⚡Megatron LLM Hub
←Back to search
Twitter

研究者称把 agent 目标限定为单次写入更易满足,配合 WCP 表基址即可任意代码执行

ai_agent浏览器漏洞利用
@Xbow @darkfloyd1014 5. Restricting it to a single write was useful because that's a very clear goal for an agent, it's a lot easier to satisfy than a more complex constraint, and since it had the WCP table base_ technique it was sufficient to get to arbitrary code at that point.