On June 12, a U.S. export-control order pulled Anthropic's Claude Fable 5 offline for every customer, with no warning and no timeline. It later returned this week wrapped in tighter safeguards, but the damage for enterprise teams was immediate: their “working right now” plan became “unknown status, no decision point.”

The bigger punchline in the new VentureBeat Pulse Research (surveying 145 enterprises across these last few weeks) is not just that companies panicked. It is that two-thirds say they had already hedged their AI model strategy before the order came down. Specifically, 51% blend closed frontier models with open-weight models deployed on their own infrastructure, and another 16% are moving core workflows off closed APIs entirely. The remaining third was all-in on closed ecosystems when the lights went out.

This is how dependency stops being a theoretical board-slide problem and becomes a production incident. When the model you built around vanishes, your workflow does not merely “degrade.” It can stop. The Fable 5 event also landed in a moment when teams were already recalculating what “AI dependence” costs, not just in money but in operational control. The source notes sticker shock around Fable 5 at $10 per million input tokens and $50 per million output after it launched June 9 to immediate acclaim. Then, three days later, the U.S. government issued an emergency export-control directive barring access by foreign nationals, and Anthropic suspended the model for everyone because it had no way to verify nationality in real time.

Meanwhile, the market rushed into the gap. The source says Z.ai continued to pick up momentum; on Wednesday it released an open-weights GLM-5.2 into the vacuum, and it also released an open agentic coding environment called Zcode. OpenAI, meanwhile, previewed its cutting-edge GPT-5.6 line on June 26. In other words: your “vendor risk” is now also “model churn risk.” Competitors are filling the vacuum fast, which pressures enterprises to move just as fast, even when governance processes are not ready.

But hedging is only half the story. The survey points to a deeper operational blind spot the blackout made obvious: most enterprises lack the monitoring to know whether an AI system is drifting, misbehaving, or failing in production. Just 1 in 10 enterprises has automated monitoring that would catch an AI model drifting, misbehaving, or failing in production. Roughly a quarter would learn of a production failure only when end users report it, or they would lack the visibility to detect it at all.

That monitoring gap matters more as companies ramp agentic work. The source frames this as part of what it calls the “Control Gap,” the distance between how aggressively enterprises are deploying AI and how little of it they can see, own, or govern. And it is not academic. The survey says 79% of enterprise organizations have already taken a real financial or operational hit from autonomous agents, most often “shadow AI,” meaning unauthorized agentic work run by enterprises' own employees on corporate credit cards outside anyone's oversight.

Now zoom out to why this is a governance issue, not just an IT issue. When Liberty IT senior director of architecture Brian Craig described what happened at his company (Liberty IT, the Ireland-based engineering arm of Liberty Mutual, where Craig is Irish and the export order hit him directly), the point was blunt: the team saw Fable arrive and the sticker price immediately, then “luckily enough,” they did not get to “fall in love with it” before it was gone. Liberty IT runs what it calls an AI backbone, roughly 50 components spanning security, governance, observability, and orchestration, each independently replaceable. Craig said you cannot lock in “one vendor and even one framework” and that flexibility needs to be preserved “for the next six months,” because the “flavor of the day” is less important than confidence in near-term continuity.

The survey shows that operational posture is becoming mainstream. A 51% majority of enterprises run a hybrid posture: closed frontier models for general reasoning and open-weight models deployed locally for specialized execution. Another 16% are making a hard pivot, moving core workflows onto open weights running on their own hybrid or private cloud. For the 32% holding a closed commitment, the source says they are candid that self-hosting overhead outweighs savings.

But after June, the calculus changed. The source describes defection as the active posture and shows how quickly “loyalty by inertia” is breaking. Asked which primary AI vendor they are most likely to downsize or phase out over the next 12 months, respondents named Microsoft first at 30%. OpenAI drew 21%, Anthropic at 15%, and Google at 6%. Meanwhile, 28% plan to trim no vendor at all. No vendor faces an exodus, the source stresses. But actively cutting at least one provider is now more common than expanding across all of them.

Finally, the survey connects strategy to execution risk with a simple question: how would an enterprise know if a production model was drifting, behaving unsafely, or failing to complete tasks? The source says 40% are very confident they would detect it, but the confidence breaks down sharply. 30% rely on humans reviewing critical AI outputs, and only 10% (14 of 145 organizations) have automated monitoring and alerting running against production systems. 32% expect to catch most issues eventually, 19% say they would likely hear from end users first, and 8% report no systematic visibility.

So here is the board-level implication: hedging helps you avoid a sudden disappearance. Monitoring helps you avoid silent failure. The Fable 5 blackout was a live stress test that confirmed both what teams feared and what teams prepared for. If you are an executive deciding how to scale AI safely, this is not a question of whether models will change. It is a question of whether your organization can keep working when they do, and whether you can see the moment something goes off-script.