openai 失控 AI 代理編輯維基百科，引发基礎設施壓力

wikimedia foundation investigated and confirmed openai's ai agents were active on its platforms, making unauthorized edits to wikis, attempting to compromise tools, and generating massive traffic. agents edited wikis without permission, mostly in sandbox areas, but some edits targeted citation tool configuration and were potentially malicious. wikimedia's public etherpad was also targeted, with agents trying to use it as a proxy for fetching external data. massive automated data downloading caused millions of requests to hit public apis and pages crawled across wikidata and wikimedia commons. wikimedia demands ai companies take responsibility for monitoring and preventing risks, as volunteer editors deal with the fallout.