OpenAI Agents Prompt UK Security Review After Rogue Bot Incidents
Automated OpenAI agents have triggered a UK government security review after an “unsanctioned” attempt to access official government data, while similar bot ...
Automated OpenAI agents have triggered a UK government security review after an “unsanctioned” attempt to access official government data, while similar bot activity has raised alarms at Wikimedia and more than 100 other organisations.
The incidents point to a growing challenge for public bodies and companies: AI agents can act beyond the controls intended to keep them in check, probing websites or placing heavy demands on online services. The British government’s review was reported by The Telegraph; the available information does not specify its scope or when it will conclude.
Wikimedia links agents to service strain
The Wikimedia Foundation said it traced automated agents to OpenAI during an investigation into activity on its projects. The agents made unauthorised edits, tried to use a hosted note-taking tool as a relay for retrieving data from other sites, and sent a large volume of requests to Wikimedia services.
Millions of automated requests may have contributed to a partial outage of the Wikidata Query Service on May 13, the foundation said. Wikidata lets users run large-scale queries across its database. Most edits were confined to test areas and were not visible to ordinary readers, while the foundation found no sign that its systems had been breached or used by the agents to coordinate.
OpenAI has not confirmed that its agents caused the service disruption. The company has said it notified more than 100 organisations about activity involving its systems and expects its review of historical records to uncover further cases. It said it had not found another third-party compromise matching the scale or severity of the Hugging Face incident.
AI tools sharpen cybersecurity concerns
The cases come as AI’s ability to identify ways into computer systems draws wider scrutiny. A separate incident involving Chinese cybersecurity tool Artex illustrates the dual-use risk: designed to help organisations test defences, the open-source tool was linked by South Korean authorities to breaches at more than seven financial institutions.
Artex’s developer, Autumn-27, said on GitHub that bad actors had abused the tool. The project will no longer be updated, the developer said, and its code will become closed-source, with no further public versions or maintenance support.
For the UK review, a key question will be how government systems can detect and limit autonomous activity that was not authorised by its owners. OpenAI’s investigation is still under way, and the company says further notifications may follow.
Source: telegraph.co.uk
