基于攻击意图的信任传递攻击检测
网络出版日期: 2026-05-13
基金资助
国家自然科学基金(U25B2026)
版权
Detecting trust-propagation attacks with attack-intent awareness
Online published: 2026-05-13
Copyright
面向数字身份的网络攻击已从单点突破向链式劫持攻击模式演化。攻击者获取具备上游信任属性的身份凭据后,可在认证协议设计规范内生成任意下游凭据,实现合法形式下的未授权访问,此类攻击被定义为信任传递攻击。与之对应,传统基于单点的异常检测对凭据滥用覆盖不足,难以有效检测信任传递攻击。为此,提出一种以数字身份生命周期为核心的一致性检测方法将数字身份活动抽象为“身份认证 — 身份授予 — 身份使用”三阶段,通过跨阶段时序约束与数据流一致性校验判定下游凭据合法性。提出自适应动态时间窗口,用于评估与判定信任传递过程的时序约束;设计基于攻击意图的行为链自动收缩机制,在同一窗口内区分主凭据与相关凭据,剔除系统日志中的各类噪声,将日志数据收敛至支撑结论的最小证据集。最后,设计的GoldenAuditer系统经评估实验验证,在检测准确率、误报率、漏报率、行为链收缩比等指标上,较传统基线方案具有显著优势,可在复杂噪声与时延条件下准确检测信任传递攻击。
孙维政 , 殷丽华 , 杨佳庚 , 骆俊杰 , 游贵鹏 , 朱金玉 . 基于攻击意图的信任传递攻击检测[J]. 网络空间安全科学学报, 2026 , 4(1) : 92 -104 . DOI: 10.20172/j.issn.2097-3136.260205
Cybersecurity attacks targeting digital identities have evolved from the single-point breach mode to the chain-driven hijacking mode. Once attackers obtain credentials with upstream trust attributes, they can derive any arbitrary downstream credentials within the design specifications of authentication protocols and gain unauthorized access in a legally compliant form. This type of attack is referred to as a trust-propagation attack. Traditional single-point-based anomaly detection has inadequate coverage for credential abuse, thus failing to effectively detect trust-propagation attacks. To address this issue, a core-centered consistency detection approach based on the digital identity lifecycle is proposed. This approach abstracts digital identity activities into three phases: identity authentication, identity authorization, and identity usage. It determines the legitimacy of downstream credentials by verifying cross-phase temporal constraints and data flow consistency. an adaptive dynamic time window is proposed to evaluate and validate the temporal constraints of the trust-propagation process. In addition, an automatic intent-driven behavior-chain contraction mechanism is designed. This mechanism distinguishes primary credentials from associated credentials within the same window, filters various types of noise in system logs, and converges log data to a minimal evidence set that supports the conclusions. Finally, the designed GoldenAuditer system is validated through evaluation experiments, which achieves higher detection accuracy, lower false positive and false negative rates, and a more optimal behavior-chain contraction ratio compared with traditional schemes. It can accurately detect trust-propagation attacks under the conditions of complex noise and time delays.
表 1 常见的信任传递攻击的技战术Table 1 Tactics and techniques of common trust propagation attacks |
| 技战术 | 描述 | ATT&CK ID |
| 黄金票据 | 离线伪造票据TGT | T1558.001 |
| 白银票据 | 伪造服务票据ST | T1558.002 |
| Kerberoasting | 离线破解服务票据以获哈希 | T1558.003 |
| AS-REP Roasting | 利用免预认证账户进行 离线密码恢复 | T1558.004 |
| DCSync(目录复制模拟) | 模拟复制导出域账户哈希 | T1003.006 |
| 算法1 自适应时间窗口算法 |
| 输入 训练期事件流 S,锚点集合 A,分桶规则 Bucket,字段集合 F 输出 各事件桶对应的窗口上界表 W 1. W = Ø 2. FOR each bucket b 3. 4. 5. FOR each anchor a ∈ A with Bucket(a) = b 6. Win = Neighbor(S, a, T) //固定长候选窗 7. G = Candidates(Grant, Win)//候选 Grant 8. 9. g*=argmax_{g∈G}Score(a,g,F) // 以锚点选择主Grant 10. a* =argmax_{x∈A0} Score(g*, x, F) 11. // 为主 Grant 选择 Auth 12. IF g* exists AND a* exists THEN 13. 14. ENDIF 15. IF g* exists THEN 16. 17. ENDIF 18. ENDFOR 19. W[b].auth_to_grant = 20. W[b].grant_to_use= 21. ENDFOR 22. return W |
| 算法2 动态窗配对与判断算法 |
| 输入:检测期事件流 S,锚点集合 A,窗口上界表 W,字段集合 F 输出:事件标注结果 R(Normal / Suspicious,附最小解释链) 1. R = Ø 2. FOR each anchor a ∈ A 3. b = Bucket(a) 4. 5. WinG =[time(a)−W[b].grant_to_use, time(a)] 6. G =Candidates(Grant, WinG) 7. IF G = Ø THEN 8. R = R ∪ { (a, Normal, Chain = {a}) } 9. CONTINUE 10. ENDIF 11. g*= argmax_{g∈G} Score(a, g, F) 12. WinA=[time(g*)−W[b].auth_to_grant, time(g*)] 13. A0 =Candidates(Auth, WinA) 14. a*= argmax_{x∈A0} Score(g*, x, F) 15. E = G \ {g*} 16. extra={ e ∈ E : NoMatch(Auth, e, WinA, F) } 17. IF a* = null OR |extra| > 0 OR Inconsistent(a, g*, F) THEN 18. R=R ∪ { (a, Suspicious, Chain = {a*, g*, a}, Extras = extra) } 19. ELSE 20. R = R ∪ { (a, Normal, Chain = {a*, g*, a}) } 21. ENDIF 22. ENDFOR 23. RETURN R |
表 2 合法闭环Table 2 Legitimate closed loop |
| 类型 | 时间示例 | 主体 | 来源地址 | 服务类别 | 选项(hex) | 备注 |
| Auth | 10:15:02 | admin | 192.168.1.10 | 认证中心 | 0x40810010 | 与 Grant* 匹配 |
| Grant* | 10:15:18 | admin | 192.168.1.10 | 目录/账户服务 | 0x40810010 | 主凭据 |
| Grant | 10:15:18 | admin | 192.168.1.10 | 中心签发/转介 | 0x60810010 | 可解释的转介 |
| Use | 10:15:30 | admin | 192.168.1.10 | 账户操作 | NULL | 锚点 |
| 算法3 基于攻击意图的行为链自动收缩算法 |
| 输入 锚点 a,候选集合 输出 最小事件解释链 C,行为链收缩比p 1. C = {a} 2. g*= argmax_{g∈G} Score(a, g, F) 3. a* =argmax_{x∈A0} Score(g*, x, F) 4. IF a* ≠ null THEN C ← C ∪ {a*} 5. IF g* ≠ null THEN C ← C ∪ {g*} 6. FOR each g ∈ G \ {g*} 7. IF Service(g) 8. DROP g 9. ELSE IF NotSameIPOrSubject(a, g, F) THEN 10. DROP g 11. ELSE 12. MARK g as Extra 13. ENDIF 14. ENDFOR 15. p=1 − ( |C| / |{events within anchor window}| ) 16. RETURN (C, p) |
表 3 信任传递断链Table 3 Broken trust-propagation chain |
| 类型 | 时间 | 主体 | 来源地址 | 服务类别 | 选项(hex) | 备注 |
| Grant* | 08:47:57 | admin | 192.168.241.137 | 目录/账户服务 | 0x40810000 | 主凭据(无匹配认证) |
| Grant | 08:47:57 | admin | 192.168.241.137 | 中心签发/转介 | 0x60810010 | 无法由认证解释 |
| Use | 08:47:57 | admin | 192.168.241.137 | 账户操作 | NULL | 锚点 |
| (无)Auth | NULL | NULL | NULL | NULL | NULL | 在窗口内未发现 |
表 4 实验环境Table 4 Experimental environment |
| 软硬件 | 相关参数 |
| 操作系统DC机 | Windows Server 2019 |
| 操作系统Client机 | Windows 10 x64 |
| 处理器 | Intel(R)Corei7-10870H@2.20GHz |
| DC/Client内核版本 | 版本1809(OS |
表 5 实验数据Table 5 Experimental data |
| 指标 | 数量 |
| 总行 | 20 699 |
| 4720事件(账户创建) | 1 489 |
| 4768事件(TGT票据申请) | 1 804 |
| 4769事件(TGS票据颁发) |
注:为匹配“身份认证 — 身份授予 — 身份使用”的生命周期一致性模型,程序仅在4720事件(账户创建)的锚点窗口内检索可关联的4768事件(TGT 票据申请)、4769事件(TGS 票据颁发),用于数字身份一致性判定,其余事件作为背景噪声不参与判定,该方式为系统日志审计领域的常用方法。 |
表 6 正常行为的事件链条与攻击行为的异常链条Table 6 Typical normal and anomalous patterns within an anchor window |
| 形态(匿名案例) | 窗内关键结构(最小证据) | 静态窗基线倾向判定 | 本文法判定 |
| 正常闭环(N) | Auth ✓ + primary(TGS) ✓ + Anchor(4720) ✓,无 EXTRA_ST | 正常/已配对 | 正常 |
| 异常分叉(A) | primary(TGS) ✓ + Anchor ✓,但同窗出现 EXTRA_ST(krbtgt) | 易“配对成功”而误报 | 可疑 |
| 异常断链(B) | Anchor ✓,出现敏感服务 ST,但在 WinA 内 无匹配 Auth | 可疑/不稳定(依赖阈值) | 可疑 |
表 7 不同检测方法的总体性能对比Table 7 Overall performance comparison of different methods |
| 方法 | 准确率 | 召回率 | F1 | TP | FP | TN | FN |
| 基线1 | 17 | 102 | 4 | ||||
| 基线2 | 3 | 1 | 18 | ||||
| GoldenAuditer | 20 | 1 | 1 |
表 8 日志处理开销对比Table 8 Log-processing overhead comparison |
| 方法 | 处理日志行数 | 耗时/s | 吞吐 (行/s) | CPU占用率 | 峰值内存/MB |
| 基线1 | 2.60 | 65% | 60 | ||
| 基线2 | 43.09 | 480.37 | 45% | 80 | |
| GoldenAuditer | 1.97 | 35% | 44 |
| 1 |
Cheng Z J, Lv Q J, Liang J Y, et al. Kairos: practical intrusion detection and investigation using whole-system provenance[C]//Proceedings of the 2024 IEEE Symposium on Security and Privacy (SP). Piscataway: IEEE Press, 2024: 3533-3551.
|
| 2 |
Microsoft. Data execution prevention[EB/OL]. (2023-05-01)[2025-12-24]. https://learn.microsoft.com/en-us/windows/win32/memory/data-execution-prevention.
|
| 3 |
Binosi L, Barzasi G, Carminati M, et al. The illusion of randomness: an empirical analysis of address space layout randomization implementations[C]//Proceedings of the 2024 on ACM SIGSAC Conference on Computer and Communications Security. New York: ACM, 2024: 1360-1374
|
| 4 |
Hasel M G, Wester I L, Paci F, et al. Mitigating privilege misuse in access control through anomaly detection[C]//Proceedings of the 18th International Conference on Availability, Reliability and Security. New York: ACM, 2023. DOI: 10.1145/3600160.3604988.
|
| 5 |
JPCERT/CC. Detecting lateral movement through tracking event logs (Version 2)[EB/OL]. (2017-12-05)[2025-12-24]. https://www.jpcert.or.jp/english/pub/sr/DetectingLateralMovementThroughTrackingEventLogs_version2.pdf.
|
| 6 |
SigmaHQ. Sigma: main Sigma rule repository[EB/OL]. (2025-12-21)[2025-12-24]. https://github.com/SigmaHQ/sigma.
|
| 7 |
Albasheer H, Md S M, Mubarakali A, et al. Cyber-attack prediction based on network intrusion detection systems for alert correlation techniques: a survey[J]. Sensors, 2022, 22 (4): 1494.
|
| 8 |
Liu Q, Bao K B, Hassan W U, et al. HADES: detecting active directory attacks via whole network provenance analytics[EB/OL]. (2024-07-26)[2025-12-24] https://doi.org/10.48550/arXiv.2407.18858
|
| 9 |
Padhiar S, Patel R. Outside the closed world: on using machine learning for network intrusion detection[C]//ICT for Intelligent Systems. Singapore: Springer, 2023: 265-270.
|
| 10 |
Della Penna S, Natella R, Orbinato V, et al. CTI-HAL: a human-annotated dataset for cyber threat intelligence analysis[EB/OL]. (2025-04-08)[2025-12-24]. https://doi.org/10.48550/arXiv.2504.05866
|
| 11 |
Luo K, Wang X, Fung P H A, et al. Universal cross-app attacks: exploiting and securing OAuth 2.0 in integration platforms[C]. 34th USENIX Security Symposium (USENIX Security '25), Seattle, WA, USA, USENIX Association, 2025: 3221-3238.
|
| 12 |
Grippo T, Kholidy H A. Detecting forged kerberos tickets in an active directory environment[EB/OL]. (2022-12-30)[2025-12-24]. https://doi.org/10.48550/arXiv.2301.00044
|
| 13 |
Matsuda W, Fujimoto M, Mitsunaga T, et al. Detection of the silver ticket for seamless single sign-on focusing on a ticket lifetime[J]. Journal of Information Processing, 2025, 33, 156- 167.
|
| 14 |
Hutchins E M, Cloppert M J, Amin R M. Intelligence-driven computer network defense informed by analysis of adversary campaigns and intrusion kill chains[EB/OL]. (2014-04-03)[2025-12-24]. https://www.nationalcyberwatch.org/resource/intelligence-driven-computer-network-defense-informed-by-analysis-of-adversary-campaigns-and-intrusion-kill-chains-2/.
|
| 15 |
Truong C, Oudre L, Vayatis N. Selective review of offline change point detection methods[J]. Signal Processing, 2020, 167, 107299.
|
| 16 |
杨佳庚, 方滨兴, 冀甜甜, 等. 锁闭保护: 基于程序行为分析的非预期执行攻击阻断[J]. 计算机学报, 2024, 47 (7): 1697- 1712.
Yang J G, Fang B X, Ji T T, et al. Lockdown-protection: unintended execution attack prevention based on program behavior analysis[J]. Chinese Journal of Computers, 2024, 47 (7): 1697- 1712.
|
| 17 |
Ning X R, Jiang J. Defense-in-depth against insider attacks in cyber-physical systems[J]. Internet of Things and Cyber-Physical Systems, 2022, 2: 203-211.
|
| 18 |
Elmagarmid A K, Ipeirotis P G, Verykios V S. Duplicate record detection: a survey[J]. IEEE Transactions on Knowledge and Data Engineering, 2007, 19 (1): 1- 16.
|
| 19 |
姚前, 张大伟. 区块链系统中身份管理技术研究综述[J]. 软件学报, 2021, 32 (7): 2260- 2286.
Yao Q, Zhang D W. Survey on identity management in blockchain[J]. Journal of Software, 2021, 32 (7): 2260- 2286.
|
| 20 |
张富, 杨琳艳, 李健伟, 等. 实体对齐研究综述[J]. 计算机学报, 2022, 45 (6): 1195- 1225.
Zhang F, Yang L Y, Li J W, et al. An overview of entity alignment methods[J]. Chinese Journal of Computers, 2022, 45 (6): 1195- 1225.
|
| 21 |
Cugola G, Margara A. Processing flows of information: from data stream to complex event processing[J]. ACM Computing Surveys, 2012, 44 (3): 1- 62.
|
| 22 |
葛琳, 季新生, 江涛. 基于关联规则的网络信息内容安全事件发现及其Map-Reduce实现[J]. 电子与信息学报, 2014, 36 (8): 1831- 1837.
Ge L, Ji X S, Jiang T. Discovery of network information content security incidents based on association rules and its implementation in Map-Reduce[J]. Journal of Electronics & Information Technology, 2014, 36 (8): 1831- 1837.
|
| 23 |
Center for Threat-Informed Defense. Adversary emulation library: OilRig emulation plan[EB/OL]. (2020-09-10)[2025-12-24]. https://github.com/center-for-threat-informed-defense/adversary_emul ation_library/tree/master/oilrig.
|
| 24 |
Abo-alian A, Youssef M, Badr N L. A data-driven approach to prioritize MITRE ATT&CK techniques for active directory adversary emulation[J]. Scientific Reports, 2025, 15, 27776.
|
| 25 |
Rajesh P, Alam M, Tahernezhadi M, et al. Analysis of cyber threat detection and emulation using MITRE attack framework[C]//Proceedings of the 2022 International Conference on Intelligent Data Science Technologies and Applications (IDSTA). Piscataway: IEEE Press, 2022: 4-12.
|
| 26 |
Virk S G, Iqbal J, Ali A, et al. Advancing security operations centers: modern use cases, MITRE ATT&CK integration, and coverage optimization in 2025[J]. Journal of Computing & Biomedical Informatics, 2025, 9 (2)
|
/
| 〈 |
|
〉 |