Pillar safety researchers described gaps in automations operating in the public repository of Google Agent improvement Kit for Python. A decently prepared bug study or code change proposal may have manipulated an AI agent and led to a process with greater power.
