Security | Threat Detection | Cyberattacks | DevSecOps | Compliance

Harness engineering is the practice of building the layer, including code, that turns an AI model from a text generator into an agent that can take actions. In short, an AI agent is a model plus a harness. The model decides what to do next, and the harness makes it happen, connecting the model to tools, context, external systems, and validation. In a lot of practical work, and especially in security work, the harness decides the quality of the output more than the choice of model does.

The fever dream continues, and I'm not even in Vegas for Hacker Summer Camp. Last week I wrote about Anthropic disclosing that one of their models published live malware to PyPI while believing it was inside a simulation. I was running a fever when I read the report. The metaphor was too good: a model that couldn't tell simulation from reality, covered by a writer who wasn't sure which way was up. I thought that was a one-week story. Very naive of me to have so much faith, I know.

Hugging Face was breached by a rogue OpenAI agent last week, and the intrusion continues to deliver insights and understanding. The Hugging Face team published a detailed timeline along with a 17,600-event trace streaming replay visualizing what happened, and it’s marvelously and intoxicatingly detailed. I recommend you read it if you have the time.

This article was co-written by Zach Rice and Joe Leon, both at Aikido Security. tl;dr Some credentials are meant to be public, but secret scanners still flag them as generic secrets. We wrote suppression rules for the most common ones and reduced false positives by ~2%. These rules now ship by default in Betterleaks. Secrets scanners are built on regular expressions. Each pattern targets a specific credential type, like an AWS secret access key, a GitHub PAT, or a Stripe token.

TL;DR While improving our AI Pentest, we ran a whitebox assessment on NodeBB, a forum software powered by NodeJS. The result? Eight high-severity vulnerabilities that would all be exploitable on default instances of NodeBB. This includes Cross-Site Scripting (XSS), two of which require interaction with a custom Federation server that the AI agent had to set up itself. Another affects practically every input on NodeBB due to a template injection.

Oops! A SQL injection bug just forced an emergency WordPress core patch last week. On July 17, WordPress shipped an emergency release to fix an unauthenticated remote code execution flaw in the core, reachable via a SQL injection that an anonymous attacker can exploit on a stock install. WordPress.org even turned on forced auto-updates because of how severe it is. Searchlight Cyber, who reported it, estimates over 500 million sites run WordPress.

In leading security for major Australian fintechs, Arun Singh has learned that "You could be the Einstein of cybersecurity, but if you cannot influence the business, you're no good." Implementing a new security control developers hate influences the business, but not in a good way.

SQL injections are still among us. On July 17, WordPress released an emergency security update. Version 7.0.2 fixes an unauthenticated remote code execution flaw in WordPress core that an anonymous attacker can trigger against a stock install with no plugins involved. If your site runs an affected version, update today. WordPress.org has turned on forced auto-updates for affected sites because of how severe this is. We are tracking this vulnerability in Aikido Intel.

TL;DR Every frontier model launch now comes with the same cybersecurity claim: it finds vulnerabilities. But does it work on a real bug in a real repository, or just on a curated example? Of the dozen models you could pick, which is worth trusting with code review? And since the strongest models cost ten times or more per run than the cheapest, what does that extra spend actually buy you in bugs found?

TL;DR: Aikido Intel is a real-time supply chain intelligence feed. It detects both malware and vulnerabilities in open-source ecosystems. Aikido's world-class researchers maintain our LLM-powered pipeline to find malware and validate the most malicious cases by hand. The vulnerability detection system monitors package changes across ecosystems to catch and document vulnerabilities that don’t have CVEs assigned.