Anthropic to offer Mythos access to EU
- Anthropic said on June 1 it will give the European Union access to Mythos, its advanced cyber-focused AI model, after months of EU requests. - The key participant is ENISA, the EU cybersecurity agency, which Reuters and Bloomberg said is set to receive Mythos access first. - Next, ENISA’s use of Mythos is expected through Anthropic’s Project Glasswing, as broader EU access discussions continue.
Anthropic said on June 1 it will offer the European Union access to Mythos, its most advanced cybersecurity-focused AI model, after months of requests from EU officials over how the system could be used and controlled. CNBC reported the move on Monday, and Reuters said the European Commission described its talks with Anthropic as “productive” regarding possible future access for EU bodies. The arrangement comes as Anthropic faces separate pressure in the United States, where the Pentagon has labeled the company a supply-chain risk in a dispute over military use of its systems. Bloomberg reported the first EU recipient is expected to be ENISA, the bloc’s cybersecurity agency. ### Which EU body is getting access first? ENISA, the European Union Agency for Cybersecurity, is expected to be the first EU institution to receive access to Mythos, according to Bloomberg and follow-up reports citing EU officials. Reuters reported on June 1 that the European Commission had held several meetings with Anthropic about possible future access for EU bodies. Project Glasswing is the program through which Anthropic has been testing Mythos before any wider release, according to Yahoo’s report of the development. That report said ENISA would join the testing program rather than receive a broad public rollout of the model. ### Why did European officials push for Mythos access? The European Commission and related EU bodies had been pressing Anthropic for access because Mythos is designed for advanced cyber work, including vulnerability discovery, according to CNBC and other reports following the talks. Reuters said the bloc had sought access after concerns that the model’s capabilities could affect European cybersecurity planning if regulators and public agencies were excluded. GovInfoSecurity reported that the EU push also followed moves by rivals to make advanced cyber models available in limited ways to public bodies. That report said OpenAI had offered the Commission access to GPT-5.5-Cyber in mid-May, adding competitive pressure around which frontier models European officials could examine directly. ### What is Anthropic saying about the arrangement? Anthropic told CNBC the EU access plan followed regulatory review and fit within the company’s broader compliance process for high-risk systems. The company has tightly limited Mythos access because of concerns the model could be misused for cyberattacks, according to CNBC’s report. The Next Web and Dark Reading, citing Anthropic and EU officials, said the arrangement was presented as a controlled-access measure rather than a commercial launch. Those reports described the EU access as part of a supervised testing framework for a small number of public-sector users. ### How does this connect to the Pentagon dispute? The U.S. Defense Department designated Anthropic a supply-chain risk in March, Politico and NBC News reported, after a dispute over how the military could use Anthropic’s models. CNN and the BBC reported that the disagreement centered on Pentagon terms Anthropic said would allow use for “any lawful use” under a proposed $200 million arrangement. Dario Amodei, Anthropic’s chief executive, had objected to those terms, according to the BBC and earlier reports on the standoff. CNBC said the EU access decision therefore lands at a time when Anthropic is expanding cooperation with European authorities while facing tighter scrutiny from parts of the U.S. national security establishment. ### Is this a broader opening of Mythos in Europe? Reuters reported only that the Commission and Anthropic had discussed “possible future access” for EU bodies, which indicates the current arrangement is limited. Bloomberg’s report pointed specifically to ENISA rather than a bloc-wide release. June 1 is also the date Anthropic disclosed it had confidentially filed for an initial public offering, according to CNBC, NPR and other outlets. That filing puts added attention on how the company manages access to high-risk models, named regulators, and government customers as ENISA begins work under the new arrangement.