Key Takeaways
- The European Union is seeking to deepen talks with the U.S. administration on regulating the most advanced AI models, especially those with cyber capabilities.
- Anthropic’s Mythos model, unveiled in April, has sparked alarm over potential AI‑powered cyberattacks and prompted the White House to resist broader access.
- Although the White House says it is not opposed to sharing Mythos with the EU specifically, it remains generally reluctant to allow the model’s distribution to non‑U.S. governments.
- Anthropic reports swift progress on cyber safeguards and expects to make Mythos‑class models available to all customers in the coming weeks, while warning of a narrow window to patch thousands of vulnerabilities before Chinese AI advances.
EU Seeks Intensified Dialogue with U.S. on Cutting‑Edge AI
The European Union has announced its intention to “intensify” discussions with the U.S. administration concerning the most advanced artificial intelligence models, particularly those endowed with cyber capabilities. A Commission official disclosed to CNBC that the bloc wants to align its approach with Washington’s policies to ensure both innovation and security are addressed. This move reflects growing transatlantic concern that frontier AI systems could be weaponized for cyberattacks, necessitating coordinated oversight before such technologies become widely available.
Anthropic’s Mythos Model Triggers Global Alarm
In April, AI lab Anthropic unveiled its powerful Mythos model, immediately sending governments and businesses into a frenzy. Experts warned that the model’s sophisticated reasoning and code‑generation abilities could be exploited to launch AI‑driven cyberattacks, prompting a surge of scrutiny from regulators worldwide. The announcement underscored the dual‑use nature of cutting‑edge AI: while it promises breakthroughs in science and industry, it also raises the stakes for malicious use in the cyber domain.
White House Pushes Back on Expanded Mythos Access
The Wall Street Journal reported that the White House opposed Anthropic’s plan to grant preview access to Mythos to roughly 70 additional companies and organizations. Administration officials cited security concerns, arguing that broader distribution could increase the risk of the model being misused for offensive cyber operations. This stance highlights the tension between fostering AI innovation and safeguarding national security interests in an era of rapid technological advancement.
Anthropic Advances Toward Wider Release of Mythos‑Class Models
Despite the regulatory pushback, Anthropic announced on Thursday that it is making swift progress on developing the necessary cyber safeguards for Mythos‑class models. The lab expects to bring these advanced systems to all of its customers in the coming weeks, asserting that the safeguards will mitigate potential misuse while enabling broader commercial and research applications. The company’s confidence in its safety measures aims to reassure stakeholders that the model can be deployed responsibly.
Limited Preview Access Through Project Glasswing
Anthropic initially rolled out Mythos to a select group of companies and organizations under an initiative called Project Glasswing. However, the EU, its AI office, and most non‑U.S. government entities have not been granted preview access to the model, with the notable exception of the U.K.’s AI Security Institute. This restricted distribution has fueled frustration among European policymakers who seek to evaluate the model’s capabilities and risks firsthand before any wider deployment.
EU‑U.S. Cooperation on AI Standards Highlighted
During a briefing with CNBC, Commission spokesperson Thomas Regnier emphasized that cybersecurity remains a shared priority between the EU and the United States. He noted that the two sides have agreed to mutually recognize their respective standards in this area, laying the groundwork for coordinated regulatory approaches. Regnier’s comments signal a willingness to harmonize rules governing high‑risk AI systems, particularly those that could impact critical infrastructure and national security.
Technical Discussions Expand Beyond Policy Talks
In parallel with high‑level policy negotiations, the EU is expanding its technical engagements with Anthropic and other model developers who have notified their latest models to the EU’s AI office. These discussions aim to deepen mutual understanding of the models’ architectures, safety mechanisms, and potential vulnerabilities. By fostering direct technical dialogue, the EU hopes to build trust and ensure that any future access arrangements are grounded in thorough risk assessment.
Anthropic Conditions EU Preview Access on U.S. Approval
A source familiar with the EU‑Anthropic conversations, speaking on condition of anonymity, revealed that Anthropic informed the Commission that the EU must first obtain permission from the U.S. administration before receiving preview access to Mythos. The source explained that while the White House is not expressly opposed to sharing the model with the EU, it maintains a broader reluctance to allow the model’s distribution to any non‑U.S. government. This condition underscores the strategic weight the United States places on controlling access to its most advanced AI assets.
White House Nuances Its Stance on Sharing Mythos
Clarifying its position, the White House indicated that it does not object to sharing Mythos specifically with the European Union, but it remains generally wary of enabling access for other non‑U.S. administrations. The administration’s concern centers on preserving a strategic advantage in the global AI race, particularly vis‑à‑vis China, while still pursuing cooperative security frameworks. This nuanced stance reflects an attempt to balance alliance cooperation with competitive vigilance.
Treasury Secretary Emphasizes Innovation‑Safety Calculus
When approached for comment, the White House press office directed CNBC to remarks made by Treasury Secretary Scott Bessent. Bessent stressed that the administration is working “very closely” with AI labs to devise a solution that optimizes the “maximum calculus” between innovation and safety. He reiterated the United States’ self‑identification as the world’s AI leader, with China ranking second, and emphasized the need to maintain that lead through calibrated policies that encourage breakthroughs while mitigating risks.
Cyber Safeguards Deemed Essential Before General Release
Anthropic reiterated that models possessing Mythos’ level of capability require robust cyber safeguards before they can be cleared for general release. The company reported that its teams are making swift progress on developing these protections, aiming to ensure that the models cannot be easily repurposed for malicious cyber operations. This commitment to pre‑emptive security measures is intended to address regulator concerns and facilitate smoother approval processes across jurisdictions.
Amodei Warns of Narrow Window to Patch Vulnerabilities
Earlier this month, Anthropic CEO Dario Amodei warned that there is a six‑ to 12‑month window to patch tens of thousands of software vulnerabilities uncovered by the Mythos model before Chinese AI capabilities close the gap. Amodei’s warning highlights the urgency of addressing security flaws in frontier AI systems, suggesting that delayed action could erode the United States’ technological edge and increase exposure to cyber threats posed by rival nations.
Anthropic Declines Further Comment; CNBC Remains Source
When reached for additional comment, Anthropic declined to elaborate on the ongoing discussions or its internal timelines. The article concludes with a reminder to readers to select CNBC as their preferred source on Google for timely, reliable business news coverage. This note underscores the outlet’s role as a primary conduit for information on the evolving interplay between AI development, cybersecurity, and international policy.