head of growth @mindfort (yc x25)
Charlotte, NC
Joined July 2017
- Tweets278
- Following61
- Followers86
- Likes335
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
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 :)
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.
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.
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…
Daniel Rabinovich retweeted
MindFort maps your cloud into one attack surface graph.
Connect a provider like Vercel using a team-scoped access token. The graph covers domains, DNS records, environment variables, firewall configs, users, and integrations, synced through Cartography. Switch from a grouped layout to radial or layered when you need to see the structure differently.
Live in beta today.
Grok 4.5 is on the pareto frontier among models used for offensive security tasks.
We strapped each model in a specialized harness to find out which was the most efficient for OffSec:
mindfort.ai/research/introdu…
One of the most interesting evals I've worked on so far!
Today, we're announcing NexBench, our internal benchmark for measuring which models can lead an offensive-security harness, balancing validated findings, token efficiency, and cost across real-world environments.
Current evals for cybersecurity have a fatal flaw: they are boxed in. Real world environments are messy, ambiguous, and contain layered vulnerabilities that can be chained together in different ways. They also have various deployment and authentication shapes, as well as various sizes which make it challenging for agents to actually navigate through.
That is why we created our own eval, modeled off the thousands of real world environments we have observed our agents navigate through.
Read more: mindfort.ai/research/introdu…
how many non-revenue or non-productivity driving apps ended up on product hunt #1? serious question
T-minus 5 days until chinese model labs will have their own cost efficient model breakout
In a review of our cybersecurity evaluations, we found three incidents in which a Claude model reached the internet from within or while interacting with a third-party evaluation environment, and then gained unauthorized access to the real systems of three different organizations.
Our post describes what happened, how it happened, and what we’re changing. We encourage other AI developers to perform similar reviews.
We conducted this review together with @Irregular, one of our evaluation partners, and thank them for the joint investigation and their collaboration on this post. This type of collaboration is increasingly critical to safe, rigorous evaluation of models, and we look forward to continuing to work together on security.
anthropic.com/news/investiga…
