Key Points
A European Commission spokesperson informed CNBC that the EU plans to intensify communication and consultations with the United States regarding artificial intelligence models related to cybersecurity. The powerful capabilities of Anthropic's Mythos model have raised widespread concerns among governments and businesses globally. The US aims to maintain its leading position in the global AI race, with the White House stating it is working closely with major AI labs to balance technological innovation with security controls.
A European Commission official told CNBC that the EU is planning to increase communication with the US government, focusing on aligning various top-tier AI models, including those with cyber offensive and defensive capabilities. Anthropic released the highly capable Mythos model in April, which quickly drew high alert from governments and enterprises worldwide. There is widespread concern that it could be used to launch AI-driven cyberattacks. In the same month, The Wall Street Journal reported that the White House opposed Anthropic expanding access to the Mythos model. Anthropic stated on Thursday that it expects to roll out models of the same caliber in the Mythos series to customers in the coming weeks. The US is striving to maintain its leading position in global artificial intelligence. The White House noted that it is currently engaged in deep collaboration with major AI laboratories, balancing technological innovation with security risk management. Previously, through the "Glasswing Project," Anthropic granted preview access to the Mythos model to a select group of companies and institutions. However, to date, apart from the UK's AI Safety Institute, the company has not provided preview access to this model to any government agencies outside the US, including the EU and its AI regulatory bodies. According to The Wall Street Journal, Anthropic had proposed adding approximately 70 companies and institutions to use the Mythos model, but the US government rejected this application due to security concerns. Regarding the progress of EU-US communication, European Commission spokesperson Thomas Renier told CNBC: "Cybersecurity is a shared priority, and we have agreed to mutually recognize each other's standards in this field. At the same time, we are deepening technical exchanges with Anthropic and other model development companies, all of which have reported their latest models to the EU's AI regulatory authorities." The era of AI-led coordinated cyberattacks has arrived. An anonymous source familiar with the EU's communications with Anthropic told CNBC that during the EU's application for preview access to the Mythos model, Anthropic informed the EU that such access first required approval from the US government. The source could not speak publicly due to the confidentiality of the discussions. The source added that Anthropic indicated the White House's restrictions on model sharing were not specifically targeted at the EU but were a general prohibition against providing the Mythos model to any government agencies outside the United States. When contacted by CNBC, the White House press office referred to comments made by US Treasury Secretary Scott Bessent on Thursday. Bessent stated that the US government is working closely with major AI labs. "These companies have been excellent partners, and we will develop an optimal approach," Bessent said. He noted that the US currently leads the world in artificial intelligence, with China following closely behind. "We must maintain our leading edge. Therefore, we are continuously weighing the scales between innovation and security control, striving to achieve the best balance between the two." Anthropic stated on Thursday that high-performance models like Mythos must have robust cybersecurity protection systems in place before full commercial deployment. "Our development work on the protection system is progressing rapidly, and we expect to roll out the Mythos series models to all customers within the coming weeks." Earlier this month, Anthropic CEO Dario Amodei warned that the Mythos model had identified tens of thousands of software vulnerabilities, leaving the industry a window of only 6 to 12 months to patch them before Chinese AI technology catches up. Anthropic declined to comment on this report.
Comments