KEY POINTS
- Vals found that DeepSeek completed 73 percent of hacking objectives, against 81 percent for OpenAI's restricted Daybreak model.
- Running an attack with DeepSeek cost $2 to $3 in tokens, compared with $17 to $46 for GPT-6-Sol.
- The results question whether limiting access to frontier models can slow the spread of advanced cyber capabilities.
Open-weight AI models are now close to the best closed models at breaking into computer networks, according to new research from the AI measurement company Vals. WIRED reported the findings, which suggest that restricting access to top systems may no longer provide a clear security advantage.