"For the past five years, much of the enterprise conversation around artificial intelligence (AI) has revolved around access - with access to application programming interfaces (APIs) from hyperscalers, pre-trained models, and plug-and-play integrations promising productivity gains. This phase made sense. Leaders wanted to move quickly, experimenting with AI without the cost of building models from scratch. " AI-as-a-service " lowered barriers and accelerated adoption."
"Today's AI stack often looks like a patchwork of third-party models. Marketing leans on generative copy tools. Developers use GitHub Copilot. Analysts query ChatGPT-like assistants. This has enabled rapid experimentation but exposes three structural weaknesses. The first is intellectual property (IP) risk. Outputs generated by external models can be legally ambiguous, which is a red flag in IP-intensive industries. Second, feeding proprietary data into external models creates security and regulatory concerns."
Enterprises initially adopted AI through third-party access—APIs, pre-trained models, and plug-and-play tools—to accelerate experimentation without building models from scratch. That approach reduced barriers but created structural weaknesses: intellectual property ambiguity, security and regulatory exposure when proprietary data is shared, and governance gaps lacking explainability and auditability. These issues are manageable for peripheral AI but become critical as AI integrates into customer interactions, product design, and supply chains. Model ownership means treating models as core assets to be trained, customized, governed, and controlled according to enterprise priorities, using bespoke training on proprietary datasets and appropriate infrastructure and governance.
Read at ComputerWeekly.com
Unable to calculate read time
Collection
[
|
...
]