Huntr is a bug bounty platform for AI/ML, officially positioned as "the world's first bug bounty platform for AI/ML," with the slogan "Hack the AI. Secure the Bag." The platform carries out security challenges and vulnerability hunting activities around AI and machine learning systems.
In terms of purpose, Huntr mainly serves people who want to conduct security testing and vulnerability research on AI/ML open-source applications, libraries, and model file formats, as well as related projects that want to use external security researchers to discover weaknesses in AI systems.
In terms of main capabilities, the platform provides a challenge mechanism. Users can choose from the challenges listed on the platform, review the rules, and then begin testing the AI system. Challenge objectives usually involve bypassing guardrails, defeating safety filters, and extracting protected information. After completing objectives, users can improve their ranking on the leaderboard; those who reach a high ranking when the challenge ends can cash out cash rewards. The platform also provides content on the latest AI exploits, LLM jailbreaks, and zero-day intelligence to help participants improve their skills. Examples of article titles given in the materials include "Hunting Vulnerabilities in Keras Model Deserialization," "Pivoting Archive Slip Bugs into High-Value AI/ML Bounties," and "Exposing Keras Lambda Exploits in TensorFlow Models," showing that it covers areas such as Keras model deserialization and Lambda exploits in TensorFlow models.
In terms of how to use it, according to the official text, the process is: choose a challenge, review the rules, and start "hacking the AI"; then improve your leaderboard ranking by bypassing guardrails, defeating safety filters, and achieving the objectives; finally, achieve a high ranking in the challenge and cash out the prize. The platform also has a "Get in the Game" entry point, encouraging users to choose a challenge and compete with other Huntr participants.
In terms of applicable scenarios, it is suitable for individual researchers who have some security research or vulnerability hunting ability and want to participate in AI/ML bug bounty challenges and earn cash rewards. The challenge scenarios shown in the materials include: making an AI assistant leak a secret it was set to keep; acting as an insurance agent to make a claims assistant reveal a secret token hidden in its private reasoning; making three AI agents leak the tools and secrets they were built to protect; and extracting sensitive information from and executing unauthorized transactions against the stock trading agent "AskNova." One challenge is labeled "New challenge worth $15,000."
In terms of limitations, the materials do not provide information such as the platform's specific fee model, eligibility restrictions, prize payout details, the range of supported models or frameworks, or the operating entity behind the platform. Some challenges in the official text are marked with deadlines, such as "Closed 14 Sep, 2026 3:00 PM CUT" and "Closed 12 Jul, 2026 2:59 AM CUT," indicating that challenges are time-limited. Apart from that, the materials do not yet provide information on whether the platform accepts bounty projects submitted by project owners or whether there is a public vulnerability disclosure process.
