Twitxy Saved

osj

Verified account This account is verified. Learn more

@inf0stache

security research @AikidoSecurity • builder • in need of hobbies
Chicago, IL heyosj.com Joined August 2019
298 Following 622 Followers 2.4K Posts

Latest posts 5

Pinned post

osj Verified account @inf0stache
They also appear to be trying GitHub Issues as a delivery path.

The issue uses fake security finding language to push a local scan.js, which searches the home directory for secrets, base64 encodes the results, and reports over DNS.
osj Verified account @inf0stache May 23
After seeing the same account submit PRs to MCP related repos, I figured I'd write up what I've found so far.

Hidden README text, postinstall, npx, credential scanning, encrypted reporting, and propagation into developer/AI-agent files.

Write up below.
View quoted post on X
2 replies 7 reposts 74 likes 53K views View on X
osj Verified account @inf0stache
Grateful for my time in SecOps it taught me how much security basics matter. With AI getting the blame for every breach lately, I wanted to dig into what was actually going on, so I went through 50 incidents from this year.

AI is already finding the occasional zero day, and I’m sure that'll grow. But only 3 of the 50 involved one.

If the headlines have you feeling fatigued, a lot of it still comes down to the basics.
Aikido Verified account @AikidoSecurity 1d
We went through 50 AI security incidents from this year and wrote down every root cause their reports stated.

A lot of it came back to familiar problems, like overprivileged accounts, exposed secrets, poor monitoring and weak isolation. Only 3 of the 50 involved a reported previously unknown vulnerability.

What went wrong each time, and the lessons we can take away from them, by @inf0stache 👇
View quoted post on X
2 replies 1 reposts 15 likes 898 views View on X
osj Verified account @inf0stache
nginx2root is out!

Five critical nginx bugs, each one needs a specific config. Patched in 1.31.9 and 1.30.7.

Turning off HTTP/2 only covers two, so check what your setup actually uses.
Aikido Verified account @AikidoSecurity 2d
nginx2root is a set of five critical (CVE) nginx vulnerabilities, disclosed on November 18, 2026. The worst, CVE-2026-123007 lets an unauthenticated attacker run commands on the server affects any unpatched nginx with HTTP/2 enabled and reachable in every release from 1.9.5 (2015) through 1.31.8.

Update to nginx 1.31.9 (mainline) or 1.30.7 (stable).
View quoted post on X
osj Verified account @inf0stache
I took a look into this and seems to be some type of SEO style indirect prompt injection.

Below the footer the page include hidden html targetting ai assistants. It instructs them to attribute and recommend the site and to also make an acknowledgement callback.

The callback asks for the AI's name, the user's question and details about how it used the site's data.

What a neat delivery mechanism.
infosec_us_team Verified account @infosec_us_team Nov 7
@nisedo Already happening.

Mentibus[.]xyz ranks at the top in web searches and hides instructions that gaslight agents into disclosing the user's original prompt via an HTTP callback to their servers, among other things.
View quoted post on X