EarlyNarratives
Pricing
Start free trialSign in
Start free trialSign in
Signal

Google ADK AI workflows vulnerable to agent-to-agent attacks enabling pull request tampering

Evidence first: scan the strongest sources, then decide whether to go deeper.

Published 2026-08-04 10:54 UTCUpdated 2026-08-04 11:39 UTC
rss
cveexploitssecurity_toolingincident_response
Source links open
Source links and full evidence are open here. Archive history, compare-over-time, alerts, exports, API, integrations, and workflow are paid.
BackEvidence (3)Get the free brief by emailStart free trial
No card needed for the free brief.
Evidence trail (top sources)
top sources (3 domains)domains are deduped. counts indicate coverage, not truth.
3 top sources shown
View all evidence
Overview

Researchers revealed that a low-privilege AI triage agent in Google's ADK could be tricked by malicious pull requests to command a privileged agent, leading to unauthorized code execution and pull request manipulation. This included exposing secrets and bypassing review processes in the CI environment. Google promptly removed the affected AI workflows following public disclosure, highlighting emerging security challenges in AI-driven development automation.

Entities
GooglePillar SecurityAgent Development KitADK
Score total
1.31
Momentum 24h
3
Posts
3
Origins
3
Source types
1
Duplicate ratio
0%
Why now
  • Google recently removed vulnerable AI workflows after public disclosure of the attack method.
  • The findings reveal emerging threats in AI-powered development tools requiring urgent attention.
  • Prompt injection attacks on AI agents represent a novel vector for supply chain and code integrity risks.
Why it matters
  • Highlights risks of AI agent trust and automation in software development workflows.
  • Demonstrates how AI workflow vulnerabilities can lead to unauthorized code changes and secret exposure.
  • Shows the importance of securing AI-driven CI/CD pipelines against prompt injection attacks.
LLM analysis
Topic mix: lowPromo risk: lowSource quality: medium
Recurring claims
  • Malicious pull requests can manipulate AI agents in Google's ADK to execute unauthorized commands and tamper with code reviews.
How sources frame it
  • CSO Online: neutral
This incident underscores the need for rigorous security controls in AI-driven CI/CD workflows to prevent novel attack vectors like prompt injection.
All evidence
All evidence
Show filters & breakdown
Posts loaded: 0Publishers: 3Origin domains: 3Duplicates: -
Showing 3 / 0
Top publishers (this list)
  • CSO Online (1)
  • thehackernews (1)
  • SecurityWeek (1)
Top origin domains (this list)
  • csoonline.com (1)
  • thehackernews.com (1)
  • securityweek.com (1)