按年被提到
20262026 · 2 次2026
AI 安全、伦理与社会影响›智能体安全与滥用·arXiv:2607.24625 · cs.CR

APPA: Recoverable Information-Flow Control for Real-World LLM Agents

Arseny Kravchenko, Vadim Liventsev, Innokentii Konstantinov, Ildar Iskhakov, Matvey Kukuy · 2026
开始阅读 ↗PDFalphaXiv 讨论
被提及 · 2 人首次 2026-07-28 · 最近 2026-09-28
2不同的人
2活跃月份
2讨论串
1被引(OpenAlex)
arXiv 摘要(候选层不写导读)

LLM agents deployed in practical workflows routinely mix private context, untrusted tool and web outputs, and external side effects. While information-flow control (IFC) provides structural defenses against prompt injection, data exfiltration, and confused-deputy attacks, conventional IFC relies on monotone taint tracking that either over-blocks benign operations or permanently strands downstream execution once an agent ingests unvetted data. We present APPA (Agentic Permissions Policy Algebra), which turns agent IFC from an abort-only barrier into a policy-governed recovery system. APPA enforces a dual-phase reference monitor at tool dispatch and protocol gateways (e.g., Model Context Protocol): before tool execution, it prospectively evaluates composite label restrictions and workflow history; upon completion, it validates realized outputs before context admission. For incremental rollout across unannotated tools, APPA incorporates gradual security typing with bounded cast resolution. To inspect untrusted data without poisoning primary agent context, APPA introduces on-demand trajectory confinement: disposable child branches absorb taint locally and exit through shape-bounded channels (attest-schema) with exact parent-label and transcript preservation, avoiding permanently partitioned multi-agent infrastructure. We prove core safety invariants: no-laundering gradual resolution, branch boundary isolation, and recovery containment against prompt-injected models. Across 6,600 controlled benchmark episodes spanning OWASP AgentThreatBench and enterprise workflows (Bench-Corp), APPA sustains 64.2-91% utility with zero observed attacks across 1,320 guarded episodes, establishing a practical defense for deployed tool-using agents.

候选层:至少 2 个不同的人、在 2 个不同月份提到过,还没到主书架的 3 人 3 个月。没有做立场标注,所以只写“被提及”。

HN 书评

推荐在前,同类按时间 · 每篇最多保留 20 条,统计用全部提及
评论motakuk · 2026-09

“Hi Hacker News! Matvey, one of the authors, is here. While building enterprise agents, we ran into a problem: the more tools you connect to the AI, the higher the chance it will run out of control and leak sensitive data. Guardrails, in theory, should prevent this, but the situation is worrying: - Non-deterministic guardrails (LLM as a judge, auto modes…”

Show HN: OpenAPPA – open-source deterministic guardrails that don't break agents原评论 ↗
被提交为 story · 1 次