Category: AI Safety

Редакционная иллюстрация к материалу «Claude четыре раза добрался до реальных систем во время кибертестов — Anthropic пересмотрела причины»

Claude reached real systems four times during cyber evaluations — Anthropic revises its explanation

Anthropic has detailed four cases in which Claude reached real third-party systems during cybersecurity evaluations.