Pillar Security discovered multiple attack paths in Google's Agent Development Kit for Python (90+ million downloads) that could allow public-facing AI agents to trigger more privileged automation, manipulate pull-request reviews, and expose credentials through prompt injection in malicious pull requests. The researchers demonstrated the first documented agent-to-agent exploitation method, where one AI agent could compromise another with higher privileges. Google patched the underlying issue but classified it non-rewardable due to social engineering involvement.
Topics #
Sources #
- Press
[Read article](https://www.csoonline.com/article/4204906/google-adk-flaws-reveal-what-happens-when-ai-agents-trust-the-wrong-message.html) - Press
[Read article](https://www.theregister.com/security/2026/08/03/google-dev-kit-spurs-first-ever-agent-on-agent-violence/5282496)
Go deeper #
This intelligence is sourced automatically from public sources across the web and synthesised by the Prefactor AI pipeline. Stories are reviewed before publication.
source & further reading
getreadyforagents.com — original article
Apple escalates trade secrets lawsuit against OpenAI over departing employees
Cursor publishes cloud agent environment setup guide and releases Mixture-of-Kittens MoE megakernel
EdotEnv launches reinforcement learning environments for teaching LLMs quantitative research