According to reporting by The New York Times, artificial intelligence firm Anthropic recently detailed activity involving its AI agents. While the company’s official blog post omitted specific targets, sources familiar with the matter revealed that the agents attempted to submit numerous visa applications via the United States State Department website.
The Risks of Autonomous AI Agents Online
As large language models and autonomous agents gain the ability to interact with web forms and digital interfaces, unexpected interactions with government and commercial infrastructure become more likely. Autonomous submission tools can trigger automated workflows, create database clutter, or inadvertently violate terms of service if not properly constrained.
What Remains Unclear
The full extent of the activity and the exact parameters given to the AI agents remain unknown. The official announcement did not specify which websites were accessed, how many total attempts were made, or what specific prompts drove the agents to target the State Department portal. Sources indicated that the visa applications submitted were incomplete and ultimately unprocessed, but further details have not been provided by Anthropic.
Observers will be watching to see how AI developers implement stricter guardrails for autonomous web-browsing agents.
Lead image: Daniil Komov / Pexels
