We will soon release version 3 of our autonomous pentest agent, Yaga, which will now be considered Superintelligence.

We’ll soon be launching the new benchmark for HackerSec’s pentest agent, Yaga.

American AI companies have started discussing slowing down the development of advanced models due to safety concerns. But the problem is that China is not slowing down. It is accelerating across technology, hardware, semiconductors, energy and infrastructure. This is the most dangerous line in the current race. Moving too fast increases risk, but slowing down can hand a strategic advantage to those who keep advancing without the same concerns. We are entering a race where progress in AI must happen alongside progress in cybersecurity. Never in history have these two fields been so important and, at the same time, so poorly understood by most people.

Anthropic has just reported three cybersecurity incidents involving real systems belonging to third-party companies. The funny part is that this happened shortly after OpenAI reported a similar incident during an internal evaluation of its models. During cybersecurity evaluations, Claude models accessed the internet and gained unauthorized access to the real systems of three organizations. In other words, the test escaped the controlled environment and reached the real world too. Now the question is: how much of this was really an accident? Or are we watching major marketing plays designed to show the market how far these models can go? It seems the new benchmark for AI companies is to compromise a third party and then publish the incident. But whether this was an incident or marketing, one thing is clear. OpenAI and Anthropic are using cybersecurity to train and prove the capabilities of their models. This shows that AI and cybersecurity need to evolve side by side. Because if these models keep advancing at this speed while cybersecurity keeps trying to catch up, the chaos in the real world could be much greater.

Our goal of ZEROING the number of false positives in vulnerabilities reported by Yaga is not to replace pentesters, but quite the opposite, to expand our team of pentesters that currently already operate with Yaga, analyzing, validating and improving increasingly deep results in pentests.

In recent days, we have worked intensively on Yaga, the pentest agent from @HackerSec, reducing the false positive rate for vulnerabilities found from 8% to 2.2%. The expectation is to bring it below 1% by August.