analiza aiqbench

The article centers on Anthropic's Claude Mythos, a model whose cyber capabilities and tiered access program are the core subject, so AI is the primary focus. The 129,000 figure comes from Anthropic's own reporting and doesn't establish how many findings were exploitable or how much they reduced real risk. Readers should weigh the access tiers and misuse safeguards, which determine who can use these capabilities, alongside the discovery numbers.