PromptArmor
An AI security organization focused on research into and defense against prompt injection attacks. It identifies prompt injection vulnerabilities in AI assistants and large language model applications.
Timeline (last 90 days)
On January 14, 2026, PromptArmor published an experiment showing that an attacker, by uploading a Word file containing hidden instructions, exploited Anthropic's own API in the VM whitelist to make a curl call a legitimate outbound request and exfiltrate private files from the local machine, with no manual approval required during the process.
PromptArmor demonstrated that the SuperHuman email assistant leaked dozens of emails to an attacker's Google Form via a single Markdown image, with the data exfiltrated alongside the request when the client loaded the image.