A new fellowship and hackathon initiative by Apart Research and Atlas Computing aims to tackle the growing security risks of AI-generated code using formal methods and secure program synthesis.
\nIn a recent post, lessw-blog highlights an urgent call for mentors for the upcoming Secure Program Synthesis Fellowship, a joint initiative spearheaded by Apart Research and Atlas Computing. The announcement outlines a structured effort to address one of the most pressing vulnerabilities in modern software development: the reliability of code generated by artificial intelligence.
To understand why this topic matters right now, we must look at the current trajectory of software engineering. As large language models (LLMs) increasingly automate code generation-a trend colloquially referred to within the community as 'vibecoding'-the gross world lines of code (LoC) are exploding at an unprecedented rate. While this accelerates development, it introduces a massive, systemic risk: the potential for vast amounts of unverified, error-prone, or actively insecure code to be deployed into critical production environments. Traditional software testing methodologies are struggling to keep pace with the sheer volume and unpredictable nature of AI-produced software. Ensuring the correctness and safety of this code requires a paradigm shift toward mathematically rigorous validation techniques that can scale alongside AI capabilities.
The lessw-blog post details how the Secure Program Synthesis Fellowship intends to bridge this exact gap. The core argument of the initiative is that advanced software correctness techniques-specifically formal methods and secure program synthesis-must be adapted and applied directly to AI systems. Formal methods involve the use of mathematical specifications to prove that a program behaves exactly as intended, leaving no room for ambiguous edge cases. By bringing together experts in AI security and formal verification, the fellowship aims to develop robust interventions focusing on specification, validation, and adversarial robustness.
The program is structured to maximize impact through collaborative, mentor-led research. It kicks off with a dedicated hackathon on secure program synthesis topics scheduled for May 22-24, 2026. This will be followed by the core fellowship, a part-time research opportunity running from June through September 2026. The organizers note that direct AI security widgets and practical products will receive special consideration during the mentor review process, signaling a strong preference for applied, real-world solutions over purely theoretical research.
For researchers, security engineers, and formal methods practitioners, this initiative represents a significant opportunity to shape the foundational safety mechanisms of future software development. By participating as a mentor or fellow, contributors can help build the guardrails necessary for a future where AI writes the majority of our code.
We highly encourage professionals in the field to explore the details of the program, the specific technical challenges being targeted, and the application requirements. Read the full post to learn more about the fellowship and how to apply before the May 5th deadline.
\n\n