**UK gov’s Mythos AI tests help separate cybersecurity threat from hype**
Last week, Anthropic announced it was restricting the initial release of its Mythos Preview model to a limited group of critical industry partners. Now, the UK government’s AI Security Institute (AISI) has published an initial evaluation of the model’s cyberattack capabilities, adding some independent public verification to those earlier reports.
**Takeaways:**
– **Initial Evaluation:** The AISI evaluated Mythos through specially designed Capture the Flag challenges and found it performs similarly to other recent frontier models on individual cybersecurity tasks.
– **Chain Attacks Potential:** Despite this similarity, Mythos could potentially excel by effectively chaining these tasks into multistep series of attacks necessary for fully infiltrating some systems.
– **Progressive Improvement:** Since early 2023, AISI has been testing various AI models in CTF challenges. Recently, Mythos Preview can now complete over 85% of the same low-level “Apprentice” tasks that earlier models struggled with.
Originally published at arstechnica.com. Curated by AI Maestro.
Stay ahead of AI. Get the most important stories delivered to your inbox — no spam, no noise.