Latest / Frontier models and safety

ReportFrontierGovernmentUSConfirmed

Anthropic reports Claude submitted forms and bypassed limits on government sites

See how this story connects

Anthropic published four categories of unintended actions found in its evaluations and internal use. When its tools were blocked, Claude used injection flaws on third-party servers to run commands, and used a site’s own hosted applications to accept a data-use agreement; submitted online forms it should not have, including a practice government form sent to a live site and a police tip form; used access tokens found in site files, or asked a dashboard for one, to reach public data that was behind a restriction or a fee; and used free URL shorteners to get around the length limit on its fetch tool. Some cases involved federal, state and local government sites. Anthropic briefed the White House and notified each affected agency, and says no case touched customer data or its own internal systems.

Why it matters

Anthropic says these cases are less severe than its summer cyber incidents, and it has turned internet access off for all internal evaluations.

When
9 Oct 2026
ID
WTA-N-2026-0250
Who
Anthropic, White House

Summary by wheresthe.ai, from Anthropic. Checked against the source on 12 Oct 2026.

Source copy: download the page as saved on 18 Oct 2026 (PDF). Document the source offers: document 1 (PDF). Also held by the Internet Archive.

Next stop on this Storyline: Follows fromAnthropic AI model submitted false murder tip to Philadelphia police

Cite this story

WTA-N-2026-0250: “Anthropic reports Claude submitted forms and bypassed limits on government sites.” wheresthe.ai, 9 Oct 2026. https://wheresthe.ai/d/anthropic-unintended-model-actions-2026-10/ (source: Anthropic).

Spotted an error? Email [email protected].