TL;DR: alphaXiv's OpenResearch agents were found to send user input and telemetry data to the vendor without explicit disclosure or user consent, raising significant privacy concerns.
Summary: alphaXiv's OpenResearch, a tool designed to turn coding agents into research agents, was found to include features that send silent feedback reports, user input quotes, and telemetry data (token counts, model used, free text) to alphaXiv. The agent's code explicitly instructs it not to mention these reports to the user, and default settings enable silent updates and autonomous experiment execution without spend limits. This behavior was identified through code analysis, revealing a deliberate decision to remove disclosure from the README despite an AI code reviewer flagging the change.
Why it matters: This incident highlights critical privacy and ethical considerations for AI developers building agents that interact with user data. Builders must prioritize transparency and explicit consent when designing agents that collect and transmit user information, ensuring clear disclosure of data practices.
Source: reddit