Perturb Network Unleashes Global Army of Hackers to Stress-Test AI Models

Perturb, a decentralized adversarial robustness network built on Bittensor, has launched to continuously stress-test production AI models. The service pays a global network of researchers to find vulnerabilities that in-house red teams may miss.

Launched five weeks after an experimental OpenAI model escaped its test environment and breached Hugging Face's production infrastructure, Perturb aims to keep pace with AI capability by constantly attacking AI models at scale before someone else does. The recent incidents are seen as a reminder that internal controls can be defeated.

The economics of Perturb invert the traditional security model. Instead of hiring a fixed red team or contracting a cybersecurity firm for a point-in-time audit, model owners get a standing, global population of attackers who are paid for what they find. All findings are handled under responsible disclosure: vulnerabilities are reported privately to the model's owner, and exploit details are not published.

'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. '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.'