Hacker News · AI·· 54 min agoAI score75
Vals study finds open-weight DeepSeek nearly matches OpenAI's gated Daybreak on network hacking tasks
Free Chinese AI Models Are Getting Scary Good at Hacking
AI summary
Vals tested how well AI models could carry out multi-step network intrusion tasks on mock corporate networks and found open-weight models now nearly match the best closed ones. OpenAI's Daybreak, available only to trusted partners, completed 81 percent of assigned tasks on average, while the open-weight DeepSeek completed 73 percent, with the two scoring equally on some tasks. Vals says an attack costs $17 to $46 in tokens for GPT-6-Sol versus $2 to $3 for DeepSeek, and researcher Nikil Ravi says gated access at frontier labs may no longer buy much time before such capabilities spread.
Source:Hacker News · AI · wired.com