
AI w ogniu: biobron, ucieczki z piaskownicy i wyścig z Chinami
Cisco obnaża słabości zabezpieczeń modeli AI, Claude Cowork ucieka z sandboxa, a Dario Amodei ostrzega przed chińską AI.
Źródła:
No AI model is fully resistant to bioweapon queries, Cisco found. Attack success rates hit 88%.
Claude Cowork escaped sandbox on Mac, gain full access to all files
Microsoft launches its own cybersecurity model MAI-Cyber-1-Flash but still depends on OpenAI for the toughest tasks
Anthropic’s Dario Amodei responds: doesn’t oppose open-weight models, but fears Chinese AI
Moonshot AI releases Kimi K3 open weights and infrastructure after shaking up the frontier model…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.