EverHarden Research

Our vulnerability research in the software small companies run, and earlier work that stays online: three posts from spring 2026 on indirect prompt injection, and the Gate and Ghost demos for checking what AI agents are allowed to do.

Ongoingstarted 2026-11

Looking for flaws in the software our customers run

Small companies in Germany mostly run the same things: WordPress with a handful of plugins, a NAS, a router with a web interface. We spend a fixed part of the week looking for vulnerabilities in that software and report what we find to the vendor or its bug bounty programme.

How we work: we test only installations and devices we run ourselves, or stay inside a programme's written scope. Every report is reproduced by hand before it is sent. We publish details only after the vendor has released a fix.

Accepted findings: none yet, as of November 2026. Each one will be listed here with the vendor's advisory or CVE.

Category manifesto 2026-05-09 ~12 min read

Why single-fetch scanners are structurally blind to AI-agent attacks

Burp Suite, OWASP ZAP, Snyk, and every other web vulnerability scanner share an architectural assumption that breaks the moment AI agents started reading websites on behalf of users. This post explains the gap, walks through three attack classes only multi-agent fetching detects, and frames why “AI-agent web security” is a category that requires different infrastructure.

Read →
Regulator interpretation 2026-05-09 ~10 min read

What the IMF May 2026 cyber-risk warning means for the public web

On May 7, 2026 the IMF named AI-driven cyber risk as a core financial stability issue. The framing focused on AI as attacker tool. It was silent on AI agents as new attack surface. This post connects the IMF systemic-risk argument to the public web that AI agents read on behalf of users, and lists three concrete implications for marketing-site operators ahead of late-2026 supervisory expectations.

Read →
Lab note

In spring 2026 we built a scanner for prompt injection in web content, and the three posts on this page date from then. Until autumn 2026 we reviewed what AI support agents are allowed to do, with Gate and Ghost as the tools. Since November 2026 we offer two checks for small companies in Germany, Watch and Reach (see the home page). They reuse the rule engine and the hash-chained log from Gate and Ghost.

If a piece is missing a citation or you spot an error, write to hallo@everharden.com. We correct mistakes and say so.