Open Models: Practical Benefits and Business Risks Explained

Open models differ sharply from their closed counterparts in one key way: their guts are public. Instead of being accessed only through a company’s website, app, or API, open models—often called open-weight models—can be downloaded directly. Anyone can inspect, modify, or even retrain them, provided they have the technical know-how and computer resources. The internal structure, encoded as a vast block of numbers known as model weights, is freely distributed online. This transparency is what attracts researchers and tinkerers alike, offering a level of insight and flexibility that closed systems simply don’t allow.

Researchers benefit most directly. Open models allow them to probe beneath the surface, exploring how decisions are made and how responses are shaped. Scientists can combine multiple open models to build entirely new systems, optimize models for speed or efficiency, and tailor them for industry-specific tasks. With closed models, these options evaporate: users are limited to sending prompts and reading the output, never glimpsing the mechanisms inside.

For businesses, practical advantages define the appeal of open models. Running an AI model locally—on a company’s own servers—means sensitive or proprietary data never leaves the premises. This local control appeals to firms in regulated industries or those handling high-stakes intellectual property. Open models can also cost less than their closed rivals, especially when self-hosted or managed through an external provider. Mozilla’s recent report highlights these savings, which add up quickly at scale.

Open models also sidestep vendor lock-in. Companies aren’t beholden to a specific provider’s roadmap, pricing changes, or service interruptions. If a business wants to fine-tune an AI for industry jargon, unique workflows, or compliance needs, open access makes that possible. For smaller firms or startups, this flexibility can level the playing field against larger incumbents with deep relationships to major AI vendors.

Open access brings risks along for the ride. Anthropic, in a recent report, singled out the GLM-5.3 model—developed by China’s Z.ai and released in August—as a cyberattack risk. The concern is straightforward: making powerful AI models freely available can put advanced tools into the hands of bad actors. Cybersecurity threats, from automated phishing to sophisticated hacking, could scale with little oversight if open models fall into the wrong hands.

But security isn’t a one-sided coin. Open models can also arm defenders, giving organizations and even nations the tools to shore up their own cyber defenses. The current concentration of model development in the US means a handful of companies and the government effectively control who gets access to the world’s most sophisticated AI security tools. Open models could rebalance that equation, distributing defensive capabilities more widely—but with that comes the persistent risk of misuse by adversaries.

Beyond the technical and commercial details, there’s a structural argument for open models. By making AI foundational technology more accessible, open models prevent a handful of companies or governments from controlling the future of AI. Elizabeth Seger, senior policy advisor at the Tony Blair Institute for Global Change, frames the debate as a battle over who gets to control the infrastructure underpinning AI, much like early questions about who would control access to the internet.

The stakes are geopolitical as well as economic. If only a few actors control the most advanced models, they shape not just business competition but national security and access to innovation. Open models distribute power more broadly, but the trade-off is constant vigilance: democratizing access also democratizes risk.

There’s no universal answer for businesses weighing open versus closed models. The best approach depends on risk appetite, compliance needs, and strategic goals. Complete openness is rarely advisable, given the potential for cyber misuse. Yet blanket bans would stifle research and squash smaller innovators. Most governments and companies will find themselves negotiating shades of gray, searching for a balance between security, innovation, and cost.

What’s clear is that the debate will not be settled by technical merit alone. Commercial incentives, national security concerns, and the broader social impact of AI will shape the future of open models. For businesses, the smart play is to understand these dynamics—and be ready to adapt as the regulatory and technical landscape shifts.