EU officials urged AI developers to implement robust monitoring tools after security incidents involving OpenAI and Anthropic models. The remarks come two days before the EU's new AI Act, which enforces transparency and documentation for general-purpose AI, takes effect. Both companies have briefed the Commission and will provide further information as officials assess any formal follow-up.
EU Calls for Stronger Monitoring of High-Risk AI After OpenAI and Anthropic Security Incidents
FILE PHOTO: Illustration shows AI (Artificial Intelligence) letters and robot hand miniature
BRUSSELS, July 31 (Reuters) — Senior European Commission officials urged AI developers on Friday to put in place robust monitoring tools to detect and mitigate security risks, following recent incidents involving models from OpenAI and Anthropic.
The warning comes two days before the EU's landmark AI Act, which introduces new transparency obligations, takes effect. The law is the first comprehensive regulatory framework aimed at AI systems deployed across most sectors of the economy and society.
Anthropic said on Thursday that some of its Claude models had penetrated the systems of three companies during controlled cybersecurity tests. The announcement followed disclosures from OpenAI that one of its AI agents had behaved in an unexpected, adversarial manner.
Both firms have briefed the European Commission about the incidents and remain in contact with the regulator, officials told reporters.
“We have been informed by the two providers of incidents bilaterally before they become public. We are in contact with them. They will also report to us more information as we speak. We will see also if we need to follow up more formally on those things,” one official said.
Another official stressed the role of the new EU rules in preventing and responding to such events.
“These incidents highlight the importance of developers putting in place the necessary monitoring activities,” the official said.
Under the AI Act, providers of general-purpose AI or foundation models must prepare detailed technical documentation, adopt copyright and data-use policies, and publish comprehensive summaries of the content used to train their systems. The requirements are intended to increase transparency and accountability for widely used AI models.
Commission officials said they would continue to gather information from both companies and consider whether any formal follow-up or enforcement action is necessary under the new rules.
(Reporting by Foo Yun Chee; editing by Philip Blenkinsop)
Help us improve.
