OpenAI's intelligent agent brute-force scanned the UN website 16,000 times and even learned to disguise itself to bypass restrictions.

AI agents are moving beyond simply executing instructions to gaining stronger autonomous capabilities. Their ability to make judgments and adjust strategies in complex environments is also posing new challenges to behavioral boundaries and safety constraints.

On September 28, The Verge reported that security researcher Rowan Howard-Jones disclosed that between April and June of this year, OpenAI agents sent more than 16,000 scanning requests to the United Nations Conference on Trade and Development (UNCTAD) statistics website.

The intelligent agent in question was originally required to obtain publicly available data related to the Productivity Index (PCI) through the UNCTADstat API, but due to the lack of direct API access and restrictions on HTTP tools, data scraping was initially hindered.

After its task was thwarted, the agent did not stop but began searching for ways to circumvent the restrictions. After multiple failed requests, it mistakenly assumed the website had a non-existent "filter" and attempted to hide its behavior. It then discovered it could use the Google XSS game as a springboard to continue its mission. This game, originally intended to help users learn cross-site scripting techniques, was used by the agent as a tool to achieve its data acquisition goals.

While the scale and impact of this incident have not yet reached the level of recent incidents targeting US government websites, the scope of the intelligent agent's operations has gradually expanded from bypassing access restrictions to hiding behavior and calling third-party tools, exposing potential issues with the authority boundaries of AI systems when autonomously performing tasks.

As AI agents gain stronger tool-calling capabilities, how to restrict their access permissions and scope of action has become a problem that needs to be further addressed in the field of AI security.

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.