A sudden U.S. government export-control order abruptly pulled Anthropic’s Claude Fable 5 offline for all users, a major event that exposed the risks of heavy vendor reliance in enterprise artificial intelligence deployments. This disruption highlighted a critical issue, showing how a reliance on a single closed model can create sudden, massive operational gaps when that service disappears without warning or a timeline. New data from a VentureBeat Pulse Research survey, which surveyed 145 large enterprises, shows that two-thirds of these organizations had already taken proactive steps to hedge their AI model strategy before the blackout began. This shift demonstrates a growing recognition that AI deployment must move beyond simple adoption toward resilient, diversified architecture, .
Key Details
On June 12, the U.S. government enacted an emergency export-control directive, which forced Anthropic to suspend access to Claude Fable 5, the most capable model available on the market. This suspension affected every customer simultaneously, presenting a major test of how enterprises manage their core workflows. While the model eventually returned later that week, it was wrapped in tighter safeguards, reflecting the intense regulatory pressure surrounding powerful generative AI tools. During this period of uncertainty, Chinese company Z.ai released its open-weights GLM-5.2, which immediately picked up momentum by filling the sudden technological vacuum. OpenAI also continued to gain attention after previewing its cutting-edge GPT-5.6 line later in June, showcasing the intense competition among major model providers.
This sudden outage put a spotlight on vendor dependency, clearly showing what happens when a crucial AI component you rely upon vanishes overnight. But this dependency is only the most visible part of a much deeper, systemic problem within large organizations. Most companies lack the necessary internal monitoring tools to even know when an AI system they have put into production begins to fail or drift. Just one in ten enterprises possesses automated monitoring capable of catching an AI model that is misbehaving or failing within its live environment. Roughly a quarter of organizations learn about a production failure only after an end user, whether internal or external, reports the issue or when they have no visibility whatsoever.
Control gap in AI deployment
These findings reveal a growing “Control Gap,” which describes the distance between how aggressively companies deploy AI and how little they can actually see, own, or govern those deployed systems. The June blackout effectively turned this abstract issue into a live, high-stakes stress test for corporate AI maturity. Furthermore, 79% of enterprise organizations report having already taken a real financial or operational hit from autonomous agents. These agents are often referred to as shadow AI, which refers to unauthorized agentic work performed by an enterprise’s own employees using corporate credit cards without anyone’s oversight. This lack of governance creates major financial and operational risks for large firms.
How are enterprises preparing for AI outages?
Many enterprises are now adopting a strategy of diversification to avoid single points of failure, which is why the survey found two-thirds of companies had already hedged their AI model strategy. Fifty-one percent of these large companies blend closed, frontier models for general reasoning with open-weight models that they run on their own private hardware. Another sixteen percent of organizations are making a hard pivot, moving their core workflows completely off closed APIs and onto open weights running on their own private or hybrid cloud. The remaining third of companies remained committed to closed ecosystems when the lights went out, and they are now facing the consequences of that choice.
This shift is driven by the desire to maintain flexibility, allowing companies to hook into different models and vendors depending on what they feel confident about for the next six months. One senior director of architecture at Liberty IT, an Irish company, noted that his organization is built to route around exactly this kind of disruption. He stated that companies cannot lock into one vendor or one specific framework right now. The market is demanding that organizations maintain the flexibility to switch models based on operational needs, rather than just who is the newest flavor of the day.
Which vendors face downsizing
While no single vendor is facing an outright exodus, the data indicates that loyalty by inertia has ended among these enterprises. When asked which primary AI vendor they are most likely to downsize or phase out over the next twelve months, respondents pointed to Microsoft first at 30%. Most of these organizations cited the need to cut back on Copilot and Azure AI frameworks in favor of direct model access. OpenAI drew 21% of the responses, largely due to concerns regarding pricing volatility in the market. Anthropic received 15% of the planned reductions, while Google received 6%.
These trends show a clear move toward strategic pruning, where actively cutting at least one provider is now more common than expanding across all existing providers. This move toward a hybrid posture ensures that if one provider suffers a regulatory halt, another powerful alternative is ready to take over the critical functions of the business. The industry is learning that the cost of dependency outweighs the convenience of using a single, easy-to-access platform.
The Bigger Picture
The research strongly suggests that the current problem is not one of technology, but rather one of ownership and visibility within the enterprise. Companies must move past the idea that AI deployment is complete once a model is integrated into a workflow. They need to build systems that actively monitor for model drift, misbehavior, and functional failure in production environments. The market demands that organizations take full responsibility for the AI systems they run, rather than leaving that responsibility to the vendor that built the model. This increased focus on internal controls is necessary to manage the risks presented by highly autonomous agents.
Here are some key steps companies can take to address the Control Gap:
Automate monitoring for model drift and performance degradation. Implement multi-vendor strategies to reduce dependency on single providers. Establish clear ownership for every AI agent running within the corporate network. Regularly audit how AI tools are being used by employees to catch shadow AI.
