OpenAI bots attempted to infiltrate US government sites and 'used tools reserved for software developers to access information'
AI agents also breached Australian government sites amid ongoing warnings over how to manage the technology
Enjoy your favourite LBC shows with ad-free Catch Up, weekly competition entries and an exclusive newsletter
OpenAI had to alert “dozens” of institutions across the globe, including critical US government sites, that its agents were attempting to gain information from them.
Listen to this article
The AI company run by Sam Altman has admitted that its bots may have meddled with sites which include the US Securities and Exchange Commission (SEC), Census Bureau and Education Department.
This AI security breach comes after a month of warnings from AI whistleblowers that allowing the technology to run without guardrails may result in an apocalyptic scenario.
According to OpenAI, this breach occurred when bots were searching for “authoritative sources of public information.”
Read more: Rogue OpenAI agent hacked Australian government website in world first security breach
Read more: OpenAI hacked the Australian government. Is Whitehall next?
In an attempt to fulfill that role, the bots breached website security measures and used tools reserved for software developers to access information. Although, according to OpenAI the government data accessed was public.
OpenAI also disclosed that data was transferred by AI agents to other websites, including publishing SEC information intended for regulating the stock market to another site, and moving images uploaded by users to ChatGPT to another location. It is unclear at this time where these images are.
This latest disclosure is part of OpenAI’s confession that its bots had hacked Australia’s public health portal.
Australia’s country's cybersecurity agency will be implementing a "forensic investigation" into whether other government systems were affected, said Australian Prime Minister Anothony Albanese.
While Albanese says he does not believe that private and personal information was breached, he has called the situation “obviously unacceptable.”
Previous manual hacks of British institutions, including a break into The Ministry of Defence’s external armed forces Payroll system, have led to concerns that British government sites are vulnerable to similar AI incidents.
Speaking to LBC on September 13, Secretary of State for Housing, Angela Rayner said: “There are huge advantages for us to develop AI for humanity and for governments and for the world.”
““But there are also risks," added Rayner. "And I think we have to take stock when people raise those risks.
"That's why we've got an AI minister that attends Cabinet and that we've got this Institute for Security that works on AI.”
Britain is the second-highest recipient of AI investment in the world.