186 Fake AI Domains Power TypoPrompting Malware Push

Introduction
Cybersecurity researchers have identified 186 lookalike domains that impersonate artificial intelligence tools and use that borrowed credibility to push malicious content at users. The domains serve ClickFix-style lures, browser-hijacking malware, and a set of extensions that take over a victim's search results. Island, which tracked the activity, has labeled this delivery pattern TypoPrompting.
The name is descriptive. A user who mistypes the address of a familiar AI tool, or clicks the wrong search result while looking for one, lands on a page built to look legitimate. Instead of the assistant they wanted, they get a prompt to verify, install, or approve something. The public reporting does not yet identify the specific domains involved, the malware families behind the hijackers, or when the campaign began.
What TypoPrompting Actually Is
TypoPrompting is a delivery pattern rather than a single piece of malware. The infrastructure is a pool of domains crafted to resemble the names of popular AI applications, and the payload behind each one varies. Some pages walk the visitor through a ClickFix sequence. Others push software that alters browser behavior. Others steer users toward extensions that quietly rewrite where their searches go.
That variety matters. A single actor running one malware family is easier to fingerprint and block. A shared delivery pattern spread across 186 domains gives an operation room to rotate payloads, retire burned infrastructure, and test which lure converts best, all while keeping the same basic trick intact.
The Three Payloads Behind the Domains
The reporting groups the malicious output into three categories, each with a different objective.
- ClickFix-style lures. ClickFix pages present a problem the visitor supposedly needs to solve, then instruct them to paste a command into a Windows dialog or terminal. Because the user runs the command themselves, the activity looks like a deliberate admin action rather than an exploit, which complicates detection and attribution.
- Browser-hijacking malware. This category covers software that seizes control of browser settings or behavior after installation. Once a browser is compromised, it becomes a pivot point for session data, credentials, and everything the user does online.
- Search-hijacking extensions. Extensions that redirect queries send traffic and data through infrastructure the attacker controls. The user sees a working search box; behind it, their queries, and potentially more, flow somewhere else.
Search hijacking deserves particular attention because extensions run with the authority of the browser profile. In an enterprise, that profile often holds authenticated sessions to internal tools.
Why AI Brand Impersonation Works
AI tools have become part of the daily workflow for many technical teams, and adoption has frequently outpaced the governance around it. Users try new assistants on personal devices, install browser add-ons to extend them, and treat the category as low-risk because it feels like productivity software rather than infrastructure.
Attackers read that behavior correctly. A person hunting for an AI tool is motivated, in a hurry, and unlikely to scrutinize a URL that looks close enough. TypoPrompting exploits the gap between how quickly a tool category grows and how slowly users learn to verify where they are going. The same pattern has long targeted banks, cloud consoles, and software vendors; pointing it at AI brands simply aims it at the traffic that is currently growing fastest.
What Defenders Should Do Now
Because the campaign leans on user interaction rather than a software vulnerability, controls that assume a malicious file or exploit will miss it. Practical steps that address the actual mechanics:
- Publish and enforce an approved AI tool list. Give users a sanctioned path to the assistants they need, distributed through bookmarks or an internal portal, so a mistyped address is not the default route.
- Filter lookalike domains at DNS and the web proxy. Block newly registered and newly observed domains, and treat close-match spellings of popular AI brand names as high risk. Feed the 186 reported domains into your blocklists where they are available.
- Lock down extension installation. Use browser policy to restrict installs to an approved store allowlist, and audit installed extensions across managed profiles for ones that alter search or homepage settings.
- Restrict clipboard-to-shell paths. Where policy allows, limit terminal and Run dialog access for standard users, and alert on shell processes spawned shortly after browser activity.
- Watch for the symptoms, not just the lure. Unexpected search provider changes, new homepage settings, and unfamiliar extensions are the fingerprints left behind after a successful hit.
- Train on the specific reflex. Most users know not to open unexpected attachments. Fewer have been told that a page asking them to paste a command is the attack.
Grivyonx Expert Analysis
The most useful detail in this reporting is not the number 186. It is that the payload changes while the pattern stays the same. Defensive programs built around blocking known-bad files will keep losing to infrastructure that swaps what it serves while keeping the same social hook. The durable control is at the boundary: control which domains a managed browser can reach, control which extensions it can load, and make the approved path to a new tool easier than the improvised one. Shadow AI adoption is a security problem long before it is a data governance problem, and campaigns like this are how that gap gets monetized.
Conclusion
TypoPrompting is a reminder that the fastest-growing category of workplace software is also the newest target for impersonation. The 186 domains researchers catalogued are infrastructure, and infrastructure gets replaced; the underlying trick, which is turning a routine attempt to reach a familiar tool into an instruction the user follows, will outlast them. Organizations that pair explicit AI tool policies with tight control over browser extensions and outbound domain access remove most of the surface this pattern depends on. Grivyonx helps teams close that gap through AI automation and DevSecOps practices that put security controls directly into the tools and workflows employees already use.
- #TypoPrompting
- #ClickFix
- #fake AI domains
- #browser hijacking
- #search hijacking
- #social engineering
- #endpoint security



