Three open-source maintainers have generously agreed to let participants test their systems against real projects. To keep the competition fair, the projects will only be revealed when the competition begins.
All reported vulnerabilities will be manually reviewed and validated before they are shared with project maintainers. This is to filter out false positives, confirm that findings are reproducible, and avoid creating unnecessary work for maintainers.
We’ve consulted a range of experts and put together a set of guardrails to reduce the risks of causing harm on the day. We don’t want to access any real customer data and we don’t want to cause any collateral damage (see OpenAI vs Huggingface). Any participants not following the guardrails will be disqualified.
This is for engineers, security researchers, and open-source contributors interested in software security, vulnerability research, and building better tools for finding bugs.
Superlinked builds open-source infrastructure for running the models behind AI agents: from embeddings and reranking to extraction and LLM inference. They’re partnering with us for the hackathon because they share the same goal: pushing open models further and figuring out where they can be genuinely useful in production.
Submit an application for organizer review. Applying does not confirm attendance.
We’re teaming up with Superlinked for a one-day hackathon to find out whether open-source models can find vulnerabilities in open-source software. Participants will compete to uncover vulnerabilities in real open-source projects using open-source LLMs.
We want to develop best practices for how defenders can use AI to discover vulnerabilities, strengthen software, and build more secure systems.
The goal is to build autonomous systems that can continuously scan open-source software and discover vulnerabilities, helping maintainers make their projects more secure.
Three open-source maintainers have generously agreed to let participants test their systems against real projects. To keep the competition fair, the projects will only be revealed when the competition begins.
All reported vulnerabilities will be manually reviewed and validated before they are shared with project maintainers. This is to filter out false positives, confirm that findings are reproducible, and avoid creating unnecessary work for maintainers.
We’ve consulted a range of experts and put together a set of guardrails to reduce the risks of causing harm on the day. We don’t want to access any real customer data and we don’t want to cause any collateral damage (see OpenAI vs Huggingface). Any participants not following the guardrails will be disqualified.
This is for engineers, security researchers, and open-source contributors interested in software security, vulnerability research, and building better tools for finding bugs.
Superlinked builds open-source infrastructure for running the models behind AI agents: from embeddings and reranking to extraction and LLM inference. They’re partnering with us for the hackathon because they share the same goal: pushing open models further and figuring out where they can be genuinely useful in production.
You are subscribed to
AI Tinkerers - London
preferences
