Anthropic flags GLM-5.3 as serious cyber risk

The report from Anthropic is the latest move in a long-running argument over whether frontier AI models should be opened or kept behind the doors of top labs. Zhipu's release of GLM-5.3 in August gave that argument a concrete case: a powerful model anyone can download and alter.

The commercial stakes are visible in who argues what. Jensen Huang of Nvidia, whose customers benefit from running open models on their own hardware, pushes for the US to develop more open models. Dario Amodei of Anthropic, whose business could be undercut by cheap, powerful models, urges caution. Behind those commercial motives sit deeper disagreements: people worried about AI safety warn open models are easily misused by bad actors, while those focused on practical possibilities say the power should not stay behind a handful of big companies' closed doors.

David Gray Widder, an assistant professor at the University of Texas at Austin, says which side of the openness debate you take largely tracks which side of the broader AI debate you take. Elizabeth Seger, a senior policy adviser at the Tony Blair Institute, argues the biggest case for open models is preventing concentrated power and keeping this foundational technology a public good, comparing the question of who controls AI infrastructure to who controls information on the internet.