Open-Weight AI Models Are Catching Up on Performance, but Still Lag Behind on Safety

The performance gap between open-source AI models and proprietary systems built by giants like OpenAI or Google is closing fast. But a new report from research organization SaferAI raises a troubling concern: as these models grow more powerful, safety measures aren't keeping pace.

The Case of GLM-5.2

According to TechCrunch, SaferAI's analysis focused on GLM-5.2, an open-weight model developed by the Chinese company Z.ai. The report concludes that this model is closing in significantly on the performance of the most advanced AI systems currently on the market—those known in the industry as "frontier" models.

The issue, researchers point out, isn't the model's technical capabilities but the absence of adequate safeguards. Unlike Western companies that invest heavily in dedicated safety teams and rigorous testing protocols, GLM-5.2 appears to have been released without comparable mechanisms to guard against misuse.

Why the Difference Between Closed and Open Models Matters

Open-weight models allow anyone to download, modify, and run the AI system locally, without oversight from the company that originally built it. This offers clear benefits for research and innovation, but it also removes the ability to impose restrictions after release—unlike closed models, which are accessible only through controlled APIs.

The SaferAI report thus revives a long-running debate within the AI research community: as the technology becomes more accessible and more powerful, the risk that bad actors will exploit models lacking proper safeguards grows accordingly.

Implications for AI Governance

Experts cited by TechCrunch warn that the rapid advancement of open-weight models like GLM-5.2 could outpace the current ability of regulatory bodies to enforce uniform safety standards. The situation becomes even more complex given that such models originate from different jurisdictions, each with its own legal frameworks and priorities for AI governance.

The Z.ai case illustrates a broader trend: the global race to develop increasingly capable AI systems is moving faster than international consensus on how these technologies should be regulated and secured.