them to your platform's query language and tune them before relying on them:
title: Pasted command recorded in the Run dialog history
status: experimental
logsource:
category: registry_set
product: windows
detection:
selection_key:
TargetObject|contains: '\Explorer\RunMRU'
selection_data:
Details|contains:
- 'powershell'
- 'mshta'
- 'http'
- '-enc'
condition: all of selection_*
level: medium
title: Script host started by Explorer with download or encoded arguments
status: experimental
logsource:
category: process_creation
product: windows
detection:
selection_parent:
ParentImage|endswith: '\explorer.exe'
selection_image:
Image|endswith:
- '\powershell.exe'
- '\mshta.exe'
- '\cmd.exe'
- '\finger.exe'
selection_args:
CommandLine|contains:
- 'http'
- '-enc'
- 'iex'
- 'frombase64string'
condition: all of selection_*
level: medium
Expect some noise from the second rule on admin machines. Both are hunting aids, not finished detections.
Beyond those two queries, these are the telemetry sources worth wiring into detections:
| Telemetry | What to look for |
|---|---|
| Process creation | explorer.exe as parent of powershell.exe, mshta.exe, cmd.exe or finger.exe with URL, encoded or download arguments; renamed system binaries running from %TEMP% |
| Registry | New RunMRU values containing script hosts, URLs or fake verification text; new Run keys written by script interpreters |
| PowerShell logs | Script block logging (event ID 4104) and AMSI detections; Base64 or nested downloaders |
| Network | Script hosts and finger.exe making outbound connections, TCP 79 leaving endpoints, first-seen domains contacted by powershell.exe or mshta.exe |
| Scheduled tasks | New tasks with innocuous names that re-run every few minutes, as with CrashFix |
| macOS | Shell sessions spawned by browsers or installers, unusual curl, base64, gunzip, osascript and dscl use, archive creation followed by an HTTP POST, new LaunchAgents |
Respond
- Contain the device with Falcon network containment, and treat it as an infostealer infection until proven otherwise.
- Reset the user's password and revoke their sessions from a different, clean device. Cookies and saved passwords may already be gone.
- Check the account: MFA methods, new inbox or forwarding rules, OAuth app consents and sign-in history.
- Look for persistence: Run registry keys and scheduled tasks. CrashFix created a task that re-ran its payload every five minutes.
- Rebuild the machine if a remote access trojan landed. Don't try to clean it.
Train people on the exact trick
One sentence does more than a module on phishing: no legitimate website will ever ask you to paste a command into Run, PowerShell or Terminal. Tell people to stop, close the page and report it, and make reporting a single click. People who report within minutes are your fastest detection.
The bottom line
ClickFix isn't new, but it is now mainstream, and it goes round the controls built for files and links. Put prevention at the first step (Run, script hosts, the browser), let behavioural EDR and XDR such as Falcon catch what gets through, protect identities so a stolen session is worth less, and give people one sentence they will remember.
Sources
- CrowdStrike: ClickFix Attacks, How They Work and How CrowdStrike Stops Them, 29 Sep 2026
- CrowdStrike: VAPOR PANDA Uses AI-Driven ARTEX to Scale a Global Espionage Campaign, 29 Oct 2026
- Microsoft Security Community: The CISO Cheat Sheet to the Microsoft Digital Defense Report 2026, 3 Nov 2026
- ESET Threat Report H1 2026, 8 Jul 2026
- Rapid7: Q1 2026 Threat Landscape Report, 21 May 2026
- Microsoft Security Blog: ClickFix campaign uses fake macOS utilities lures to deliver infostealers, 6 May 2026
- Microsoft Security Blog: New ClickFix variant CrashFix deploying Python RAT, 5 Feb 2026
- Microsoft Security Blog: Think before you Click(Fix), 21 Aug 2025
- SentinelOne: Caught in the CAPTCHA, 22 May 2025
- Proofpoint: Around the World in 90 Days, State-Sponsored Actors Try ClickFix, 17 Apr 2025