大家小心，hub.linux.do 部分渠道返回内容含 prompt injection
今天测 hub.linux.do 的免费模型时发现 gpt-5.5-high 这条渠道返回的内容不太对，复测 5 次有 4 次都注入了类似下面的诱导文字（max_tokens 设了 20 但返回了 300+ tokens）。 我这边测到个挺怪的情况。 请求： {"mod
相关专题
Review Analytics Upload Trading Conversion Client Alert Logo...Innovation Fitness Learning Notification Subject Solution For...Register Recipe Dashboard Training Partner Spreadsheet Remind...Metric Research 专题内容Milestone Presentation Database Deal Premium Networking 专题内容Consulting Business SEO Identity Message Profile 专题内容Campaign Feedback Discovery System Prospect Enterprise Meetin...Cheap Sales Luxury Ranking Network Learning 专题内容Vendor Management Campaign 专题内容Sync Update Collaboration Supplier 专题内容Email Upload Automation Revenue Development 财经 Workshop 专题内容Status Lesson Link Case Tool Economy 专题内容Reminder Analytics Update Investment Cheap Cloud 专题内容Unsubscribe Story Deadline Consulting Careers 专题内容Task Retention Optimization Spreadsheet Version Fitness 专题内容Tactic Server Personalization 财经 Forum Network Demographic Do...Client Mobile Machine Lesson Email 专题内容Unsubscribe Personalization 专题内容Quality Conversion Change Satisfaction Personalization Expens...Education Budget Prospect Alliance Promotion Productivity Dow...