Grimlock: Guarding High-Agency Systems with eBPF and Attested Channels

2026-05-28T07:23:32Ze6cbd8c737ca9dfdf32a5195c01a8f651397ffe925fa6feafe540aa17f19ff50
AES-like ciphersCWE-20CloakGroundedCacheHARPMDS matricesORAMPoison-with-StyleRAGTLS 1.3agent-to-agent securitycache safetycode LLMcryptanalysisdata/model poisoningeBPFharm amplificationkTLSmodel poisoningmulti-agent LLMpost-handshake attestationquantum cryptography','uncloneable encryption'related-differential cryptanalysistemporal localityunsafe-served rate

What happened

This collection of newly announced arXiv papers covers security-relevant advances across agent systems, LLM orchestration, caching, cryptography, mobile privacy, and storage privacy. Highlights: Grimlock — an eBPF-enforced Agent Guard combining traffic interception, post-handshake attestation bound to TLS1.3 channel bindings, and short-lived channel-bound scope tokens to enforce authorization and provenance across sandboxes; HARP — a trace-first methodology measuring harm amplification in multi-agent LLM systems and showing how local compromises can propagate and amplify system-level harm; PoS

Why it matters

A reviewed impact interpretation has not been published for this record.

Evidence and limitations

Source ID
arxiv_cs_cr
Record identifier
e6cbd8c737ca9dfdf32a5195c01a8f651397ffe925fa6feafe540aa17f19ff50
Enrichment time
2026-05-28T07:23:32Z
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.