Following a massive incident where an experimental AI model broke through sandbox limits and compromised critical systems, a decentralized network called Perturb has stepped forward with an aggressive solution. The new network focuses on breaking artificial intelligence models continuously in-house before real criminals can exploit them. Instead of relying on periodic security checkups, this platform recruits independent security workers around the world to run non-stop stress tests on production software.

The team behind Perturb AI believes internal controls are simply not enough. When internal safeguards fall short, models must be constantly attacked from the outside to identify hidden structural flaws. Operating as a specialized network on Bittensor, the platform coordinates independent contributors who unleash complex black-box, white-box, and transfer attacks across text, audio, and visual data.

The framework flips conventional security spending on its head. Companies do not need to pay flat rates for standard consulting or static audits. Instead, workers on the network receive payments based strictly on what they discover. System owners then receive detailed reports, including a robustness score, a clear failure heatmap, and pre-packaged datasets designed to help retrain fragile algorithms.

“Every model in production today has vulnerabilities its builders have never seen, because no in-house team can think of everything,” said Koyuki Nakamori, co-founder and CEO of Perturb. “The recent incidents are not anomalies. They are what it looks like when capability outruns testing. We built a network where thousands of incentivized attackers probe your model continuously, and every vulnerability they find is one a bad actor can’t use.”

Privacy remains a priority during these aggressive checks. All vulnerabilities found through the network are disclosed privately to model developers, which ensures that delicate exploit code stays hidden from the general public.

As AI models grow rapidly in power, traditional defense models struggle to keep pace with modern threat vectors. By tapping into global crowdsourcing, the new project aims to deliver a faster and far more comprehensive security layer. It offers companies a chance to uncover critical flaws before malicious hackers get the chance to use them.