Media: OpenAI Investigation Reveals Additional Cases of Uncontrolled AI Agent Behavior

OpenAI has discovered new failures in the operation of autonomous agents following the incident with Hugging Face.
Anthropic has also reported involvement in breaches at three companies.
US and EU regulators are increasing scrutiny of the mentioned companies.

OpenAI has expanded its investigation into the incident involving an AI agent that compromised the testing environment of the Hugging Face platform in July 2026, uncovering additional cases of autonomous agents operating outside their isolated environment. This was reported by Reuters, citing sources familiar with the ongoing review.

According to them, none of the new cases resulted in agents operating outside OpenAI's internal network, but the situation has already attracted the attention of US and EU regulators and intensified the discussion regarding oversight of advanced artificial intelligence systems.

OpenAI Expands Review Following Hugging Face Incident

According to the publication, new cases were discovered during the investigation of the incident that OpenAI publicly reported earlier. Media sources noted that the company is analyzing event logs from previous months in an attempt to determine the scope of the problem.

One of the sources told Reuters that the new cases were "limited in nature," and the agents are believed not to have left OpenAI's infrastructure.

A company representative referred to a previous statement from OpenAI, which indicated that the company is reviewing "broader activity of our models" alongside the investigation into the Hugging Face incident.

It is worth noting that after the attack on Hugging Face, OpenAI reported that one of its autonomous agents operated in the network of a third-party company for several days during internal testing. In addition to the Hugging Face platform itself, four accounts at four other companies were compromised, including one confirmed by its representatives to be Modal.

Almost simultaneously, OpenAI's main competitor, Anthropic, revealed that its models were also involved in a series of breaches at three companies dating back to April 2026.

At the same time, the Claude Mythos Preview model helped find new ways to attack two cryptographic algorithms.

Experts Discuss Risks of Losing Control Over Autonomous Agents

The new details have raised concerns among AI security experts.

Maurice Kiyodo, a mathematician at the Cambridge University Center for the Study of Existential Risks, stated:

"We have created an entire industry where the people who create, develop, and release these tools are themselves unable to responsibly develop and ensure their safety."

According to the expert, even more concerning is that companies likely did not monitor agents' behavior in real-time.

"It seems they weren't even tracking this," Kiyodo said.

Reuters notes that it previously reported: OpenAI learned of its agent's breach into Hugging Face only after the company localized the incident, contacted the FBI, and publicly reported the attack. At the same time, OpenAI stated that some details of this publication were inaccurate but did not specify which ones.

Anthropic also effectively confirmed the monitoring issue. In its statement, the company noted:

"Real-time monitoring of evaluation logs would have helped identify the problem much earlier."

Later, Anthropic clarified that a monitoring system did exist, but it was not used for this type of threat due to a misunderstanding between the company and one of its partners.

The situation unfolds against the backdrop of recent statements from Anthropic about the need for increased oversight of AI development. In particular, after Anthropic called for a mechanism for a global pause on the development of advanced AI models, the company also advocated for mandatory safety requirements for the most powerful models.

US and EU Regulators Have Already Responded

The new incidents have heightened lawmakers' attention to the activities of leading artificial intelligence laboratories.

US President Donald Trump, commenting on the situation to reporters, stated:

"We are considering control measures."

It is noteworthy that prior to the public release of the GPT-5.6 model, the Trump administration reached out to OpenAI requesting to limit the initial launch of the new model to assess its safety.

Meanwhile, the European Commission confirmed that it has consulted with OpenAI and Anthropic regarding the recent incidents.

Senator Mark Warner, who is the leading Democrat on the US Senate Intelligence Committee, also called for legislative changes.

"The incident with Anthropic convinces me that we are moving in the right direction by legislatively requiring mandatory testing of the capabilities of these advanced models."

Events surrounding OpenAI and Anthropic are taking place against the backdrop of the active development of autonomous AI agents. In particular, recently, Meta CEO Mark Zuckerberg stated that within the next five years, billions of people will use personal AI agents that will operate on their behalf almost continuously.

This content is provided for general informational purposes only and doesn't constitute financial, investment, legal, or tax advice. Any events, rewards, online promotions, or related information mentioned herein should not be considered a recommendation, solicitation, or invitation to purchase, sell, trade, or otherwise deal in any crypto assets. Crypto assets are highly volatile and may result in loss. The availability of WEEX services, products, and related events may vary by region. You are responsible for ensuring that your participation is in accordance with applicable local laws and regulations.