SEMLA: Securing Enterprises via Machine-Learning-based Automation
| Reference number | |
| Coordinator | Kungliga Tekniska Högskolan - DIVISION OF SOFTWARE AND COMPUTER SYSTEMS |
| Funding from Vinnova | SEK 9 063 845 |
| Project duration | November 2023 - October 2025 |
| Status | Completed |
| Venture | Advanced digitalization - Enabling technologies |
| Call | Cyber security for industrial advanced digitalization 2023 |
Important results from the project
The project developed AI-based methods to improve secure and efficient software development. We built Prometheus, combining large language models with verification to reason about code, and advanced LLM-based security analysis, including a first benchmark for network configuration. We also delivered faster LLM inference and scalable model techniques. The results are broadly applicable to trustworthy AI-based software development with long-term industrial impact.
Expected long term effects
In the longer term, the project’s results are expected to have impact beyond network automation. The methods and systems developed are broadly applicable to software development, particularly in how AI and LLMs can be combined with verification, performance modeling, and system-level guarantees. By addressing challenges such as correctness, security, scalability, and cost early on, the project explored the future of AI-based coding in a timely manner.
Approach and implementation
The project was structured around clearly defined tasks covering AI-assisted security analysis, code generation and verification, and ML infrastructure. The activities were appropriate and the project was implemented as intended. Work progressed according to schedule with no significant delays. Collaboration between partners worked well. An important outcome was the confirmation that verifying complex software remains an open research challenge, requiring further work.