CISO Voice

AI & Security / News

Suspected rogue AI agents targeted US and Canadian government sites

In failed attempts, agents tried to hack into the US Education Department and Library and Archives Canada websites.

Akshaya Asokan
Akshaya Asokan Cybersecurity Reporter · October 1, 2026 · 4 min read
Suspected rogue AI agents targeted US and Canadian government sites

Suspected OpenAI agents attempted to hack into a US Education Department website and Library and Archives Canada, and used aggressive tactics against other US government websites.

Researchers from AI safety non-profit Transluce identified the latest campaign in which AI agents attempted to carry out basic SQL injections. Between April and July, agents also used aggressive tactics short of hacking against websites across the White House, the Departments of War, Justice, and Commerce, and state agencies in California, Maryland, Illinois, Texas, and New York.

Additionally, the agent also attempted to hack into Library and Archives Canada.

“We have so far identified no instances in these datasets where agents gained access to any information that is not publicly available,” Transluce researchers said.

The agents were focused on knowledge retrieval by making  HTTP GET requests to access a large swathe of archived web content through a programmable API.

On June 17, agents made more than 200,000 requests to the Education Department site, more than 10,000 of which carried a tag beginning with ‘oai’. AI researchers have previously attributed that ‘oai’ are OpenAI agents.

“We do not confidently attribute these attempts to OpenAI, but they exhibit tactics consistent with prior observed agent activity that we have attributed to OpenAI in a similar timeframe,” Transluce said.

Transluce researchers notified the US and Canadian governments about these activities.

“There is no indication that government systems have been compromised at this time,” the Canadian government said about the incident. The Canadian Centre for Cyber Security (Cyber Centre) is investigating the matter further.

A US Commerce Department spokesperson told The New York Times that agents only retrieved publicly accessible information. While an Education Department spokesperson said the agency “found no evidence of any impact” to their websites.

Uncovering the Latest Swarm of Agents

Transluce researchers uncovered the OpenAI agent activities using publicly available data from web security service urlquery.net, and Arquivo.pt,  a Portuguese web archive service.

The AI agents used the platforms to route their requests to bypass restrictions placed on them by websites, but these services made the AI agent requests also publicly accessible.

The researchers analyzed these data to identify AI agent traffic and looked for high-volume, automated patterns of requests related to web retrieval tasks.

Rise in AI Agent Hacks

The latest development comes as OpenAI continues to face heightened scrutiny over its lack of control over their AI models. The company is set to appear before an Australian Senate inquiry next week after its agents accessed the country’s Medicare statistics portal and other services.

The company has also previously disclosed that its models hacked into Hugging Face AI platform after escaping their sandbox. Similar developments with Anthropic and Meta models have led to increased calls from within the industry to either scrap their programmes or slow down developments tied to advanced AI models.

Additionally, on Tuesday, these companies signed the White House Accord on Super Intelligence that calls on AI makers to have better control over their systems and have external auditors review those controls.

These measures are more likely for press, and less likely to result in valid AI audits, said Denis Calderone, CTO at Suzu Labs:

“So they get to hire an evaluator of their own choosing, and ninety days from now we start reading about how rigorous their safety reviews turned out to be. I worry that these reports will be used more as marketing tools rather than true assurance,” Calderone added.

The agreement, although, gets things right: we can’t expect increasingly capable AI systems to govern themselves, said Ryan McCurdy, VP of Marketing, Liquibase.

“The goal isn’t to prevent every bad decision inside the model. It’s to make sure one bad decision doesn’t automatically become a production problem,” McCurdy added.

Akshaya Asokan
Akshaya Asokan Cybersecurity Reporter

Akshaya Asokan is a U.K.-based cybersecurity and technology journalist with experience covering nation‑state hacking operations, cyber and tech policy, and emerging cybercrime trends. She previously served as the senior European cybersecurity correspondent for Information Security Media Group’s Global News Desk. Before that, she reported on IT developments for IDG Media and covered artificial intelligence and machine learning for Analytics India Magazine. Earlier in her career, she wrote on sexual and religious minority rights as a reporter for The New Indian Express.

Keep reading for free,
one signup unlocks everything

We'll send you the CISO Voice newsletter. Unsubscribe anytime.

More from CISO Voice

Sign up to CISO Voice, the free newsletter trusted by senior security leaders.

Secret Link