CapChain: A Capability-Token Access Control Architecture with Verifiable Provenance for Multi-Agent LLM Systems
CapChain: архитектура управления доступом на основе токенов возможностей с проверяемым происхождением для мультиагентных систем на базе больших языковых моделей
2026-08-04
SCID: 54.1/cdkpsp3w
Discuss with AI
MAGPIE contextual-privacy benchmarkcapability-aware reducercapability-token access controlmulti-agent LLM systemsverifiable provenance
Figures from the paper
Abstract (AI)
Multi-agent large language model (LLM) systems built on LangGraph-style frameworks share an unencrypted state store with no field-level read access control, so one agent can read another’s private fields. The 2026 OWASP Agentic Top 10 and CVE-2025-67644 show that this layer is already being exploited. We propose CapChain, a replacement for the step that merges each agent’s output into shared state, with three primitives: capability tokens that cryptographically bind which agent may read which field, a tamper-evident provenance log with logarithmic-cost audit, and a capability-aware reducer that decides whether each field may be forwarded, redacting and logging unauthorized merges. We evaluate the reducer’s attribute–purpose gate on the MAGPIE contextual-privacy benchmark, with scoring based on two deterministic tiers: verbatim matching and keyword-and-entity salience. Across 150 conditions (25 scenarios × 3 open-source models × 2 instruction modes), the Salient tier triggers the redactor for 4.67% of inter-agent messages. This rate counts the reducer’s gate activations (redaction decisions), not leakage prevented. By relocating enforcement from the language model to the state-transition layer that graph-structured orchestrators expose, CapChain brings cryptographic, capability-based access control and audit-by-construction provenance to a structural locus that prompt-level defenses cannot reach.
Key Findings
1
CapChain replaces shared-state merging with capability tokens, tamper-evident provenance logging, and a capability-aware reducer for multi-agent LLM access control.
2
Capability tokens cryptographically bind agents to permitted fields, while the reducer redacts and logs unauthorized field forwarding during state transitions.
3
On the MAGPIE contextual-privacy benchmark, the Salient-tier gate activated redaction for 4.67% of inter-agent messages across 150 conditions.
4
The provenance log supports tamper-evident auditing with logarithmic cost, providing audit-by-construction for agent interactions.
5
The reported 4.67% measures reducer gate activations and redaction decisions, not the amount of leakage prevented.
Research Object
shared-state merging in multi-agent LLM systems built on LangGraph-style frameworks
Research Subject
field-level access-control enforcement, unauthorized-merge redaction, and verifiable provenance auditing at the state-transition layer
Publication Details
Publication Date
2026-08-04
Journal
Publisher
ISSN
Cited by
1
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest
References available in scid.ai4
Not What You've Signed Up For: Compromising Real-World LLM-Integrated Applications with Indirect Prompt Injection2023
AgentLeak: A Benchmark for Internal-Channel Privacy Leakage in Multi-Agent LLM Systems2026
TRiSM for Agentic AI: A review of Trust, Risk, and Security Management in LLM-based Agentic Multi-Agent Systems2026
SAGA: A Security Architecture for Governing AI Agentic Systems2026