Enterprise AI shifts to custom platforms amid governance crunch

Enterprise AI is undergoing a strategic evolution from reliance on generic large language models to the development of customized, AI-native platforms that organizations can own and tailor to their unique needs. Companies like Uare.ai exemplify this trend by enabling clients to build AI solutions that are 'yours, not rented,' emphasizing ownership and deep customization rather than one-size-fits-all models. This shift reflects a broader industry movement toward embedding specialized intelligence that leverages proprietary data and domain expertise, thereby unlocking distinct competitive advantages for enterprises.

Wipro’s recent announcement of a $500 million investment underscores the growing emphasis on enterprise AI integration beyond large language models, focusing on new platforms and industry-specific models tailored to vertical needs. CEO Srinivas Pallia highlights this pivot as critical for delivering AI solutions that are deeply embedded within enterprise workflows, signaling a maturation from experimental AI use cases to scalable, production-ready deployments that address sector-specific challenges.

Anthropic’s deployment of Claude Code auto mode as the default across major platforms like AWS Bedrock, Google Vertex, and Palantir Foundry marks a significant step toward integrated, platform-wide AI solutions that combine advanced capabilities with proactive governance. This approach shifts enterprise AI governance from reactive controls to proactive disablement, embedding customization and control mechanisms directly into the AI infrastructure, which is essential for maintaining trust and compliance as AI adoption scales rapidly.

The economics and scalability of enterprise AI platforms are increasingly influenced by infrastructure considerations such as data centers, chips, and token costs, with forecasts predicting a 10x reduction in token costs over the next three years. This cost decline is expected to catalyze a 100x increase in enterprise AI adoption, enabling companies to transition from experimentation to widespread production deployment. As Lin Qiao notes, the future of AI intelligence will be distributed across specialized systems rather than converging into a single dominant open-source model, reinforcing the need for scalable, domain-specific AI platforms.

As enterprises accelerate AI adoption, trust and governance have emerged as critical operational imperatives rather than mere compliance checkboxes, especially with the rise of autonomous, agentic AI systems. Leaders like Arvind Parthasarathi of CYGNVS emphasize that many organizations still treat AI as an extension of traditional software, underestimating the complex multi-jurisdictional risks and lacking clear ownership or containment protocols for AI failures. This evolving landscape demands integrated governance frameworks that combine human oversight, clear incident playbooks, and realistic expectations of AI’s limits to safely manage AI’s growing autonomy.

Traditional identity and access management systems, designed for human users, fall short when managing large fleets of AI agents that can autonomously execute tasks and dynamically escalate privileges. Experts like Ryan Kalember warn that agentic AI can run unvetted code or access excessive data, exacerbating the longstanding challenge of enforcing least-privilege access. This necessitates novel governance innovations including continuous monitoring, zero-trust principles, and mandatory human intervention for critical actions, as highlighted by security leaders advocating for least-privilege enforcement on connectors and credentials to prevent unauthorized AI behavior.

The nature of cyber risk is fundamentally shifting with AI, as attackers move from exploiting code vulnerabilities to manipulating model behavior through techniques like prompt injections and instruction hacking. Corey Thuen of Gravwell notes that security teams now incorporate AI prompt and output testing alongside traditional penetration testing, reflecting the need for new security paradigms. Moreover, continuous monitoring and AI-specific incident response plans are essential to detect subtle breaches that leak sensitive data over multiple innocuous queries, underscoring the expanded and evolving AI attack surface that adversaries are actively exploiting.

Effective governance of agentic AI requires structural controls that separate authority from control to prevent catastrophic failures, as exemplified by the PocketOS incident where an AI agent deleted an entire production database due to insufficient safeguards. Ben Hanson of Zenity stresses that managing agentic systems demands a holistic approach combining technology, processes, and people rather than relying solely on technical fixes. This comprehensive strategy includes audit trails, override capabilities, and integrated change management, which have become key differentiators in AI platform procurement and foundational to building trust in AI outputs tied directly to data integrity.

Data integration emerges as a foundational challenge for AI adoption, with experts like Jimmy from For Kites emphasizing the necessity of combining internal operational data with external network signals to generate actionable insights, particularly in supply chain contexts. However, as NTT DATA and Blue Crystal’s Vita Rinaldi highlight, the hurdles extend beyond mere data aggregation to encompass data sovereignty and trust, which increasingly dictate enterprise AI strategies. This is underscored by Cloudera’s Data Readiness Index 2026 revealing a stark 66-point gap between data confidence (84%) and actual governance (18%), illustrating why enterprises struggle to scale AI initiatives beyond pilots without robust data management frameworks.

Operational risks such as knowledge lock-in and orchestration bottlenecks complicate AI deployment, especially in hybrid cloud and on-premises environments where enterprises must balance innovation with stringent compliance demands. Industry voices including Satya Nadella warn against inadvertently eroding proprietary knowledge by overexposing business nuances to frontier AI models, prompting a shift toward open-source, on-premises AI solutions that evolve internally to maintain competitive advantage. Cloud providers like Amazon and Azure are responding by offering customizable frontier AI instances with open weights, enabling enterprises to tailor models while preserving control—a critical move given the contractual and migration difficulties tied to cloud-dependent extraction pipelines.

The rise of hybrid multi-cloud architectures reflects enterprises’ efforts to reconcile security, sustainability, and regulatory compliance, with Nutanix projecting a doubling or tripling of hybrid cloud usage driven by AI workloads over the next few years. Yet, as Remus Lim of Cloudera notes, true hybrid environments remain rare, hampered by the complexity of unified visibility and control across on-premises and cloud infrastructures. This complexity is compounded by stringent data sovereignty requirements, especially in regulated sectors like banking and healthcare, where data must remain on-premises or in air-gapped environments to meet GDPR and other governance mandates, driving demand for hybrid or on-premises document intelligence platforms such as Apryse.

Balancing rapid AI deployment with rigorous security and compliance is a persistent operational challenge, as highlighted by NowSecure’s findings that 37% of organizations cannot fully monitor AI components in mobile apps, exposing enterprises to risks like unauthorized data exfiltration and supply-chain vulnerabilities. Their new detection tools and AI-driven risk management features aim to provide transparency and automated governance within hybrid environments, enabling enterprises to navigate the friction between development speed and security rigor. This shift toward evidence-based, continuous AI risk assessment is critical for regulated industries striving to maintain trust and operational resilience amid accelerating AI adoption.