
Building the final frontier of security swarms @mindfort
Charlotte, NC
Joined July 2017
- Tweets305
- Following62
- Followers164
- Likes350
To anyone concerned about ARTEX affecting their org:
We built a closed source agent swarm platform you can try for free, with guardrails, scoping, that you can deploy on any API, MCP, Portal, etc.
Hundreds of coordinated agents, acting like ARTEX, without malicious side effects.
Begging any MOTS, Security Engineers, AppSec, to really give it a shot.
You'll be extremely surprised at what you'll find, once hundreds of harnessed agents are given time, resources, and crazy compute power.
Building in house in such a short time frame is really difficult (and hard to keep agents governed), in the age of machine-speed attackers, who are running abliterated models 24/7.
@mindfort has your back.
3
587
Now with ARTEX existing, code security by itself is no longer an option. Neither is DAST. If attackers are using agent swarms against you everyday, you must be doing the same. Evaluate vendors accordingly.
2
324
Shocker: start testing everyday in RUNTIME.
People here keep saying, “We need to rewrite all the human code with AI so it’s secure,” but seem to be leaving out a few steps. A bunch of AI pilled, AI everything startups have been getting popped through web/API bugs this year. There’s clearly another step between “rewrite all the code with AI” and poof, everything is secure.
3
285
Spent a good bit of time with ARTEX / Strix this week on some targets I built to simulate enterprise apps, to really test their capabilities.
TLDR: Realized how much better agent swarm platforms were than 5-10 agents deployed by ARTEX / Strix, we are really only scratching the surface of how bad these attacks are going to get, once attackers gain access to more compute, and can coordinate mass amounts of agents.
Two interesting findings about ARTEX + Strix:
2. When running these in a code harness (which most users will do), I found that the harness would oftentimes rewrite the scopes of the test for ARTEX/Strix, which affected results in some cases.
If you are using any code agent while using these tools, you should really be monitoring every trace that is happening, things can really start to get fuzzy if the harness starts going outside of it's scope that it was initially designed to execute on.
18
5,037
1. Both ARTEX / Strix struggled with finding escalation vulnerabilities and business logic vulns, and got noticeably slowed down by the heavy rate limiting I introduced.
1
15
5,808
The past few days, I've been testing a multitude of open source pentesting harnesses on a target I built, and the result difference compared to security agent swarm platforms is absurd.
Crazy how much better/faster/stronger 100+ agents are than 10 agents split up.
It is most definitely a compute endgame. Teams that aren't building or using infrastructure to deploy hundreds of these agents weekly are going to soon be cooked.
Will share some interesting findings soon :)
1
7
319
It's a battle of who can use tokens in the most efficient way possible.
There’s a huge component to speed and efficiency (and designing amazing harnesses). Wasted time and excess spend with an inefficient harness and using huge models for executing agent actions means fewer exploitable vulnerabilities found, and slow speed for defenders.
The more efficient the token spend, the more frequently you can run these agents, and a powerful harness can also achieve much better results over time.
The future of cybersecurity is fleets of autonomous and semi-autonomous agents defending your company's estate 24/7. Yes, lots of buzzwords, but I truly believe it.
Your defense will be limited only by the token and compute budget you're willing to spend.
For now, though, most companies don't have access to this tech, and they'll need help deploying it. Ideally using a model-agnostic harness/platform.
1
3
131
The final frontier of SEO is research. Technical SEO, really good keyword data, is no longer a moat and has been democratized ten-fold with agents. I'm worried that as crawlers reward original insights, companies will push to release slop research, just to boost their citations.
1
5
142
The team at @morphllm has been incredible to work with. Really excited to be working with them, and giving more defenders access to frontier cyber.
We're excited to announce a new partnership with the team at @morphllm to bring the latest abliterated open models to MindFort Swarms.
In response to the rapid increase of agent-led cyberattacks, we needed a partner that was able to meet the inference requirements needed to run large scale agent swarms.
MindFort is designed to let teams safely deploy security swarms to quickly find vulnerabilities in their products and services.
Try today: mindfort.ai/morph
2
86
With OpenAI and Anthropic severely limiting cyber programs for GA, I’m proud to lead one of the only companies to truly democratize security agents for all.
You can literally run the same continuous swarms that Fortune 500s already do, and your first run is on us :)
1
3
191
AI pentesting companies have gaslit consumers into justifying 4-5k per pentest. Genuinely puts us back to square one, since it's not sustainable to run continuously, and doesn't accomplish what we consumers sought out to do, trying to run swarms weekly or even daily or multiple targets to harden their surfaces.
Very proud of to be building security swarms that benchmark at ~1/30th of the cost, while beating every ai pentesting company you can think of.
1
1
212
Echoing again that defenders need to be continuously testing their infrastructure and applications with same tools that attackers use. Cyber models are getting cheaper, stronger, and faster.
I remember when we tested GPT 5.6 for cyber - it was powerful but not cheap whatsoever, not practical for long-running use.
A few months later, it's a different story.
We tested DeepSeek V4.1 Flash 1017 on NexBench, and it placed 1st in overall performance and cost to performance ratio, beating our previous #1 (Grok 4.6), with 30% less dollar spend.
NexBench tests a model's ability to recon, exploit, and chain vulnerabilities in hyper-realistic environments, while measuring price-to-performance ratio, token efficiency, and more.
Our full write-up is coming soon, but one this is clear: open models are becoming the most powerful tools defenders have, besting many of the models available from frontier labs for offensive work.
3
129
Daniel Rabinovich retweeted
We tested DeepSeek V4.1 Flash 1017 on NexBench, and it placed 1st in overall performance and cost to performance ratio, beating our previous #1 (Grok 4.6), with 30% less dollar spend.
NexBench tests a model's ability to recon, exploit, and chain vulnerabilities in hyper-realistic environments, while measuring price-to-performance ratio, token efficiency, and more.
Our full write-up is coming soon, but one this is clear: open models are becoming the most powerful tools defenders have, besting many of the models available from frontier labs for offensive work.
Daniel Rabinovich retweeted
Last week, we evaled Grok 4.6 and 4.7 in our agent harness to test the models' ability to reason about, write, and execute real exploits against a target in a messy environment.
Grok 4.6 scored incredibly well, becoming the best model we've ever tested at recon, pentesting, and reporting validated findings.
Even more impressively, it did so at a $0.88 cost per finding, making it our 2nd most cost-efficient model for cyber use.
See our updated findings here: mindfort.ai/research/nexbenc…







