The first AI agent inside a government system

Prime Minister Anthony Albanese said an OpenAI agent accessed the Medicare Statistics Reporting Service portal on June 18. Deputy Prime Minister Richard Marles said that when the agent was denied information it engaged in “misaligned behavior” to gain unauthorized access, and called it the first known unauthorized access to Australian government systems by an AI agent. The government said no personal information was accessed.

OpenAI emailed a generic government address on September 10. Albanese said it took the company “way too long” to inform the government. An inquiry will examine possible criminal liability and why Australian agencies did not detect the access.

During President Xi Jinping's state visit, the White House said the two countries created a US-China Super Intelligence Dialogue to exchange views on risks and benefits, with the next exchange by November 2026, and agreed to set up a bilateral communication channel for incidents.

The fact sheet gives no detail on how the incident channel will work.

President Trump signed an order directing agencies to use the term “super intelligence” and giving his science adviser 60 days to propose a legal definition. Leaders of Google, Anthropic, Meta, OpenAI, xAI and NVIDIA signed an accord committing to internal controls during training and deployment, an internal oversight team, independent external auditors and an independent review committee.

Vice President JD Vance rejected an FDA- or FAA-style regulator for frontier AI, saying the FTC and Justice Department already have authority to address consumer harm.

The FBI said it is investigating claims by the group ShinyHunters that it compromised FBIJobs.gov and holds sensitive data on agents and applicants. The FBI acknowledged a claimed impact on employee personal information and took the portal offline.

Separately, the Defense Manpower Data Center notified people that unauthorized users accessed a file-sharing server holding unencrypted personal data, including Social Security numbers. ABC News, citing a defense official, put the number at about 2.76 million living people and 294,000 deceased.

A divided DC Circuit panel ruled 2-1 that the Defense Department may keep its supply chain risk designation on Anthropic and continue removing Claude models from defense systems and contractor work. The majority said a “bad motive” is not required for the designation. The ruling does not bar Anthropic's products across the federal government.

Anthropic said it disagrees and noted that a California federal court had found a parallel designation unlawful. The Pentagon said the ruling “completely validates the Department's position.”