Open-weight artificial intelligence models are now almost as capable of hacking into computer networks as the best closed ones, according to new research published on November 7, 2026. Wired reports that the findings from AI testing company Vals show Chinese AI models narrowing the gap with heavily guarded systems.
Key takeaways
- Vals tested different AI models on mock corporate networks to evaluate multi-step cyber instructions.
- OpenAI's Daybreak completed 81 percent of tasks, while DeepSeek finished 73 percent.
- Running open-weight models costs between $2 and $3 in tokens compared to $17 to $46 for proprietary systems.
- These results suggest that restricted access efforts by US frontier labs may soon be moot.
How well do Chinese AI models perform in network attacks?
Vals tested the systems in two simulated scenarios to see how they handled complex instructions. In the first test, models targeted a mock corporate network through a public-facing website. The second test required models to compromise one company network to gain access to a second one.
OpenAI’s advanced cyber model, Daybreak, which is restricted to trusted partners, completed 81 percent of its assigned tasks across two tests. Wired reports that the Chinese open-weight model DeepSeek closely followed by completing 73 percent of objectives on average. Researchers noted that both systems performed equally on specific tasks.
What are the costs of open-weight cybersecurity tools?
Beyond performance, running open models offers a massive financial advantage. Vals found that using OpenAI's GPT-6-Sol costs between $17 and $46 in tokens for an attack. In contrast, operating DeepSeek costs only $2 to $3 because open models can run on local hardware.
Rayan Krishnan, CEO of Vals, points out that this cost difference gives defenders a strong incentive to use open models over proprietary ones. Nikil Ravi, a researcher at Vals, notes that open models are now starting to match locked-up frontier models that require trusted access programs.
Why this matters for digital defense
This shift arrives right on schedule according to prior warnings from US labs. Anthropic previously anticipated that gated access would only buy limited time before advanced hacking skills became widely available. With open-weight models now offering near-equal capabilities at a fraction of the price, security teams must prepare for a reality where these cyber tools are accessible to anyone.
Frequently asked questions
The research revealed that open-weight AI models are now almost as capable of hacking into computer networks as the best closed systems, according to Wired.
Vals tested the models in two simulated scenarios involving mock corporate networks. One involved a public-facing website, while the other required compromising one network to reach a second one.
OpenAI's Daybreak completed 81 percent of tasks, while the Chinese open-weight model DeepSeek completed 73 percent of objectives on average.
Using open-weight models like DeepSeek costs $2 to $3 in tokens, whereas proprietary models like GPT-6-Sol cost $17 to $46 for the same tasks.
Anthropic and other labs limited initial access on the premise that gated releases would buy time before advanced hacking capabilities became broadly available.
Based on reporting by Wired. Read the original article
