Opinion: AI has rewritten the economics of red teaming
Original title: We are not ready for what AI has done to red teaming
The author says AI-assisted teams logged 100-plus true positives on one app, versus 7 to 10 findings in a traditional engagement.
A CSO Online opinion column argues that the urgent change is red teaming with AI, not only testing models for prompt injection. It says frontier models let small offensive teams outperform larger traditional operations from two years ago, across setup, attack and reporting. Cited figures, aggregated by DayBlink Consulting and described as representative of Robert Pimentel’s experience, include 90-plus findings in one after-hours exercise versus 60 to 70 a year for a 3-4 person team, and scans of millions of IPs in 2-3 hours for roughly $20k-$30k. The column says reporting still needs human review and that judgment, not raw AI output, separates strong operators from “slop cannons.”
Unconfirmed: The productivity numbers are the author’s and DayBlink Consulting’s, not independently verified.
- The column separates AI red teaming of models from using AI in traditional offense.
- It claims external-app reconnaissance rose from 150-200 findings a year to 100-plus a month.
- Nine to ten months of C2 logs can draft a deconfliction report in under 20 minutes.
- Operators still must verify findings and ATT&CK mappings before delivery.
- An LLMs-as-judges workflow is described as checking actions before they run.
- Organizations
- DayBlink Consulting
From the original article
6 min read at csoonline.com
Something shifted in offensive security over the last eighteen months that most organizations haven’t fully absorbed yet. It isn’t that attackers got smarter. It’s that the economics of attack changed entirely — and those of us who run red teams have had a front-row seat to the transformation.
csoonline.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.