BRUSSELS – OpenAI, the company behind ChatGPT, is in talks with the European Commission to give it access to its most advanced network architecture. This has the ability to identify software flaws.
The unexpected move is a boon for the European Union, which has been struggling for weeks to identify cybersecurity risks posed by new AI models. But it’s also a setback for Anthropic, OpenAI’s competitor, which hasn’t yet approved EU access to its Mythos design, which specializes in cybersecurity.
The head of the program at OpenAI, former British Chancellor of the Exchequer George Osborne, wrote to the Commission on Sunday night into Monday to give access to this AI model capable of exploiting programming errors, he told POLITICO. The company has also started “a process of contacting member states”, added George Osborne.
“The idea is (…) to work with them to ensure that the institutions, the public services, the important elements of the European infrastructure (were well protected),” Osborne explained during this exclusive interview.
The move immediately eased concerns in Brussels, where cybersecurity and AI officials had been unable to locate the Anthropic AI model, Mythos, for weeks.
European Union Commission spokesman Thomas Regnier praised the “transparency” of OpenAI, adding that it will help EU officials “monitor the deployment of this model” and “potentially address security concerns.” The commission has already had discussions with OpenAI, and further discussions will follow throughout the week, added Thomas Regnier.
ENISA, the European Commission’s cybersecurity agency based in Greece, also confirmed it had contacted OpenAI.
The OpenAI version is a cybersecurity-focused variant of GPT-5.5, the company’s highest build, which was launched about three weeks ago.
EU officials and government watchdogs could access a more permissive version of the cyber security model – that is, faster to identify security vulnerabilities on request. OpenAI’s flagship model, on the other hand, includes all kinds of security measures to prevent users from turning it into a hacking tool.
OpenAI’s move comes as EU officials are growing impatient with Anthropic, one of OpenAI’s main competitors. Anthropic is blocking the distribution of its Mythos model, with advanced cybersecurity capabilities, to several trusted US-based technology and cybersecurity companies and 40 unidentified organizations.
A group of 30 EU parliamentarians has been pushing the Commission’s cybersecurity agency, Enisa, to gain access to Mythos, and four countries, including Spain, have asked the Commission for more information and better coordination on the model.
According to Thomas Regnier, the Commission has had several meetings with Anthropic, but the discussion “is not yet at the same stage as the solution proposed by OpenAI”.
For George Osborne, AI start-ups like OpenAI should not be “the only ones to decide who is cyber-secure in this world”.
Anthropic declined to comment.
This article has been updated to include comments from the European Commission and ENISA.
This article was first published by POLITICO in English, then edited in French by Tiphaine Saliou.
