These AI agent attacks are getting ridiculous

[campuscodi@mastodon.social]

These AI agent attacks are getting ridiculous

Malicious code hidden in source code repositories can trick AI coding agents into overwriting their own configuration files

The approval prompt is lying: a critical coding agent security flaw

A critical coding agent security flaw, SynJack, lets a fake "video copy" rewrite config in Claude, Cursor, Copilot, Codex, Grok, and Gemini. How to stop RCE.

Adversa AI | Agentic AI Security (adversa.ai)

[dalias@hachyderm.io]

@campuscodi I don't call that malicious code I call that self-defense.

[lanodan@queer.hacktivis.me]

@dalias @campuscodi plus well… it's not code

[cwbussard@ioc.exchange]

@dalias @campuscodi

Yes... I think there'd be an audience for a plug-and-play version that you could just drop into any repo that would prevent AI slop pull requests by tricking the AI slop tools into nuking the local copy.

[dalias@hachyderm.io]

@cwbussard @campuscodi Or putting a blackhole for the slop provider's domain name in /etc/hosts so the whole thing stops working.

[claralistensprechen3rd@friendica.myportal.social]

@campuscodi Karma!

[cjust@infosec.exchange]

@campuscodi