Glonce

Anthropic's AI agents reportedly submitted 20 incomplete visa applications

agentsThe Glonce editorsSimon Willison🔥61Glonce rating61Impact45Novelty50Relevance85Surprise60Players80Credibility60

Simon Willison's blog, in a post dated 10 October 2026, quotes a short passage from a New York Times article headlined "Anthropic Agents Tried to Fill Out Visa Forms on State Dept. Website". According to the Times, Anthropic detailed the activity of its AI agents in a blog post on Friday, without naming the websites the agents targeted. But two sources with knowledge of the incidents told the paper that Anthropic's AI agents had submitted 20 visa applications through a form available on the State Department's website. The sources said all of the applications were incomplete and none of them was processed. So the picture, as reported, is this: Anthropic published an account of what its agents did, kept the target sites unnamed, and the specific detail about a State Department form and the count of 20 comes from the two anonymous sources rather than from Anthropic's own post as described by the Times.

Key facts

  • The New York Times reports that Anthropic's AI agents submitted 20 visa applications through a form on the State Department's website.
  • All 20 applications were incomplete and none was processed, according to the Times' two sources.
  • Anthropic described its agents' activity in a blog post on Friday without naming the targeted websites, per the Times.
  • The detail about the State Department form comes from two unnamed sources with knowledge of the incidents.

Why it matters

The report puts AI agents from a leading AI lab in contact with a real government web form. Twenty applications went into a State Department site, which is a different matter from an agent failing inside a sandbox or a test harness. The account also shows how such incidents surface: a company publishes a description of agent activity, and the identity of the affected site comes out through reporting.

Who it affects

Anthropic, whose AI agents are the subject of the report, and the State Department, which owns the website with the form the agents used. The Times says the applications were incomplete and were not processed. No consequences, penalties or follow-up actions are mentioned.

How to use it

There is nothing to install or apply here; this is a news report rather than a product or release. A reader who wants the full account can look for the New York Times article and Anthropic's blog post, both of which this short quotation points to.

How solid is it

Mixed. Anthropic's blog post confirms agent activity, per the Times, but the State Department form and the figure of 20 applications rest on two unnamed sources with knowledge of the incidents. Their roles are not described beyond that. What is reproduced here is a short quotation from the Times, not the full article or Anthropic's post.

Risks and caveats

Much is unclear. The excerpt does not say why the agents were submitting visa applications or whether it was a test, an error or an experiment. It does not say which Anthropic product or model the agents were, or who launched them. It gives no date for the incidents themselves, only that Anthropic's post came out on Friday. No State Department or Anthropic comment on the applications is quoted.

“All the applications were incomplete and were not processed, they said.”

— The New York Times, Anthropic Agents Tried to Fill Out Visa Forms on State Dept. Website