SWE (Cybersecurity)

📋 Description

• • Join a groundbreaking initiative at the forefront of AI and cybersecurity by contributing your specialized software engineering skills to high-impact research collaborations with leading AI organizations.
• • This unique opportunity allows you to play a pivotal role in constructing sophisticated training datasets that are essential for enhancing the reasoning capabilities of AI models, specifically in tackling complex software engineering and security challenges.
• • You will leverage your deep expertise in secure software development practices and your keen ability for vulnerability analysis to help shape the future of intelligent systems and their interaction with code.
• • The project centers around SWE-bench-style tasks, meticulously derived from real-world, open-source software repositories, providing a practical and challenging environment to explore how advanced AI models approach intricate software engineering and security problems.
• • A significant aspect of your role will involve reconstructing and annotating the reasoning steps within AI model-generated trajectories, especially for models that do not inherently expose their internal thought processes.
• • By applying your own engineering acumen and understanding of the task context, you will infer, document, and validate the likely decision-making processes that underpin each step taken by the AI.
• • Key responsibilities include the design of benchmark tasks, which involves identifying specific vulnerability classes, categorizing them by type, subtype, and difficulty level, and rigorously validating the intended exploit behavior.
• • You will be instrumental in creating or validating small, self-contained, runnable codebases, often referred to as environment repositories, which may incorporate data ingestion mechanisms and prompt/tool usage scenarios where trust boundaries are intentionally challenged.
• • A critical component of the project is the validation of attacks through the development of precise exploit scripts, accompanied by clear and comprehensive documentation of the resulting unsafe behavior.
• • Furthermore, you will implement or meticulously verify patches designed to prevent identified exploits, ensuring that these fixes are not only effective in mitigating the vulnerability but also robust against potential bypasses.
• • Your work will culminate in the production of detailed task metadata, encompassing crucial elements such as severity mapping, precise file and line number references, thorough impact analysis, concise remediation summaries, and relevant supporting documentation.
• • You will also conduct rigorous review and quality assurance checks to guarantee that file paths are resolved accurately, line ranges are correct, sensitive information is not inadvertently leaked, and that the implemented patch successfully blocks the intended exploit.
• • This role offers a flexible, part-time commitment, allowing you to contribute between 15 to 25 hours per week, with the potential to scale up to 40 hours per week based on project needs and your availability.
• • The engagement is structured as an independent contractor role, offering the freedom to work remotely from anywhere in the United States and manage your own schedule.
• • Payments are processed weekly via Stripe or Wise, ensuring timely compensation for your valuable contributions.
• • This is an exceptional chance to apply your software engineering and cybersecurity expertise to a project that directly influences the development of more capable, secure, and reliable AI systems, making a tangible impact on the future of technology.