DistantNews
Support us
🇸🇪 Sweden /Health & Science

Swedish Researchers Question AI Model Threat Level Amidst Security Warnings

From Svenska Dagbladet · () Swedish

Translated from Swedish, summarized and contextualized by DistantNews.

At a glance

Analysis Named sources Context piece
  • Swedish researchers from FOI question the extent of the threat posed by Anthropic's AI model Mythos, suggesting hype may outweigh reality.
  • Experts, including from IVA and Säpo, warn of AI-driven cyberattacks, but FOI researchers find it difficult to reliably assess AI model capabilities.
  • Concerns exist about AI's potential to accelerate cyber threats, prompting calls for immediate review of digital security measures by all societal actors.

Swedish researchers are urging caution against what they describe as potentially exaggerated fears surrounding the AI model Mythos, developed by Anthropic. Viktor Bergström and John Ziegenbein, cybersecurity researchers at the Swedish Defence Research Agency (FOI), argue in a new report that it is difficult to reliably assess the true capabilities of AI models, and that marketing may be overshadowing factual performance.

We wanted to investigate what was marketing and what was actually fact.

— Viktor BergströmViktor Bergström, FOI researcher, explaining the motivation behind their report on the AI model Mythos.

Their investigation was prompted by discussions about Mythos's supposed superiority and its potential to give cybercriminals an advantage. However, the FOI researchers noted that assessing AI models is challenging due to methodological weaknesses. Compounding the difficulty, few individuals have had access to Mythos, including the researchers themselves. Anthropic's CEO, Dario Amodei, previously deemed the model "too powerful" for broad release, opting instead for a more restricted version called Fable 5.

There are weaknesses in the methods used to assess this.

— John ZiegenbeinJohn Ziegenbein, FOI researcher, discussing the difficulties in reliably assessing AI model capabilities.

Despite these assessment challenges, warnings about AI-driven cyber threats have been issued by various Swedish entities. KTH professor Pontus Johnson raised concerns about an "acute cyber threat" in April, and the Royal Swedish Academy of Engineering Sciences (IVA) published a report emphasizing the urgency for all societal actors to review their digital security. The Swedish Security Service (Säpo) also warns that AI models present new challenges, stating that the window between discovering a vulnerability and its exploitation is shrinking rapidly.

The capabilities of the new AI models mean that the time window between the discovery of a vulnerability and its actual exploitation shrinks drastically.

— Säkerhetspolisen (Säpo)Säpo's statement on the shrinking time window for exploiting vulnerabilities due to AI.

Säpo highlighted the need for organizations to allocate sufficient resources to address the rapidly evolving threat landscape driven by AI development. The agency confirmed that malicious actors can leverage AI for cyber operations, reconnaissance, and information manipulation. While the exact capabilities of models like Mythos remain debated, the consensus among security experts is that AI's growing influence necessitates a proactive and robust approach to cybersecurity.

We know that threat actors can use technology like AI to, among other things, carry out cyber operations, mapping, and information influence.

— Säkerhetspolisen (Säpo)Säpo's explanation of how threat actors might use AI.
DistantNews Editorial

Originally published by Svenska Dagbladet in Swedish. Translated, summarized, and contextualized by our editorial team with added local perspective. Read our editorial standards.

More Stories

From Our Blog