OpenAI confirms agents were involved in May RubyGems incident

OpenAI confirmed that its agents used RubyGems during a May incident in which more than 2,000 packages were submitted over two days, according to reporting based on public package records and researcher analysis. The packages used RubyDoc.info documentation builds to run scripts and retrieve public data, the report says. Researchers also found attempts to exploit a then-unreported route that could expose API keys. RubyGems says the route was viable but found no evidence that keys were used.