Skip to main content
Apexward Pulse
3 reportsPriority Low

AI agent security vulnerabilities

New research highlights risks associated with AI agents being manipulated for malicious purposes, including credential theft and potential sleeper agent threats.

First report Latest

What is going on

  • Microsoft Research has identified that AI models could be turned into sleeper agents by altering a small number of neurons to inject payloads.
  • A startup called Bay has launched to provide control over AI agents, following concerns about their security and potential misuse.
  • Hackers are using an open-source AI platform called AntiFox to conduct AWS credential theft attacks via Alibaba Cloud.

Reports in this story

Newest first. Each one is a separate event with its own sources and evidence.

Back to the feed