A single prompt sent to a public-facing AI agent could have exposed every Amazon Bedrock AgentCore agent in the same AWS account and region, according to new research from Zenity Labs. The attack chain, named AgentCorruption, gave the researchers access to private chats, source code, long-term...