The Auto Industry Must Now Manage Accounts That Do Not Belong to Humans

The digital transformation of the automotive industry has entered a new phase. Until now, the industry’s transformation has largely centered on electric vehicles, autonomous driving, connected cars and software-defined vehicles. Now, the focus is shifting toward AI agents that can operate across vehicle development, production, supply chains, finance, insurance, logistics and customer service.

The problem is that AI agents are not simple support tools. They retrieve data, access systems, execute tasks and call other services and APIs. They do not log in like human employees, but they can perform work like human employees. If traditional corporate security systems were designed mainly to manage human accounts, the next challenge is how to manage “non-human identities” such as AI agents, bots, API tokens, service accounts and automation scripts.

Samjong KPMG addressed this issue directly in its June 2026 report, AI Agent Adoption in the Mobility Ecosystem: Agentic AI Gateway. The report argues that existing identity and access management systems are no longer sufficient as real-time data connections expand across vehicles, factories and enterprise systems, and as generative AI and AI agents spread throughout organizations.

The report’s message is clear. In the age of AI agents, companies cannot ask only who accessed a system. They must also ask what accessed the system, with what authority, what data it retrieved and what decisions it executed. Non-human actors are beginning to perform real work inside corporate systems.

The automotive industry stands at the front line of this shift. Cars are no longer just mechanical products. They are defined by software and connected to sensors, cloud systems, over-the-air updates, vehicle data, battery management, charging infrastructure, insurance, finance, maintenance and used-car markets. The journey of a vehicle from design to production, sale and operation involves automakers, parts suppliers, logistics providers, insurers, financial institutions, maintenance networks and software vendors.

In such a complex ecosystem, data constantly moves between organizations and systems. AI agents are expected to automate these connections. They can detect supply chain risks, analyze parts inventory, review contract terms, check vehicle repair histories, respond to customer inquiries and organize insurance review materials.

But as the number of AI agents grows, new risks also emerge. If companies do not clearly know which systems AI agents can access, over-privileged automation accounts may expose internal data. Unused API tokens or service accounts can become attack paths. If an AI agent accesses the wrong data or executes the wrong command, operational errors and security incidents can occur at the same time.

Samjong KPMG describes this as the problem of Non-Human ID. Non-Human ID refers to the identity of digital actors that are not people. AI agents, bots, applications, API keys, service accounts, automated workflows and robotic process automation accounts all fall into this category. They are not registered in human resources systems like employees, and they are not naturally governed by employment events such as resignation or department transfer. Yet they operate inside enterprise systems with real privileges.

Traditional Identity and Access Management, or IAM, was designed mainly around human users. When an employee joins a company, an account is created. Permissions are assigned based on role. When the employee leaves, the account is removed. AI agents and automation accounts are much harder to manage this way. In many cases, companies do not clearly know who created them, what business purpose they serve, what privileges they hold or when they should be retired.

This is where the report introduces its 3C framework: clarity, control and confidence. Clarity means the ability to identify what AI agents exist, where they operate and what permissions they have. Companies must inventory AI agents and automation accounts scattered across the organization and identify their owners, purposes and access rights.

Control means the ability to limit and manage the behavior of AI agents. Not every AI agent should be able to access every dataset. Agents should receive only the minimum permissions required for their task, and companies must define usage boundaries and execution conditions. Rules for granting, changing and revoking privileges must also be clear. In a connected mobility ecosystem, this control must cover not only internal systems but also data exchanges with external partners.

Confidence means the ability to verify and audit the behavior of AI agents. Companies must be able to record which data an AI agent accessed, which APIs it called and which decisions it executed. When a problem occurs, they must be able to trace the cause and distinguish normal behavior from abnormal behavior. Trust is not created by declaration alone. It requires logging, auditing, monitoring, anomaly detection and clear accountability.

To implement these 3Cs, the report argues for a layered and open architecture. This means separating but connecting the AI agent ecosystem, the control layer and the data layer. In the AI agent ecosystem, companies identify and register agents. In the control layer, they apply permissions and policies. In the data layer, they protect sensitive and operational data while tracking access histories.

The concept of an Agentic AI Gateway is especially important. It refers to a gateway that AI agents must pass through when accessing enterprise systems and data. Just as human workers need access cards and security gates, AI agents need digital gates that verify and control their access.

An Agentic AI Gateway verifies the identity of AI agents, checks permissions, controls data access and records execution history. It is not merely a security tool. It is closer to a governance mechanism for operating AI agents. As the number of agents grows, companies will struggle to control each agent individually. A centralized structure is therefore needed to manage identification, authorization, auditing and policy enforcement.

The report illustrates this concept through contract management. Contract management in the automotive sector is complex because it involves parts suppliers, logistics firms, financial institutions, insurers and maintenance networks. Contract terms, pricing, delivery timelines, quality standards, risk clauses and regulatory compliance must all be managed together. AI agents can analyze contract information, identify risky clauses, compare negotiation terms and track changes.

But contract data is sensitive. If the wrong AI agent gains access to contracts or can view external partner information, the consequences could be serious. To use AI agents in contract management, companies must first define which agents can access which contract data. They must then apply least-privilege controls, record access history and be able to block access immediately when needed. The 3C framework is a way to systematize this process.

The discussion extends beyond the automotive industry. Every industry is considering AI agent adoption. Financial firms are using AI agents for investment analysis and risk management. Hospitals are using AI for clinical support and document processing. Manufacturers are introducing automation agents for production planning and quality control. But once AI agents begin handling internal enterprise systems directly, every industry must face the same questions. Under whose responsibility does the agent act? What permissions does it have? Who can trace and stop it when it behaves incorrectly?

The importance of Samjong KPMG’s report lies in shifting the AI adoption debate from productivity to governance. Many companies have expected AI to accelerate work and reduce costs. But when AI agents begin executing real work, control and accountability become as important as efficiency. The more fast-moving agents a company deploys, the faster mistakes can spread.

The report also proposes an implementation roadmap. The first step is to define the organization’s risk appetite and governance goals. Companies must decide how far AI agents will be allowed to act, which tasks will require human approval and which data should remain restricted. This is not only a technology decision. It is a management decision that should involve executives, security teams, legal teams, compliance functions and business units.

The second step is to assess and use existing internal technical capabilities. Many companies already have data management, account management, access control, security monitoring and cloud operations capabilities. The question is whether these capabilities are connected in a way that fits an AI agent environment. Existing IAM, cloud security and data governance systems must be integrated with AI agent management.

The third step is to build an integrated management system for AI agents and Non-Human IDs. Companies must identify automation accounts, API keys, service accounts and AI agents across the organization and register them in a central repository. They need to see who owns each account, what purpose it serves, which systems it accesses and what permissions it holds.

The fourth step is to establish a standard process from account creation to retirement. AI agents also have a lifecycle. They are created, granted permissions, deployed, modified and eventually retired. If this process is not standardized, unnecessary accounts remain active and excessive privileges accumulate. That becomes the seed of future security incidents.

The core idea behind the Agentic AI Gateway is not to block AI. It is to create the foundation for using more AI agents safely. For AI agents to be deployed in real business operations, executives, security teams and business departments must trust them. Without trust, AI remains stuck in experiments and pilots. With trust, AI can become part of organizational operations.

The automotive industry already carries complex supply chains, regulations and safety responsibilities. Adding AI agents can improve efficiency, but it also makes accountability more complicated. Who is responsible if an AI agent automating parts procurement selects the wrong supplier? How should companies verify an AI agent that misses an important contract clause? How should they control an AI agent that accesses vehicle data and collects too much sensitive information? Without answers to these questions, AI agent adoption will struggle to scale.

The report’s answer is clarity, control and confidence. These are both technical principles and management principles. Companies must know where AI agents exist, limit what they can do and verify what they actually did. Only then can AI agents become trusted digital colleagues rather than risky automation.

The mobility ecosystem will continue to adopt more AI agents. AI can intervene across vehicle development, parts procurement, production, logistics, sales, finance, insurance, maintenance and customer management. But as the number of AI agents increases, the number of identities companies must manage will also explode. An era is beginning in which more non-human accounts than human accounts may move through enterprise systems.

This changes the standard of enterprise security. In the past, the key question was, “Who logged in?” In the future, the question will be, “Which agent executed what action with what permission?” The automotive industry’s AI transition can no longer be evaluated only by model performance or automation level. Governance capability over AI agents will become a source of competitiveness.

Samjong KPMG’s message is clear. AI agent adoption is becoming a structural trend, not an optional experiment. But AI agents without control can become a new attack surface rather than a new source of productivity. If mobility companies want to use AI in real operations, they must first build systems to identify, control and trust non-human digital actors.

The starting point of the AI agent era is not a smarter model. It is a safer gateway. The Agentic AI Gateway is the first blueprint for building that gateway.