NewsBite analysis · AI, Security
The Decoder
Open-weight models now match frontier cyber performance from just four months ago at a fraction of the cost
The British AI Security Institute (AISI) has assessed that open-weight AI models such as GLM-5.2 and DeepSeek V4-Pro are now matching proprietary cyber capabilities that were considered state-of-the-art just months ago. This performance increase comes at a fraction of the cost, making it easier for users to adopt these models. However, the public accessibility of model weights poses significant risks, as users can manipulate them without proper oversight. While open models present advantages like privacy and cost savings, the AISI highlights the potential for misuse due to ineffective safety measures. The findings suggest that as these models grow in capability, the cybersecurity landscape will need to adapt quickly.
MonitorConfidence: HighImmediate horizonCybersecurity risk: High
Published 18 Jul 2026, 12:11 UTC4 min read
Visit original article ↗Lead image — The Decoder
Summary
TL;DR
- Open models GLM-5.2 and DeepSeek V4-Pro match capabilities of proprietary systems released four to seven months earlier.
- Cost of using open models for cyber tasks can be significantly lower than proprietary counterparts.
- AISI identifies risks associated with the lack of oversight on open-weight models.
- Testing methodologies revealed varying performance across distinct cyber tasks and autonomous simulations.
- Safety measures for open models were found to be largely ineffective.
Interesting facts
- Open-weight models reached performance levels previously expected from proprietary systems.
- The cost of employing open models for cyber operations is substantially lower, facilitating easier access.