iTechGuides is reader-supported. When you buy through links on our site, we may earn an affiliate commission. As an Amazon Associate I earn from qualifying purchases. Learn more
Companies are reporting more concern about controlling, understanding and governing their AI systems, but current evidence does not show a broad retreat from black-box models—or that such a retreat is happening “faster than ever.” AI use is still expanding. What is changing more clearly is the pressure to see how systems depend on vendors and infrastructure, manage data and accountability, and make models easier to oversee or replace.
What the current numbers do—and do not—show
Several recent company-published surveys point to control and governance challenges. They measure what particular groups of respondents reported, however; none tracks how many companies replaced opaque models with interpretable ones over time.
| Finding | Who was surveyed or measured | What it can support |
|---|---|---|
| 71% said switching their primary AI vendor or model would be difficult; 91% said they did not fully understand their AI dependencies across vendors, models and infrastructure. | IBM Institute for Business Value survey of 1,000 senior executives in 16 countries and 17 industries, June 2026. | Respondents report switching friction and limited dependency visibility. These are self-reports, not counts of companies that switched or tried to switch. |
| 68% said meeting data-residency and sovereignty requirements across geographies was challenging. | The same IBM Institute for Business Value survey, June 2026. | Data location and jurisdictional control are reported concerns; this does not establish that respondents changed models. |
| Two-thirds said they were accountable for AI systems they did not fully control. IBM also reported 10% higher AI return on investment in 2025 among organizations that designed early for workload portability and replaceable models. | IBM Institute for Business Value, 2026. The ROI figure is a reported association. | Responsibility can exceed perceived control. The ROI association does not show that portability caused higher returns. |
| Governance ownership shifted from data and analytics functions (17% to 13%) toward dedicated AI governance roles (14% to 17%) between 2024 and 2025. | Stanford AI Index 2026, citing McKinsey survey data. | Organizations appear to be changing who owns governance; this is not evidence that opaque models were removed. |
| AI adoption was described as accelerating. | OpenAI’s December 2025 report combined de-identified enterprise usage data with a survey of 9,000 workers across almost 100 enterprises. | Adoption growth is compatible with stronger scrutiny; it does not show a shift away from black-box models. |
These results come from different publishers and populations, so their percentages should not be compared as though they came from one survey. IBM, Cisco, OpenAI and Deloitte publish research with commercial interests. Treat their findings as evidence about the populations and methods they describe, not as universal measures of corporate AI use.
Recommended Free Tools
What “black-box AI” means—and what it does not
“Black box” is a plain-language description for an AI system whose behavior or outputs are difficult to understand. It is not a single technical category with a universally agreed threshold. A system can be hard to inspect for different reasons: its model may be complex, its documentation incomplete, its data and infrastructure spread across vendors, or its operation poorly monitored.
#1 Best Overall
- Transparency can mean access to information about a system’s model, data, operation, dependencies and governance. It is broader than an explanation for one output.
- Explainability asks whether people can understand why a system produced an output or behaved in a particular way. An explanation generated after the fact can be useful without exposing the model’s internal computation.
- Interpretability concerns how directly a model’s behavior can be understood from the model itself. A complex model does not become intrinsically interpretable simply because a separate tool supplies a plausible explanation.
- Control includes practical authority over deployment, configuration, data handling, monitoring and replacement. Better explanations alone do not guarantee that an organization has that authority.
Explainability for language models remains an active research problem, especially for high-stakes uses. A 2023 policy review also describes questions of feasibility and usability across jurisdictions. Neither point means every AI system is legally required to reveal its internals; applicable rules depend on the jurisdiction and use case.
Why scrutiny is increasing
Vendor and infrastructure dependencies
An organization may rely on several vendors, models and infrastructure layers to deliver one AI service. If teams cannot see those dependencies or move a workload without substantial redesign, they may have limited practical choice even when alternatives exist. The IBM survey findings indicate that executives report this kind of friction, but they do not establish how often companies have overcome it.
Data location and operating across borders
Organizations operating in multiple geographies may need to know where data is processed and stored, and which providers or infrastructure handle it. A model can be understandable yet still fail to meet a data-location requirement; conversely, keeping data in a required location does not make a model explainable.
Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchWindows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallAccountability without full control
When executives are accountable for systems they say they do not fully control, organizations have a reason to clarify ownership, access, oversight and escalation paths. Cisco’s 2026 study summary frames AI ambition as outpacing readiness and identifies transparency and explainability as governance responsibilities. Its survey covered more than 5,200 privacy-responsible IT, technology and security professionals across 12 markets. That is Cisco’s survey framing, not proof of a model-replacement trend.
Rank #3
Moving from pilots to broader use
Deloitte’s 2026 release describes enterprise AI as shifting toward scaling and broader sanctioned access. Wider deployment makes inventory, review and operating controls more consequential; it does not show that companies are leaving opaque systems. The UK Department for Science, Innovation and Technology has cautioned that business AI-use estimates vary with definitions and with which tasks or roles are counted, so adoption statistics need careful interpretation.
Four separate ways an organization can respond
“Moving away from black-box AI” can refer to different changes. These approaches address different risks and should not be treated as substitutes for one another.
| Approach | Question it addresses | What it does not solve by itself |
|---|---|---|
| Model and workload portability | Can the organization move a workload between models or providers without extensive redesign? | Portability does not make a model interpretable or ensure the organization understands its dependencies. IBM’s findings indicate concern, not that most firms have achieved portability. |
| Governance and observability | Can teams inventory AI systems, track dependencies and assign oversight? | Better ownership and monitoring do not reveal every internal model computation or guarantee vendor independence. |
| Data and jurisdictional controls | Can the organization manage where data is handled and address requirements across the places it operates? | Data control does not itself explain model behavior or make a provider easy to replace. |
| Interpretable models or post-hoc explanations | Can people understand a model’s behavior directly, or receive an explanation about a complex model’s output? | A post-hoc explanation is not the same as intrinsic interpretability, and neither ensures portability, data sovereignty or organizational control. |
The right response depends on the risk. A team concerned about regulatory accountability may prioritize inventory and governance ownership; a team exposed to provider lock-in may examine portability; a high-stakes application may need stronger evidence that its behavior can be understood. One improvement should not be assumed to settle the others.
Is explainable AI replacing black-box AI?
The available evidence does not establish that. It shows AI adoption continuing, reported concerns about dependency and control, and some movement toward dedicated governance ownership. It does not provide a comparable longitudinal measure defining “black-box AI” and counting corporate replacements over time, nor a verified company case study documenting a switch from an opaque model to an interpretable replacement.
So the defensible conclusion is narrower than the headline’s superlative: companies face stronger reasons to make AI systems governable, observable and portable, but a broad or unprecedented exodus from black-box models has not been demonstrated. “More governance” is a supported description of the pressure; “faster than ever” replacement is not.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
