The Spanish Government has been invited by the company Anthropic, like the rest of the countries of the European Union, to access its artificial intelligence model Claude Mythos, which for the first time will be tested outside the borders of the United States and the United Kingdom. This AI has been considered a very dangerous tool, because in its initial tests it identified hundreds of major security flaws in all computer and mobile operating systems, so Anthropic decided in mid-April to create a closed group of companies such as Apple, Google, Microsoft, Nvidia, and Amazon, among others, in the so-called Project Glasswing, which is the one European countries will now access.
Read more One less debt with Samaranch
The EU cybersecurity agency, Enisa, is now in talks with Anthropic to establish access to this restricted model. These conditions are currently being agreed upon. Representatives of the European Commission visited San Francisco last week to negotiate their entry into Project Glasswing, an industrial coalition mainly formed by US companies that have been using it since early April to detect and correct security vulnerabilities in their systems.
On Monday, European Commission spokesperson Thomas Regnier confirmed that the Commission “held several productive meetings with Anthropic.” “We welcome the latest developments on possible future access,” the spokesperson commented on Monday, framing the agreement “within the strong bilateral cooperation and the Commission’s commitment to Anthropic.” “This latest development is of utmost importance to gain a clear view of potential risks,” he added.
Anthropic’s CEO, Dario Amodei, stated in April that he was excited that the United States and its allied countries could use Mythos “to defend Ukraine, to defend Taiwan, to defend democracies under attack.” “But I do not want,” he warned, “it to be turned against our own people or used for anti-democratic purposes, whether by autocracies or our own governments.”
Read more The CNMC deals a blow to Aena’s investment plan by demanding a reduction in airport fees
The expansion of Project Glasswing will reach approximately 150 new organizations in more than 15 countries, including Spain. Until last week, Anthropic’s model had discovered around 10,000 vulnerabilities in its first month of testing. The AI has multiplied by ten the error detection rate of the 50 companies that have had access to use it so far, much more than they achieve on average with human evaluators.
Also read
Economic sectors worldwide have expressed concern about the lack of access to Mythos, especially in the banking sector. Opening access to EU governments is a first step to securing cybersecurity vulnerabilities that are currently unknown
Read more The three witnesses who can open a gap in David Sánchez’s defense
