IBM Launches Bug Bounty Program for Granite AI Models with HackerOne Partnership.

IBM has partnered with HackerOne to launch a bug bounty program for its Granite AI models, offering rewards of up to $100,000 for successful vulnerabilities found. The program aims to improve AI safety and security, and IBM is already using its Granite Guardian tool to block 99.97% of jailbreak attempts. Analysts have a Moderate Buy consensus rating on IBM stock, with an average price target of $294.69 and 20.2% upside potential.

IBM (IBM) has partnered with HackerOne to launch a bug bounty program focused on its Granite AI models, offering rewards of up to $100,000 for successful vulnerabilities found. The program aims to enhance AI safety and security, with IBM already using its Granite Guardian tool to block 99.97% of jailbreak attempts. Analysts have a Moderate Buy consensus rating on IBM stock, with an average price target of $294.69 and 20.2% upside potential.

The Granite AI models, developed by IBM, are used in various enterprise applications. The bug bounty program invites ethical hackers to test the models by attempting to make them behave in unintended ways. Successful findings will be studied by IBM's internal team of experts in AI safety, security, and governance to improve the models. The program will also launch with Granite Guardian, an open-source safety tool designed to work alongside any large language model.

The partnership with HackerOne leverages the company's community of researchers who have proven valuable in testing the safety and security of real-world AI systems. The goal is to probe edge cases, expose novel failure modes, and identify risks before they become apparent. The program will evolve over time, with the potential for bounty rewards to increase based on the scope of activities.

IBM's Granite models are already considered one of the most secure open-source families. The Granite Guardian models hold six of the top 10 spots on the GuardBench, an independent measure of how well guardrail models can detect harmful and hallucinated content. When paired with a Granite LLM, the success rate for jailbreaking the model drops to 0.03% on the HarmBench red-teaming framework.

The first cohort of researchers will be invited by HackerOne to test their skills against Granite in the coming weeks. The program is expected to continue, with the findings from the bug bounty program helping to shape the future of AI security and maintainability.

References:
[1] https://www.tipranks.com/news/ibm-partners-with-hackerone-to-launch-a-bug-bounty-program
[2] https://research.ibm.com/blog/granite-hackerone-bug-bounty
[3] https://www.marketscreener.com/news/ibm-further-strengthens-granite-for-enterprise-deployment-with-hackerone-ce7c50dedf88f127