MosaicLeaks: Can your research agent keep a secret?

# MosaicLeaks: When AI Research Tools Accidentally Spill Secrets Researchers at Mosaic discovered that AI agents designed to help with research tasks can accidentally leak sensitive information—like passwords, private documents, or confidential data—when they search through files or browse the internet to complete their assignments. This is a real security risk for companies using these tools because the AI might expose secrets without realizing it's doing something wrong. The finding highlights an important gap: we need better safeguards to make sure AI assistants don't compromise confidential information while doing their jobs.