The field of code generation and security is rapidly evolving, with a focus on developing innovative solutions to improve the reliability, efficiency, and security of code generation systems. Recent research has explored the use of large language models (LLMs) to generate high-quality code, as well as techniques to improve the security and robustness of these systems. Notable advancements include the development of novel architectures and frameworks that enable more effective code generation, such as MemoCoder and PurpCode, which have demonstrated significant improvements in code quality and security. Additionally, research has highlighted the importance of evaluating and mitigating potential security risks associated with code generation systems, such as vulnerabilities to adversarial attacks. Overall, the field is moving towards more sophisticated and secure code generation systems, with potential applications in a wide range of areas, including software development, cybersecurity, and artificial intelligence. Noteworthy papers include MemoCoder, which proposes a multi-agent framework for collaborative problem solving and persistent learning, and PurpCode, which introduces a post-training recipe for training safe code reasoning models.
Advances in Code Generation and Security
Sources
Decompiling Rust: An Empirical Study of Compiler Optimizations and Reverse Engineering Challenges
MemoCoder: Automated Function Synthesis using LLM-Supported Agents
PurpCode: Reasoning for Safer Code Generation
Mut4All: Fuzzing Compilers via LLM-Synthesized Mutators Learned from Bug Reports
Running in CIRCLE? A Simple Benchmark for LLM Code Interpreter Security
MOCHA: Are Code Language Models Robust Against Multi-Turn Malicious Coding Prompts?
When Prompts Go Wrong: Evaluating Code Model Robustness to Ambiguous, Contradictory, and Incomplete Task Descriptions
Security Challenges in AI Agent Deployment: Insights from a Large Scale Public Competition
LanternNet: A Novel Hub-and-Spoke System to Seek and Suppress Spotted Lanternfly Populations
Vulnerability Mitigation System (VMS): LLM Agent and Evaluation Framework for Autonomous Penetration Testing
Curiosity by Design: An LLM-based Coding Assistant Asking Clarification Questions
MultiAIGCD: A Comprehensive dataset for AI Generated Code Detection Covering Multiple Languages, Models,Prompts, and Scenarios
Introducing HALC: A general pipeline for finding optimal prompting strategies for automated coding with LLMs in the computational social sciences
DeepGo: Predictive Directed Greybox Fuzzing
RedCoder: Automated Multi-Turn Red Teaming for Code LLMs
Fuzzing: Randomness? Reasoning! Efficient Directed Fuzzing via Large Language Models
CodableLLM: Automating Decompiled and Source Code Mapping for LLM Dataset Generation
CodeEvo: Interaction-Driven Synthesis of Code-centric Data through Hybrid and Iterative Feedback
Prompt Optimization and Evaluation for LLM Automated Red Teaming
Ensemble Fuzzing with Dynamic Resource Scheduling and Multidimensional Seed Evaluation
IFEvalCode: Controlled Code Generation
Protecting Vulnerable Voices: Synthetic Dataset Generation for Self-Disclosure Detection
AutoBridge: Automating Smart Device Integration with Centralized Platform
Geak: Introducing Triton Kernel AI Agent & Evaluation Benchmarks