A single prompt was enough to hijack every AI agent in an AWS account, Zenity researchers found

AchievementResearchOct 8, 2026

Zenity Labs reported a chain of vulnerabilities in Amazon Bedrock AgentCore to AWS on December 25, 2025, allowing a single prompt sent to one public agent to take over every agent in the same AWS account and region. The flaws, dubbed "AgentCorruption," exposed credentials, source code, private conversations, and agent memory. AWS has since tightened metadata access and default execution role permissions, though Zenity Labs recommends custom minimal-access roles.

Receipt № 22881 source · awaiting confirmation ◐

Evidence

1source· awaiting independent confirmation

No score is assigned. Sources and their independence are shown in the citation chain below.

Citation chain · 1 source

01med

the-decoder.comBloghttps://the-decoder.com/a-single-prompt-was-enough-to-hijack-every-ai-agent-in-an-aws-account-zenity-researchers-found/

PRIMARY

AWSCompanyAmazonCompanyAmazon Bedrock AgentCoreCompanyZenity LabsCompany

Canonical: https://the-decoder.com/a-single-prompt-was-enough-to-hijack-every-ai-agent-in-an-aws-account-zenity-researchers-found/

A single prompt was enough to hijack every AI agent in an AWS account, Zenity researchers found

Evidence

Citation chain · 1 source

Related