KAIJU: An Executive Kernel for Intent-Gated Execution of LLM Agents
2026-04-06T08:51:48Z•5882a85c681c53b18eba5dc7e4b0b4e66a87af06adf796a5b980f2008c5bb9a4
GBQAHTTP fingerprintingIaC disambiguationLLM agentsMITRE ATT&CKMPIRust safetySZZ algorithmTTP extractionagent architectureautomated program repairbenchmarksdeveloper analyticshigh-performance computinginfrastructure-as-codeintent-gated executionmulti-agent systemsprogram synthesisprompt injectionruntime tracessecurity controlstool calling
What happened
Collection of new CS-SE arXiv papers focused on LLM-driven software engineering, agent architectures, and automated program analysis. Key contributions include KAIJU — an Executive Kernel with Intent-Gated Execution (IGX) to decouple LLM planning from tool execution and enforce authorization boundaries; a systematic review of automated TTP extraction and its methodological gaps; Ambig-IaC, a disagreement-driven approach to disambiguate Infrastructure-as-Code generation; and improvements to MPI bug detection/repair using FSL, CoT and RAG to raise detection accuracy. Other papers present a Petri
Why it matters
A reviewed impact interpretation has not been published for this record.
Evidence and limitations
- Source ID
- arxiv_cs_se
- Record identifier
- 5882a85c681c53b18eba5dc7e4b0b4e66a87af06adf796a5b980f2008c5bb9a4
- Enrichment time
- 2026-04-06T08:51:48Z
- 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.