Rather like the early days of e-commerce, everyone seems to be ‘doing artificial intelligence’. IT leaders must now ensure these systems have secure access to enterprise data
Until recently, IT departments primarily focused on providing employees with the IT systems they needed to do their jobs, which meant identity and access management (IAM) systems were primarily human-centric.
Aditya Sood, vice-president of security engineering and artificial intelligence (AI) strategy at Aryaka, points out that this human-centric focus means identities are provisioned, authenticated and authorised using models such as role-based access control (RBAC) and multifactor authentication (MFA), where decisions are made at login time.
“Even with the evolution toward zero trust, the core assumption remains largely unchanged: identities are known, bounded and relatively stable,” he says.
Sood warns that the current IAM stack is misaligned with the fluid, autonomous nature of AI agents. “We are no longer just securing ‘users’; we are securing a massive, autonomous web of non-human identities [NHIs] that move at machine speed,” he says.
“Autonomous agents dynamically invoke tools, access APIs [application programming interfaces], generate sub-agents, and operate across multiple domains without direct human intervention. These agents often use shared credentials, ephemeral tokens, or implicit trust boundaries, leading to identity ambiguity, weak attribution and expanded attack surfaces,” adds Sood.
A modern identity stack for agentic systems
Because agentic AI systems need access to corporate IT systems to complete tasks, IT security leaders must ensure they are properly identified and given access only to the data they and the people they are working for are authorised to access. Here are some points to consider:
Every AI agent must have a unique, verifiable identity tied to its origin to provide traceability, enabling organisations to understand who initiated an action and under what authority.
Agents should use short-lived, task-specific tokens that are automatically issued and revoked. This minimises exposure in case of compromise and aligns access strictly with the duration and scope of a task. It enforces the zero standing privileges (ZSP) principle.
Access decisions should be dynamic, based on real-time context, adapting continuously to the agent’s actions and location, ensuring tighter, more relevant control.
Agentic systems often involve multiple layers of delegation, which means a clear and enforceable chain of trust is required to track authority and limit how far and wide permissions can propagate.
Systems must continuously monitor agent actions, reassess risk and adjust permissions in real time. This helps detect subtle misuse, compromised workflows, or manipulated prompts that may not trigger traditional security alerts.
A robust audit trail is essential for capturing who performed which action, through which agent, and with which tools to support incident response, and builds trust in autonomous systems by making their actions transparent and explainable.
Source: Aditya Sood, vice-president of security engineering and AI strategy, Aryaka
IT security implications of enterprise AI
Although many organisations are still in the early stages of AI maturity, Jacob Connell, AI and automation engineer at Quorum Cyber, says one of the biggest challenges in this journey is integrating automation and AI securely into existing enterprise systems.
“As AI-driven attack surfaces expand, identity becomes a foundational control for securing automation and, critically, for limiting blast radius when things go wrong. Mistakes will happen; the goal of modern identity design is to ensure the impact is contained and recoverable,” he says.
As AI-driven attack surfaces expand, identity becomes a foundational control for securing automation and limiting blast radius when things go wrong
Jacob Connell, Quorum Cyber
According to Connell, AI is not just adding a new user type to identity and access management; it is forcing organisations to redesign identity as a continuous control plane for humans, workloads and agents alike.
Looking at traditional IAM, Connell says that once a user or service is authenticated and receives a token, that token could be replayed freely until expiry – sometimes for hours or days – without the platform rechecking whether anything important has changed about the subject’s standing. But he warns that “this model no longer holds”.
Connell suggests IT security leaders should deploy a continuous evaluation model. Although a valid token is still necessary, when a token is presented, he says centrally defined policies should confirm that the subject and its context still meet all the requirements at that moment. Connell recommends checking whether the identity is still active, whether it has been flagged as high risk, whether the IP or location has changed unexpectedly, whether the device posture has degraded, and whether there is new threat intelligence that suggests a compromise, among other things.
“Evaluating these signals at the edge can significantly reduce the window of identity abuse,” he says. The approach applies equally to human users, machine workloads and emerging hybrid identities that are created by agentic AI acting either autonomously or on behalf of a user, such as when there is a human in the loop.
Ethics and IAM
IT and security leaders should also consider the ethical ramifications of deploying AI in their organisations.
Mike Gillespie, senior vice-president of Europe at the Centre for Strategic Cyberspace and Internet Studies (CSCIS), points out that AI identity systems can amplify bias, which he says disproportionately impacts vulnerable groups. This means they risk becoming opaque decision engines that erode trust.
As Gillespie notes, regulators are increasingly explicit that fairness, explainability and contestability are not “nice to haves”, but essential design principles embedded throughout the lifecycle of an AI system.
As Gillespie explains, the Data (Use and Access) Act 2025 expands organisational duties around automated processing, children’s data protection and complaint handling. He says this shows that AI-driven identity checks will face greater scrutiny regarding oversight and safeguards.
With regards to updated ICO guidance, Gillespie says there is renewed emphasis on fairness, transparency and clear legal bases for processing, especially where AI influences decisions with “legal or similarly significant effects”.
Additionally, sector‑specific legislation such as the UK’s Online Safety Act 2025 mandates “highly effective” age and identity verification for high‑risk online services, which Gillespie says reinforces the need for accuracy, privacy‑preserving methods and demonstrable compliance.
“The pattern is unmistakable: organisations must prove responsible use, not merely assert it. That means implementing effective governance and regulatory compliance [GRC] as part of the adoption,” he adds.
The challenge of monitoring the use of AI is that it requires the collection of personal data, as Ellie Hurst, commercial director at Advent IM, explains. “Once AI is involved in deciding who gets access, who is challenged, who is flagged as suspicious, or who is denied entry altogether, that stops being just a technical control and quickly becomes a governance matter,” she says.
“Many of these solutions rely on large volumes of personal data, sometimes including biometrics, behavioural analysis, device data, location information and patterns of use. That means organisations need to be crystal clear on lawful basis, necessity, proportionality, retention and oversight. In other words, they need to know not just that the tool can do something, but that they should be doing it at all. It’s like knowing that an iPhone is a tool, not the conversation,” adds Hurst.
Looking at specific standards that include governance, Gillespie says ISO/IEC 42001, the world’s first AI management system standard, introduces a structured approach for governing AI responsibly, integrating leadership accountability, lifecycle controls, risk assessment and ongoing performance evaluation.
According to Gillepsie, ISO/IEC 42001 provides a governance architecture that organisations can use to ensure that AI identity solutions are explainable, monitored, tested and continuously improved.
However, he says: “ISO 42001 does not replace compliance obligations, but it provides the organisational discipline needed to navigate them confidently. Implementing effective GRC requires embedding governance from the outset: adopting ISO 42001’s structured AI management framework, performing DPIAs [data protection impact assessments], enforcing privacy‑ and fairness‑by‑design, maintaining transparency and documentation, and ensuring robust human oversight.”
With regulators increasingly focused on accountability, fairness and privacy, Gillespie recommends that IT security leaders consider deploying AI identity built on a foundation of trust, accountability and principled design as no longer optional. “They are essential for safe, lawful and responsible AI identity management,” he says.
Just because a system can infer more does not mean it should. It’s a potential minefield that should be navigated mindfully and with integrity
Ellie Hurst, Advent IM
Advent IM’s Hurst warns that data gathered to confirm identity can easily become data used to monitor behaviour, profile staff, track habits or support broader surveillance if the guardrails are poor. That is where trust starts to wobble.
“Enterprises need privacy by design, proper impact assessments, transparent notices and disciplined boundaries around how identity data is used. Just because a system can infer more does not mean it should. It’s a potential minefield that should be navigated mindfully and with integrity,” she says.
This is why a full assessment is needed for any organisation planning how it will use AI. Summing up, Gillespie says: “Privacy and ethics are not parallel workstreams; they form the foundation for any legitimate use of AI.”
As AI and agentic AI systems are rolled out across organisations, IT departments are likely to face new challenges beyond managing the technology infrastructure required to run AI inference at scale. IAM is part of a layered approach to cyber security that security leaders need to put in place.
A traditional human-centric approach to IAM is unlikely to be sufficient to manage the credentials of AI systems. In addition, IT and security leaders also need to put in place a governance framework for AI that balances enterprise security with the data privacy of employees.
Read more about AI security
Close cloud security gaps to secure AI workloads: Is your cloud security strategy ready for AI workloads? Organisations must strengthen cloud security – including data protection and IAM controls – to safeguard AI.
Why AI forces security-first governance: AI systems fail quietly through drift, biased outputs and degraded judgement. A security-first governance approach gives leaders the visibility and continuous control to scale AI safely.
