When the U.S. government abruptly pulled Anthropic’s Claude Fable 5 from the market on June 12, enterprises that had bet everything on a single AI model faced a sudden crisis. The export control order left customers with no warning and no timeline for return. Yet new survey data reveals that two-thirds of enterprises had already built a safety net before the blackout hit.
VentureBeat Pulse Research surveyed 145 enterprise organizations during the weeks spanning the Fable 5 outage. The results show a clear pattern: 51% of enterprises run a hybrid strategy that blends closed frontier models with open-weight models deployed on their own infrastructure. Another 16% are moving core workflows entirely off closed APIs. Only one-third remained fully reliant on a single closed ecosystem when the lights went out.
The blackout revealed deeper problems with AI monitoring
The Fable 5 shutdown underscored what happens when a vendor you depend on disappears overnight. But vendor dependency is only the most visible symptom of a broader issue. Most enterprises lack the monitoring infrastructure to know when an AI system in production starts failing. Just 10% of organizations have automated monitoring that would catch a model drifting or malfunctioning in real time. Roughly a quarter would only learn of a failure when end users report it, or have no visibility at all.
The survey also found that 79% of enterprise organizations have already suffered a real financial or operational hit from autonomous agents. The most common cause is shadow AI, where employees run unauthorized agentic workflows on corporate credit cards without oversight. Other failures include infinite loop billing, where recursive workflows rack up thousands in token costs in a single incident, and agents degrading production databases with unthrottled queries.
This gap between deployment speed and governance capability is what researchers call the Control Gap. June’s blackout turned that gap into a live stress test for many companies.
Most enterprises still lack a single owner for AI governance
The biggest barrier to governing AI across platforms is not technical but organizational. 32% of respondents cited the absence of a single owner or accountable team as their top challenge. Vendor opacity followed at 25%, and missing tooling at 16%. A lack of talent ranked dead last at 5%. The skills exist, but the mandate does not. Only 38% say a central team actually governs AI behavior across their platforms today. 21% say ownership is unclear or contested between teams, and 17% say no role holds formal accountability at all.
The fragmentation is made worse by the number of platforms claiming to be the primary AI layer. Fully 85% of enterprises run two or more such platforms across ERP, ITSM, productivity suites, and data systems. 36% describe an open contest between four or more. Just 8% have consolidated to one.
During the blackout, some enterprises showed how effective a hedged strategy can be. Liberty IT, the engineering arm of Liberty Mutual, runs what it calls an AI backbone with roughly 50 independent components for security, governance, and orchestration. Senior director Brian Craig noted that the company could route around the Fable 5 shutdown because its architecture allowed flexibility across multiple models and vendors. Morgan Stanley applies a similar principle with human accountability layered on top of automation for its end of day P&L processes.
The economic pressure to hedge will only increase. Per token inference costs are falling 70 to 80% each year, while agentic workloads consume 100 to 500 times the tokens of earlier LLM tools. Enterprises are already pairing smaller specialized models with semantic routing to avoid burning premium tokens on commodity work. Meanwhile, 73% report little to show for their custom fine tuning investments of the past 18 months.
The bottom line is that replaceability is spreading faster than ownership. Two-thirds of enterprises have adapted on model dependency by hedging. But the internal problems remain. Assigning a single accountable owner costs nothing and requires no vendor. Yet most companies still have not done it. For more on how AI is reshaping critical industries, read our coverage of AI in healthcare.
