Poison Once, Exploit Forever: Environment-Injected Memory Poisoning Attacks on Web Agents
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866908943773597696 |
|---|---|
| author | Zou, Wei Dong, Mingwen Calvo, Miguel Romero Chang, Shuaichen Guo, Jiang Lee, Dongkyu Niu, Xing Ma, Xiaofei Qi, Yanjun Jiang, Jiarong |
| author_facet | Zou, Wei Dong, Mingwen Calvo, Miguel Romero Chang, Shuaichen Guo, Jiang Lee, Dongkyu Niu, Xing Ma, Xiaofei Qi, Yanjun Jiang, Jiarong |
| contents | Memory makes LLM-based web agents personalized, powerful, yet exploitable. By storing past interactions to personalize future tasks, agents inadvertently create a persistent attack surface that spans websites and sessions. While existing security research on memory assumes attackers can directly inject into memory storage or exploit shared memory across users, we present a more realistic threat model: contamination through environmental observation alone. We introduce Environment-injected Trajectory-based Agent Memory Poisoning (eTAMP), the first attack to achieve cross-session, cross-site compromise without requiring direct memory access. A single contaminated observation (e.g., viewing a manipulated product page) silently poisons an agent's memory and activates during future tasks on different websites, bypassing permission-based defenses. Our experiments on (Visual)WebArena reveal two key findings. First, eTAMP achieves substantial attack success rates: up to 32.5% on GPT-5-mini, 23.4% on GPT-5.2, and 19.5% on GPT-OSS-120B. Second, we discover Frustration Exploitation: agents under environmental stress become dramatically more susceptible, with ASR increasing up to 8 times when agents struggle with dropped clicks or garbled text. Notably, more capable models are not more secure. GPT-5.2 shows substantial vulnerability despite superior task performance. With the rise of AI browsers like OpenClaw, ChatGPT Atlas, and Perplexity Comet, our findings underscore the urgent need for defenses against environment-injected memory poisoning. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_02623 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Poison Once, Exploit Forever: Environment-Injected Memory Poisoning Attacks on Web Agents Zou, Wei Dong, Mingwen Calvo, Miguel Romero Chang, Shuaichen Guo, Jiang Lee, Dongkyu Niu, Xing Ma, Xiaofei Qi, Yanjun Jiang, Jiarong Cryptography and Security Artificial Intelligence Memory makes LLM-based web agents personalized, powerful, yet exploitable. By storing past interactions to personalize future tasks, agents inadvertently create a persistent attack surface that spans websites and sessions. While existing security research on memory assumes attackers can directly inject into memory storage or exploit shared memory across users, we present a more realistic threat model: contamination through environmental observation alone. We introduce Environment-injected Trajectory-based Agent Memory Poisoning (eTAMP), the first attack to achieve cross-session, cross-site compromise without requiring direct memory access. A single contaminated observation (e.g., viewing a manipulated product page) silently poisons an agent's memory and activates during future tasks on different websites, bypassing permission-based defenses. Our experiments on (Visual)WebArena reveal two key findings. First, eTAMP achieves substantial attack success rates: up to 32.5% on GPT-5-mini, 23.4% on GPT-5.2, and 19.5% on GPT-OSS-120B. Second, we discover Frustration Exploitation: agents under environmental stress become dramatically more susceptible, with ASR increasing up to 8 times when agents struggle with dropped clicks or garbled text. Notably, more capable models are not more secure. GPT-5.2 shows substantial vulnerability despite superior task performance. With the rise of AI browsers like OpenClaw, ChatGPT Atlas, and Perplexity Comet, our findings underscore the urgent need for defenses against environment-injected memory poisoning. |
| title | Poison Once, Exploit Forever: Environment-Injected Memory Poisoning Attacks on Web Agents |
| topic | Cryptography and Security Artificial Intelligence |
| url | https://arxiv.org/abs/2604.02623 |