June 16, 2026

Goals

• Find out more exploitable regions inside the libtriton.so library.

What I worked on

• Read about the linux internals.
• Learnt how to use vLLM for serving LLM models.

Key Insights

• vLLM has a provision of serving LLM models using kernel optimizized using triton.

Questions / Confusions

• None

Ideas / Hypotheses

• None

Papers / Resources

• None

Next Steps

• Try to corrupt the LLM output accross different architectures.

Introspection

• I need to go home.