LLMs can generate vast amounts of code to accomplish the goals dictated by a user. But many are using LLMs to generate code in programming languages that were designed to build anything, not just business logic or “enterprise” applications that LLM users are wanting to build. It is possible that an undertrained user would not be able to recognize generated code that does more than what was asked to do (e.g.: a server listening on a port or a file removal process), whether due to a deficient LLM or due to malicious intervention in the code generation process.

Furthermore, those generic languages have package systems that might introduce supply-chain attacks in the generated code or in the process of generation.

Why ask LLMs to generate Python, C#, Java code when the same goal can be accomplished by a more constraint language? This session will discuss languages that popular LLMs can target as alternatives to generating code in languages that have more power to do damage.