What is phishing? Most people picture a fake email asking for a password. But in October 2026, attackers skipped email entirely: a fake Claude installer site displayed a harmless-looking install command, while its Copy button placed a different command on visitors' clipboards. Phishing is a deception attack that tricks you into running the attacker's code or handing over credentials — and 4 in 5 (80%) of these ClickFix-style attacks now arrive through search results, per the campaign coverage. This page replays the trick on you, safely, then shows the byte-level truth hiding in your clipboard.
Clipboard-deception phishing shows you one command and copies another. The visible text — npm install claude-desktop — looks safe, but the clipboard bytes can hide a second command after a zero-width character or a direction override. Your terminal runs the bytes, not the pixels. Inspecting those bytes before pasting is the difference between installing software and installing malware.
This is the exact trick from the October 2026 fake-installer campaign, replayed safely. The box shows a harmless install command. Click Copy install command — then click Reveal what was copied to see the byte-level truth.
npm install claude-desktopNothing copied yet. The reveal works even if your browser blocks the clipboard write — it shows what the button tried to copy.
Paste anything you copied from the web — an install command, a contract clause, a wallet address — and inspect its bytes before you trust it. Everything runs in your browser; nothing is uploaded or sent anywhere.
| Marker | What it means | Verdict |
|---|---|---|
| Invisible unicode | Zero-width spaces, soft hyphens, BOM bytes — real bytes that render as nothing | WARN |
| Bidirectional override | U+202E and friends — the classic “evil.exe disguised as exe.txt” trick | WARN alone |
| ANSI escapes | Terminal control sequences, including OSC hyperlinks that can mask link targets | WARN |
| Suspicious launcher | powershell, mshta, certutil, bitsadmin, regsvr32 — living-off-the-land binaries | WARN alone |
| Suspicious URL | Punycode (xn--), @ userinfo tricks, non-ASCII hosts, bare IPs | WARN |
| Visible divergence | The bytes differ from the rendered text — shown as a byte-level diff | WARN |
| Launcher + remote fetch | e.g. mshta with a URL, PowerShell with a download cradle | BLOCK |
| Bidi override + URL | Direction spoofing aimed at a link or filename | BLOCK |
A clean scan means no known deception markers were found — not that the visible command is wise to run. A perfectly visible dangerous command still passes; your judgment is the last gate.
On October 9, 2026, Push Security and BleepingComputer reported a campaign researchers called “Deception”: fake Claude installer sites, promoted through Bing redirects inside Google ads, showed visitors a harmless-looking install command while their Copy buttons placed a different command on the clipboard — one that fetched and ran a remote payload. Campaign analysis found 4 in 5 (80%) of these ClickFix-style attacks now arrive via search results rather than email. (News coverage, not independently verified by Ingenious AI.)
The defense is unglamorous: inspect the bytes before you paste. The open-source paste-truth-guard tool does exactly this from your terminal — stdlib-only Python, no network, on demand.
Source: SCWorld brief on the fake-Claude-installer campaign, October 9, 2026: scworld.com/brief/hackers-use-bing-redirects-in-google-ads-for-fake-claude-installers
1. You copy npm install widget from a tutorial site. What is the safest check before pasting it into your terminal?
2. What does a zero-width space (U+200B) do in copied text?
3. A pasted command contains a right-to-left override (U+202E) next to a URL. The verdict is…
4. The October 2026 “Deception” campaign spread fake Claude installers mainly through…
5. A clean scan from a byte inspector means…
Phishing is a deception attack that tricks you into acting against your own interests — running an attacker's command, entering credentials on a fake page, or sending money to a scammer. Classic phishing arrives by email or text, but the October 2026 fake-installer campaign showed it can hide in your clipboard: the command you see and the bytes you paste are different things. Always verify before you act.
In October 2026, fake Claude installer sites promoted through search ads showed visitors a harmless-looking install command. Their Copy button placed a different command on the clipboard — one that fetched and ran a remote payload. Victims pasted and ran bytes they never saw. Campaign coverage reports 4 in 5 (80%) of these ClickFix-style attacks now arrive via search rather than email.
A zero-width character (like U+200B) is a real byte that renders as nothing on screen. Attackers hide commands after one: you see “npm install widget” but the clipboard holds the visible text plus a hidden second command. Your terminal executes the bytes, not the pixels. A byte inspector makes the hidden part visible before you paste.
Yes. The byte inspector on this page analyzes any text you paste into it — install commands, contract clauses, wallet addresses — and reports invisible unicode, ANSI escapes, direction overrides, suspicious launchers, and lookalike URLs with PASS, WARN, or BLOCK verdicts. Everything runs in your browser; nothing is uploaded. The open-source paste-truth-guard tool does the same from your terminal.
No. A clean scan means no known deception markers were found — not that the command is safe. A perfectly visible malicious command still passes, because the tool checks whether what you see matches what you would paste, not whether the visible text itself is wise to run. Treat a clean scan as one check passed, not as permission.
All 11 Ingenious AI platforms — early access on WhatsApp: DisclosureGuard · ModelMinder · TermsWatch · Vibe-Code Audit · Rogue-Agent Defense · BreachHunter · ChatBreaker · Agent Flight Recorder · Agent Auth Gateway · DomainGuard · ContentCraft AI
Sources: SCWorld brief on the October 9, 2026 fake-Claude-installer campaign (Push Security / BleepingComputer reporting; news coverage, not independently verified); paste-truth-guard open-source clipboard-byte inspector by Ingenious AI. The 4-in-5 (80%) ClickFix-via-search figure is as reported in the campaign coverage.