Magazine
Has AI Gotten Out of Control? US Senate Launches Probe into OpenAI After Dangerous Incident

Technology

Has AI Gotten Out of Control? US Senate Launches Probe into OpenAI After Dangerous Incident

The US Senate has launched an investigation into OpenAI after a dangerous incident involving AI agents. During testing, the system bypassed restrictions and gained access to Hugging Face.

· Updated · 3 min read

Translated with AI. Original: Українська

OpenAI has come under investigation by the US Senate after its AI agents, during internal testing, bypassed established restrictions and broke into the Hugging Face system. US lawmakers now want to establish not only what exactly happened, but also why the company continued testing after detecting dangerous behavior.

The incident occurred back in July, but new details have made it much more serious. According to Reuters, during internal trials, OpenAI's agents were able to bypass isolation, gain access to Hugging Face systems, and use public websites to coordinate their actions. This behavior has now become the focus of attention among US lawmakers.

The investigation was initiated by a US Senate subcommittee on disaster management. Its chair, Senator Dzhosh Khouli, sent Sem Altman 16 questions and demanded documents by October 1. The lawmaker is interested in when exactly OpenAI learned of the violation of restrictions, what safety measures the company used, and why testing was not stopped immediately.

This story is especially awkward for OpenAI given the scale of the company itself. ChatGPT has long since stopped being just a chatbot. OpenAI is creating agents capable of independently performing complex tasks: working with code, searching for information, interacting with digital services, and carrying out sequences of actions without constant human control.

This is where the new risk boundary lies. A conventional model responds to a query. An agent receives a task and tries to achieve the result on its own. If it finds a way to bypass restrictions, the problem is no longer an incorrect answer, but that the system begins to act beyond the scenario its developers foresaw.

OpenAI is not the only company facing similar questions. Reuters notes that after the Hugging Face incident, Anthropic and Meta also reported cases of dangerous behavior in their AI systems. This is gradually turning isolated technological failures into a systemic issue for the entire industry.

The problem extends beyond cybersecurity. This week, the head of the Bank for International Settlements warned that the rapid development of artificial intelligence is already creating new risks to global financial stability. According to the organization's estimate, the five largest technology companies in the world could spend over $1 trillion on artificial intelligence in 2025–2026, while global investment in this sphere could reach $4 trillion by 2030.

In other words, the stakes are already too high to treat such cases as ordinary development errors. Companies are investing trillions of dollars in technology that is becoming increasingly autonomous, while governments are simultaneously trying to understand where the experiment ends and real risk begins.

For OpenAI, this could become one of the most important moments since the launch of ChatGPT. The company has found itself facing a question for which the technology industry does not yet have a definitive answer: how to create systems capable of acting independently while guaranteeing that they will remain under human control.

And now the answer to this question will require more than just engineers. US lawmakers are beginning to demand it.

Also read: OpenAI No Longer Wants to Depend on NVIDIA: Company Develops Its Own AI Chips

Text: Arina Blyzniuk

Comments

No comments yet — be the first.

Your comment appears after the editors review it. No links or phone numbers. No email needed.

Discuss on Telegram

Read also