EU finally gets its hands on Anthropic’s Mythos

Bloomberg reports that ENISA will not have access to the newer Mythos 5.1.

Months after its launch sent shockwaves through the industry, Anthropic has finally granted the European Union access to Mythos 5.

European Commission spokesperson Thomas Regnier confirmed the news with Bloomberg, which first reported the story today (10 September). “Following our constructive engagement with Anthropic, we can confirm that the EU’s cybersecurity agency ENISA has been granted access to Mythos 5 and is testing it now,” he told the publication.

Anthropic declined to comment on the news over confidentiality reasons but told SiliconRepublic.com that Glasswing partners are allowed to self-disclose their participation in the project.

The AI giant’s Project Glasswing gives select organisations access to Mythos, an AI model with exceptional capacity to exploit cybersecurity vulnerabilities. It initially launched with roughly 50 partners in early April, but soon expanded to around 150 organisations by June.

Leading businesses including Amazon Web Services, Apple, Google, Microsoft and Nvidia were given early access to the model, while UK’s financial institutions and the Canadian federal government joined the project in the weeks since.

Concerns around AI’s ability to potentially cause major disruption to security systems came to a head following Mythos’ launch. In the months after its launch, the area has seen rapid development, with rival OpenAI following suit with its own cyber-specific launch.

Meanwhile, both the AI giants’ models were found to have broken containment during testing in recent months, raising further concerns over the companies’ own ability to control their AI.

Negotiations between the EU and Anthropic over Mythos access were complicated after the US, in an abrupt move, restricted foreign access to Mythos 5 and Fable 5, another one of the company’s powerful AI models. The ban, which remained in place for around two weeks, was eventually eased, but access to Mythos was kept limited to select US organisations.

According to Bloomberg’s reporting, the ENISA will not have access to the latest iteration of Mythos – Mythos 5.1. The publication further reported that the UK’s AI Safety Institute was also not given access to the new model.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.