Signal › Feed

Dropbox Just Made Its Largest Acquisition Ever — $350 Million for Tigris Data and a New Business Entirely. DBX Agent Cloud Is the AI Infrastructure Bet Drew Houston Has Been Building Toward.

The gap between AI value delivered and enterprise preparedness for the risks that come with it has never been wider. Here's what 3,700 global leaders told Kyndryl about the state of AI in November 2026.


Kyndryl's third annual AI readiness report, published November 10, 2026, surveyed 3,700 senior business leaders across 21 countries and found something that most enterprise AI narratives are not structured to explain: AI is delivering more value than ever — 64% of leaders now report positive ROI, up 10 points from 2025 — and enterprise preparedness for the risks that accompany it is at a three-year low. External risk readiness fell seven points to 24%. The gap between reported value and demonstrated readiness is not a rounding error. It is the defining organizational challenge of enterprise AI at scale.

This is not the finding that gets amplified in vendor earnings calls or at industry conferences. Those venues favor the 64% ROI headline. But the full readiness report is built around the other number: the 24% that tells you how prepared the global enterprise is for the risks it is running. That is the number that determines whether the ROI gains are durable.

The ROI-Readiness Paradox That Defines Enterprise AI in 2026

The intuition that rising ROI and falling readiness cannot coexist in the same system is wrong, and understanding why requires tracing the typical enterprise AI deployment sequence.

The first wave of enterprise AI deployments — which built the ROI track record that now produces 64% positive sentiment — was characterized by bounded scope, narrow data access, and limited system integration. A document summarization tool that runs on a closed file system. A code review agent that operates within a sandboxed development environment. A customer service chatbot that handles a defined set of query types with scripted escalation paths. These deployments deliver ROI because the return is real and the risk is contained. The architectural assumptions they require are modest, the attack surface is small, and the workforce changes required to operate them are incremental.

The second wave — the wave that is currently underway — is structurally different. AI agents are being deployed into core enterprise processes: live customer data, financial systems, HR workflows, supply chain operations. The scope is not bounded in the same way. The data exposure is not limited in the same way. The architectural assumptions are not modest in the same way. And the organizational changes required to operate this wave of AI safely are not incremental — they are transformative.

The result is a gap that grows as deployment scales. Each additional AI system brought into production on the first-wave architecture adds to an accumulating pile of risk exposure that the existing governance, security, and workforce infrastructure was not designed to manage. ROI rises because more use cases are live. Readiness falls because the systems those use cases run on were never designed to carry this load, and the organizations that deployed them never retooled to manage the risk that comes with scale.

This is not a hypothetical future concern. The Gartner finding that 40% of enterprise AI agents were canceled before reaching production is partly a symptom of this dynamic: enterprises that deployed agents without the governance infrastructure to manage them had to cancel deployments when the risk exposure became visible. Kyndryl's report quantifies the structural conditions that produce those cancellations.

What the 7-Point Readiness Decline Actually Means

"External risk readiness" in Kyndryl's framework is not a single metric — it is a composite score that aggregates preparedness across several risk categories: cybersecurity threats, competitive disruption from new market entrants or cross-industry players, regulatory and compliance changes, and infrastructure constraints. A score of 24% means that across all of these categories, only roughly one in four enterprise leaders reports that their organization is prepared to manage the external risks their AI deployments face.

The seven-point decline from last year — moving from roughly 31% to 24% — is alarming not primarily because of the endpoint but because of the direction. Readiness is falling as AI deployment is accelerating. The gap between exposure and preparedness is widening, not narrowing. At the current trajectory, enterprises that continue scaling AI deployments without parallel investment in readiness infrastructure will reach a point where the risk exposure outstrips their ability to respond when something goes wrong.

The practical implication is straightforward: for every enterprise that is in the 64% reporting positive ROI, there is a question about whether that ROI is being generated on a foundation that is becoming progressively more fragile. The enterprise AI governance gap documented in mid-2026, in which a majority of enterprises had moved AI into production without the governance structures to manage it safely, is not a problem that resolves on its own as AI scales. Kyndryl's data suggests it is getting worse.

Security Tops the Risk Register: Why 66% of CIOs Are Rebuilding Architectures

Kyndryl identifies security as the top external risk category undermining AI readiness, and the 66% of CIOs who acknowledge that their current architectures require redesigning to scale AI effectively gives that finding concrete organizational form. These are not CIOs who are unaware of the gap — they are CIOs who understand the problem and are acknowledging, on the record, that the system they are running AI on was not designed to carry this load securely.

What does that architectural redesign requirement look like in practice? The core challenge is that most enterprise architectures were designed around a model in which software systems have defined inputs and outputs, predictable behavior within those parameters, and clear boundaries between internal and external data flows. AI systems — particularly large language models and autonomous agents — do not operate within those boundaries in the same way. They can synthesize information across data sources in ways that were not anticipated by the access controls designed for earlier systems. They can produce outputs that contain information that should not have been accessible to the querying system under the enterprise's data governance policies. They can be manipulated through inputs that look like normal user queries but are actually designed to override their operating constraints.

Redesigning enterprise architecture to manage these risks requires changes at multiple layers: network segmentation that isolates AI systems from data they do not need to access, access control frameworks that account for the non-deterministic behavior of AI systems, logging and monitoring infrastructure that can detect when AI systems are behaving outside expected parameters, and incident response playbooks that have been updated to include AI-specific attack scenarios. The 34% of CIOs who do not yet acknowledge the redesign requirement are either not operating AI at sufficient scale to have encountered these issues, or have encountered them and have not yet translated the experience into an acknowledged architectural priority.

The 9% Pacesetters: Three Traits That Separate the Leaders from the Rest

The Pacesetter definition in Kyndryl's framework is specific to the point of being demanding. Organizations qualify as Pacesetters only if they meet all three of the following criteria simultaneously:

Role redesign around AI. The organization has updated its organizational structure to reflect how work actually flows through AI-augmented processes. This means job descriptions, team structures, and performance metrics have been updated — not just supplemented with AI tools, but restructured around the assumption that AI is handling some categories of work that humans previously handled. The distinction matters because an organization that has handed AI tools to a workforce whose roles, incentives, and workflows were designed for pre-AI work will not realize the same gains as one that has redesigned the work itself.

Formal change management. The organization has implemented a structured program so that employees understand the new operating model: what AI is authorized to do, where human oversight is required, how accountability is allocated when AI systems produce errors, and what escalation paths exist when AI outputs are uncertain or contested. This goes beyond training — it means the workforce has a shared understanding of the new operating environment and has been given the tools to operate effectively within it.

Measured workforce readiness. The organization has verified, through assessment rather than assumption, that its workforce can operate within the AI-augmented environment at the level the deployment requires. Nine percent of organizations have met all three criteria. The 91% who have not are, by definition, running AI at scale on an organizational model that was not designed for it.

The Pacesetter gap is ultimately a resource allocation question. Building the organizational infrastructure to meet all three criteria simultaneously requires deliberate investment in the functions that are most commonly underfunded in technology deployments: change management, organizational design, and workforce development. The enterprise AI failure analysis documented earlier this year consistently identifies the absence of these investments as a primary cause of deployment failures that look, on the surface, like technology problems.

The Competitive Disruption Finding Nobody Is Talking About

One of the most underreported findings in Kyndryl's 2026-27 report is the 10-point decline in concern about competitive disruption from outside the enterprise's own industry — from 33% to 23% of leaders identifying cross-industry disruption as a primary external risk.

On one reading, this is rational recalibration. The competitive disruption narrative of 2024 and early 2025 — in which AI-native players from outside traditional industry boundaries would disintermediate established enterprises across every vertical — has not materialized at the speed that was predicted. The more immediate competitive pressure has come from within industry verticals, as AI has accelerated the advantage of digitally mature incumbents over slower-moving peers rather than producing the cross-industry disruption that was widely anticipated.

On another reading, the decline in concern is complacency that will prove costly. The enterprises that stopped worrying about cross-industry competitive disruption in 2026 are making a bet that the structural changes AI enables in adjacent industries will not reach their market before they have time to respond. That bet may be correct on a two-year horizon, but the enterprises with the lowest cross-industry disruption concern are building the least resilience to it, and the history of technology-driven disruption suggests that the transitions, when they arrive, tend to arrive faster than the incumbents who discounted them expected.

The 10-point decline in concern does not mean the risk has diminished. It means the risk has been deprioritized relative to the nearer-term pressures of internal AI governance, security hardening, and workforce readiness. Whether that prioritization trade-off is appropriate will depend on what the next 24 months produce.

The Enterprise AI Readiness Gap by Industry and Geography

Kyndryl's 2026-27 report covers 21 countries, and while the specific industry-level readiness scores are not broken out in the press release, the aggregate data points provide a framework for estimating where the readiness gap is widest and narrowest.

DimensionReadiness (estimated)Key driver
Security posture24% average across all sectorsAI-specific attack vectors outpacing defensive tooling
Infrastructure architecture34% (CIOs acknowledging redesign need = inverse of readiness)Legacy systems not designed for AI-scale data flows
Workforce change management~9% at Pacesetter levelOrganizational redesign consistently underfunded
Competitive disruption response23% view as primary riskDeclining priority vs. internal governance needs
Regulatory compliance readinessVaries sharply by jurisdictionEU AI Act enforcement timeline creating compliance urgency

Geographically, the pattern that emerges from Kyndryl's earlier research — in which enterprises in regulated financial services and healthcare markets have higher readiness for specific risk categories because regulation has forced investment in compliance infrastructure — likely persists in the 2026-27 edition. The lowest readiness is typically concentrated in enterprises in sectors where AI deployment has accelerated fastest without commensurate regulatory pressure to formalize governance: mid-market software, logistics, and manufacturing.

The cross-geographic pattern is more uniform than the industry pattern. The readiness decline from 2025 to 2026 appears consistent across geographies because the underlying driver — AI deployment scale outrunning governance investment — is a universal dynamic. The enterprises that are bucking this trend are the ones that made governance investment a first-class priority in their AI programs from the beginning rather than treating it as a trailing cost of deployment.

A Five-Step Enterprise AI Readiness Audit for Q4 2026

The Kyndryl data provides a diagnostic framework. What does the response look like? The following five-step audit is designed to close the gap between AI ROI reporting and actual enterprise readiness within a 90-day planning cycle.

1. Architecture review against the AI risk inventory. Map every AI system in production against the enterprise's existing security, data governance, and access control architecture. For each system, identify whether it was deployed on an architecture designed for AI-scale data flows or patched onto a pre-existing system. The 66% of CIOs who acknowledge needing architectural redesign have implicitly completed this step — the ones who have not need to run it before the redesign can begin.

2. Security posture audit with AI-specific scope. Standard enterprise security audits were not designed to test for AI-specific attack vectors: prompt injection, model extraction, training data poisoning, and jailbreaking techniques that exploit model behavior rather than infrastructure vulnerabilities. Run a targeted AI security review that covers these categories explicitly, and update the incident response playbook to include AI-specific scenarios.

3. Workforce change management assessment by function. Measure workforce readiness at the team or function level rather than the enterprise level. Identify which functions are operating AI tools without the organizational infrastructure to manage them safely — no documented escalation paths, no updated job descriptions, no accountability framework for AI-generated outputs. These are the highest-risk deployments regardless of the ROI they are producing.

4. External risk ownership mapping. Assign explicit executive ownership to each of the four external risk categories: security, competitive disruption, regulatory compliance, and infrastructure constraints. The organizations with the lowest readiness scores are almost always the ones where external risk ownership is diffuse — where it is everyone's problem and therefore no one's priority.

5. Pacesetter gap analysis. Score the organization against the three Pacesetter criteria — role redesign, change management implementation, measured workforce readiness — and identify which criterion is furthest from current state. Treat the gap in that criterion as the primary organizational input to 2027 AI strategy planning. The enterprise ROI measurement gap documented in Kyndryl's data will not close until the organizational infrastructure that enables reliable ROI measurement is in place.

The five steps are not a remediation program — they are a diagnostic sequence. The output is a clear picture of where the enterprise sits on the readiness continuum and which investments will close the gap most efficiently. For the 76% of enterprises that are not currently prepared for external AI risks, the Q4 2026 planning cycle is the highest-leverage moment to change that.

Takeaway: Kyndryl's third annual readiness report confirms what the most rigorous enterprise AI practitioners have been observing in production for the past 18 months: AI is delivering ROI, and enterprise preparedness for the risks that accompany that ROI is not keeping pace. The 64%/24% split — 64% of leaders reporting positive returns, 24% of enterprises ready for external risks — is not a temporary gap that closes automatically as AI matures. It is a structural condition produced by deployment sequences that prioritized speed over governance, and it widens as deployment accelerates without parallel investment in readiness infrastructure. The 9% Pacesetters have demonstrated that the gap can be closed: role redesign, change management, and measured workforce readiness are not theoretical aspirations but operational practices that separate the enterprises building durable AI advantage from the ones accumulating risk they cannot yet see.

Frequently Asked Questions

What did Kyndryl's 2026 AI readiness report find?

Kyndryl's third annual readiness report, published November 10, 2026, surveyed 3,700 senior business leaders across 21 countries. The headline finding is a tension between two diverging trends: 64% of enterprise leaders now report positive returns on AI investments, up 10 percentage points from 2025, suggesting that AI is delivering measurable business value at scale. At the same time, enterprise readiness for external risks — the threats and disruptions that arrive from outside the organization — fell seven percentage points to just 24%. The report identifies security as the top external risk category, finds that 66% of CIOs believe their current architectures require redesigning to scale AI effectively, and observes that only 9% of organizations qualify as Pacesetters: the cohort that has fully realigned roles, implemented change management, and built genuine workforce readiness to operate AI at enterprise scale.

Why is enterprise AI readiness falling even as ROI improves?

The divergence between rising ROI and falling readiness is not contradictory — it is a predictable consequence of the deployment sequence most enterprises follow. Early AI deployments are targeted at high-value, low-risk use cases: automating repetitive workflows, summarizing documents, accelerating code review. These deployments deliver measurable ROI because the scope is bounded, the data exposure is limited, and the attack surface is small. But as enterprises scale AI — moving from isolated pilots into core processes, integrating agents with live enterprise data, and deploying AI across customer-facing systems — the risk profile changes dramatically. More deployments mean more data flowing through AI systems, more third-party model dependencies, and more architectural assumptions that were never stress-tested at scale. The architectures that worked for the first wave of AI deployments — often patched onto existing systems rather than redesigned around AI — accumulate technical debt and security exposure faster than governance teams can address. ROI improves because more use cases are live. Readiness falls because the infrastructure those use cases run on was never designed to carry this load.

What is a Pacesetter organization in Kyndryl's readiness framework?

Kyndryl defines Pacesetters as the 9% of organizations that have simultaneously met three criteria. First, they have redesigned roles around AI rather than simply adding AI tools to existing job descriptions — meaning the organizational structure has been updated to reflect how work actually flows through AI-augmented processes. Second, they have implemented formal change management programs so that the workforce understands the new operating model: how decisions are made, what AI is authorized to do, where human oversight is required, and how the accountability structure has changed. Third, they have built measurable workforce readiness — not just provided AI training, but verified that employees can operate effectively within the new AI-augmented environment. The 9% Pacesetter figure is significant because it establishes how few organizations have treated AI transformation as a systematic organizational change rather than a technology procurement decision. The 91% who have not yet reached Pacesetter status are, by definition, running AI at scale on a workforce and organizational model that was not designed for it.

How should enterprises close the AI readiness gap in 2026?

Closing the gap between AI ROI and AI readiness requires treating readiness as an engineering problem rather than a training problem. The five highest-leverage actions for Q4 2026 are: first, conducting a full AI architecture review to identify where existing systems were patched to accommodate AI rather than designed for it, and prioritizing redesign of the highest-risk segments; second, running a formal AI security audit that maps every model dependency, every data flow through AI systems, and every access control assumption — the 66% of CIOs who say their architectures need redesigning should treat this audit as the prerequisite to any further deployment; third, mapping workforce readiness at the team level rather than the enterprise level, because readiness gaps are almost always concentrated in specific functions or geographies; fourth, assigning explicit ownership of external risk categories — security, competitive disruption, regulatory change — to named executives who are accountable for readiness metrics; and fifth, benchmarking against the Pacesetter criteria to identify which of the three defining characteristics is furthest from current state and treating that gap as the primary organizational priority for 2027 planning.

What are the biggest external risks to enterprise AI in 2026?

Kyndryl's 2026-27 readiness report identifies security as the top external risk undermining enterprise AI readiness, a finding that tracks with the surge in AI-specific attacks documented across the industry over the past 18 months. Beyond security, the report identifies three other major external risk categories. Competitive disruption from outside the enterprise's own industry registered a 10-point decline in concern — from 33% to 23% of leaders viewing it as a primary risk — but the report's context suggests this reflects a narrowing of attention to nearer-term threats rather than a genuine reduction in the disruption risk itself. Regulatory and compliance uncertainty is a growing readiness burden, particularly for enterprises operating across multiple jurisdictions as AI-specific regulatory frameworks move from guidance to enforcement. Infrastructure constraints — specifically the gap between the architectures enterprises currently operate and the architectures required to scale AI reliably and securely — are the fourth major external risk category, and the one that the 66% of CIOs who acknowledge needing architectural redesign have identified as most requiring immediate investment.