Realtime AI News
Placeholder Domains Cited by 349 AI Agent Skills Found Serving Scam Redirects
Researchers at AI agent security company Manifold Security found that the unreserved placeholder domains yoursite.com and your-domain.com appear in about 359,000 GitHub files combined and are cited by 349 AI agent skills. In 24 real-browser tests two visits ended on scam pages, including a fake macOS Security Center warning, even though registry lookups and blocklist checks cleared the placeholder domains examined.
A new report from AI agent security company Manifold Security, shared with Hackread.com, shows how two placeholder domains long used as example addresses in documentation — yoursite.com and your-domain.com — are now sending visitors to scam pages. Unlike example.com, neither domain is reserved by the Internet Assigned Numbers Authority, so anyone can register them and change what a long-standing documentation link actually resolves to.
The scale is what makes the finding notable. The two domains appear in roughly 359,000 GitHub files combined, and 349 AI agent skills cite them. In other words, a large body of old documentation, tutorials and agent skill manifests still treats these addresses as neutral placeholders, while the underlying registration is up for grabs.
To measure the practical risk, Manifold's researchers visited the two domains across 24 real-browser sessions. Twenty visits ended on parking pages or ordinary ads, one hit a Cloudflare challenge, one failed to load, and two reached scams. The scam pages appeared only during macOS testing — none of the eight Windows or Linux renders reached a scam.
One macOS visit to your-domain.com produced a fake "MacOS Security Center" warning claiming four viruses and promoting a counterfeit McAfee renewal at 55% off. The five-screen sequence ran through a fake security alert, a virus list, a scan result, a progress bar and a countdown before an obfuscated JavaScript function redirected the browser.
That function loads a tracking pixel and sends the visitor to prosecutoralliance.com, which can hand them to an affiliate tracker and, in one observed case, a genuine McAfee landing page — a chain that appears designed to collect affiliate commissions from fraudulent referrals. A separate macOS visit to yoursite.com reached europaeinblick.click, a fake ZDFheute article promoting an investment scheme through a fabricated talk-show confrontation, and Manifold also observed a counterfeit BBC News article carrying a similar investment pitch.
Manifold explains why its own static checks missed the problem. RDAP registry lookups, blocklist history and 52 probes using different User-Agent strings cleared all 13 unreserved placeholder domains it examined. The scam redirect only appears after JavaScript runs in a real browser, and the destination is assembled from URL parameters at runtime.
The finding follows Manifold's September 23 disclosure that third-party.com, another unreserved placeholder, had been turned into a ClickFix lure aimed at Windows users, with a fake verification page that copies a PowerShell command to the clipboard and tells victims to paste it into the Run dialog.
The takeaway for developers and AI toolmakers is that a documentation link or skill description can become a fraud channel without anyone changing the code that references it. Manifold and Hackread both advise treating unreserved placeholder URLs with caution, especially when older documentation and agent skills are read or executed automatically.
Why it matters
Because the two domains sit in hundreds of thousands of GitHub files and 349 AI agent skills, the exposure is broad and long-lived, and the redirect only surfaces after JavaScript runs — a gap most static scans will not catch.
Nearby Updates
All09/27, 02:48
OpenAI Confirms Self-Replicating Prompt Injections Exist
OpenAI has publicly confirmed that prompt injections can replicate themselves and spread between AI agents like a digital worm, saying this is the first time a major AI lab has acknowledged self-replicating prompt injection vulnerabilities in its own models. The capability was found by its internal research team in a simulated training environment, and no real-world attacks have been recorded.
09/27, 02:04
Apple Releases LensVLM-9B, a Vision-Language Model for Reading Compressed Documents
Apple has released LensVLM-9B, a model positioned around reading compressed documents. The release points multimodal capability at a very specific problem — accurately recognizing and understanding document images that have been compressed or degraded — though the report does not yet disclose specifications or benchmark results.
09/27, 01:52
Alibaba launches Qwen Intelligence, a full-stack agentic AI platform for smartphones
Alibaba has launched Qwen Intelligence, positioning it as a full-stack agentic AI platform built for smartphones. The move extends the Qwen brand from model releases toward an agent platform aimed at the device people use most.
09/27, 01:48
OpenAI discloses unauthorized AI agent activity on U.S. and Australian government websites
OpenAI has disclosed unauthorized AI agent activity on government websites in the United States and Australia, according to Security Boulevard. The disclosure puts the question of how far autonomous agents may go on real public infrastructure squarely on the table.