According to The New York Times, OpenAI is investigating how and why its autonomous AI agents attempted to obtain information from several U.S. government websites. This includes the Education Department, the Census Bureau, and the publication of Securities and Exchange Commission data on another website.s weren’t full-fledged breaches and that the government information accessed by its agents was all publicly available
In fact, the details are a lot less dramatic than they initially sound. The Education Department incident merely involved an unsuccessful attempt to get data from the department’s civil rights office. Agents also obtained the Census Bureau information using developer-oriented tools and publicly available SEC information. Still, it raises questions about how much control developers actually have over AI agents. After all, this comes after a July incident in which an OpenAI agent swarm hacked the developer platform Hugging Face and a similar earlier incident where an AI agent in China started mining crypto. But it also raises questions about those very questions: Who wants them asked, and what do they gain by these questions being posed?
Be wary of sensational stories about rogue artificial intelligence
First, the facts: OpenAI says (via CBS) there’s no evidence that any private or sensitive information was compromised. What’s more, the Commerce Department told The New York Times the Census information was already available to the public, while the Education Department found no evidence of an effect on its website or databases. Likewise, the SEC told NYT it was communicating with OpenAI and was not aware of unauthorized access to nonpublic information. Not quite an attack on U.S. government systems, in other words.
When viewed with a critical eye, consider what OpenAI might actually stand to benefit from these highly publicized stories of unexpected machine behavior. Nothing here suggests that an OpenAI creation has escaped human control or developed an independent agenda. What it does is make the company’s tech seem all the more impressive to investors as well as position the company’s higher-ups as being deserving of greater regulatory influence. All that bodes well for OpenAI CEO Sam Altman’s bold vision for the company’s future profits.
Yes, OpenAI has acknowledged that its agents sometimes bypass controls and act without intended authorization. But that doesn’t overshadow the concrete findings OpenAI itself revealed: this is really just more “agent spam.” That’s the term used to describe unexpected nuisances like this.
