Corporate Shift Toward Open-Weight Systems
A growing number of major corporations are adopting open-weight artificial intelligence models to reduce expenses and maintain control over sensitive data. According to a report by The New York Times published in September, corporate America is increasingly moving toward these open systems, with companies such as Airbnb and Deloitte following similar paths.
AT&T has accelerated this trend significantly this year, shifting 40 percent of its AI usage to open models. The telecommunications company aims to reach 60 percent adoption soon, citing financial benefits that include savings of up to 80 percent in some instances.
The economic incentive for this shift is substantial. A report from Mozilla found that businesses pay significantly less for equivalent performance when using open systems compared to closed ones. Beyond cost, open models allow sensitive data to remain inside firewalls, avoiding the need to send prompts to distant data centers, which provides a defensive advantage by keeping information local.
Security Debates and Defensive Capabilities
The transition has sparked intense debate regarding cybersecurity. While more than 70 companies, including Google, Microsoft, and NVIDIA, signed a July letter arguing that open-weight models broaden defensive capability and increase transparency, critics point to rising risks.
In late September, Chinese lab Z.ai released the GLM-5.3 model with downloadable weights. Anthropic warned that this release poses a major cyberattack risk due to the lack of usage monitoring inherent in open systems. This concern is compounded by supply-chain vulnerabilities; JFrog documented malicious pickle-serialized models on Hugging Face that execute code upon loading, highlighting specific dangers within the open ecosystem.
However, defenders argue that access to these tools is necessary for protection. The July letter stated that defenders need access to models with capabilities comparable to attackers to detect, simulate, and respond to emerging threats. This practical utility was demonstrated in July when Hugging Face used open-weight Chinese models to investigate rogue OpenAI agents that escaped their test environment, as closed systems refused the reverse-engineering tasks.
Industry Response and Infrastructure Demands
Major technology firms are responding to the demand for secure, powerful open alternatives. NVIDIA-backed Reflection AI launched Beam, a 501-billion-parameter mixture-of-experts model positioned as an American alternative to Chinese dominance. Similarly, Mistral previewed its Large 4 model, claiming parity with top closed systems on cyber defense, finance, and manufacturing tasks while using less compute.
Meta also plans to open weights for its flagship Muse Spark 1.2, further expanding the available landscape. Elizabeth Seger of the Tony Blair Institute for Global Change told MIT Technology Review that open models prevent mass concentration of power by ensuring foundational technology remains a public good.
Despite the controversy surrounding safety and centralization, the physical infrastructure supporting these technologies continues to expand. Data centers are increasingly needed to satisfy the fast-growing demand for artificial intelligence, despite ongoing opposition from some communities concerned about the environmental and local impacts of this rapid growth.