Mmastodon TechnologyAI first seen 2 d ago, last 2 d ago, peak #1
Researchers show one prompt injection can hijack AWS Bedrock agents
Original: Zenity Labs has shown what one prompt could do to AWS Bedrock AgentCore. Agents could be talked into fetching credential
Security researchers at Zenity Labs demonstrated that a single prompt injection against AWS Bedrock AgentCore could trick an AI agent into retrieving credentials from the instance metadata service, handing the attacker the agent's IAM role. That role was broadly scoped, allowing invocation of every other agent, raising concerns about over-permissive defaults in agentic cloud infrastructure.
Why now: The disclosure highlights how easily AI agents in cloud environments can be weaponised through prompt injection, a pressing concern as more companies deploy them.
Zenity LabsAWSBedrock AgentCoreIAM
Rank over time, top of the chart is #1. 2 snapshots from 2 d ago to 2 d ago.
Evidence
- Zenity Labs has shown what one prompt could do to AWS Bedrock AgentCore. Agents could be talked into fetching credentials from the instance metadata service. One prompt injection, and the attacker held the agent's IAM role. That role was wide. It could invoke every other agent… · adrianhollister@infosec.exchange · 2
API: https://socialmediatrends-api.osmike.com/v1/trends/2211963