Model2Kernel: Model-Aware Symbolic Execution For Safe CUDA Kernels
2026-03-27T08:52:01Z•dcbb2cbb7eab95d3a01e88b96dabf58e0ae30065ca116e1922b3822e3c85d318
AI-safetyCUDAGPULLM-inferenceRAGdynamic-analysisformal-verificationmemory-corruptionmemory-safetymodel-securityprogram-synthesisself-evolving-agentsstatic-analysissymbolic-executiontrusted-AIvulnerability-discovery
What happened
Collection of recent CS submissions with strong relevance to software/hardware correctness and AI safety. Most notable is Model2Kernel: a model-aware dynamic analysis plus CUDA-specialized symbolic execution system that found 353 previously unknown memory-safety bugs in CUDA kernels used for LLM inference (only nine false positives), with implications for weight corruption, crashes, and adversarial attacks in GPU-accelerated inference stacks. Other papers cover formal/design-time verification for trustworthy AI, verified synthesis of self-evolving agents (SEVerA) that enforces hard contracts,L
Why it matters
A reviewed impact interpretation has not been published for this record.
Evidence and limitations
- Source ID
- arxiv_cs_pl
- Record identifier
- dcbb2cbb7eab95d3a01e88b96dabf58e0ae30065ca116e1922b3822e3c85d318
- Enrichment time
- 2026-03-27T08:52:01Z
- AI-assisted enrichment
- Yes
This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.