"An AI agent with access to a production database carries the same privileged access risk as a human administrator with equivalent permissions – and it operates at machine speed." — Takanori Nishiyama, Senior Vice President, APAC & Japan Country Manager, Keeper Security
Measured harm: IBM's 2026 findings and APAC data exposures
Shadow AI is not a theoretical risk; it is already a measurable vector in breach investigations. IBM's 2026 Cost of a Data Breach Report found that 43 percent of breached organisations had experienced an incident involving shadow AI, up from 20 percent a year earlier, and that breaches involving shadow AI cost an average of US$5.39 million. The report's figures are echoed by concrete incidents in the Asia‑Pacific region: Panasonic Holdings in Japan disclosed an employee uploaded personal records of 334,000 persons to an unsanctioned generative AI tool; Bee Cheng Hiang in Singapore exposed over 95,000 customer email addresses via an unreviewed AI‑generated bulk mail script; and a contractor in Australia uploaded a spreadsheet to ChatGPT, exposing sensitive personal and health records of 3,000 NSW Reconstruction Authority flood recovery applicants.
How Non‑Human Identities (NHIs) proliferate in enterprises
Modern AI tools create identities and credentials as part of ordinary use. Employees can connect applications, automate workflows and process data within minutes, and every such integration introduces service accounts, API keys, tokens and secrets. These are often created outside traditional IT provisioning, lifecycle management and credential rotation; they can live in browser extensions or configuration files beyond the security perimeter. The result, as the source warns, is an accumulation of unmanaged and unmonitored Non‑Human Identities (NHIs) that often have persistent access and unclear ownership long after the employee who created them has left the organisation.

The cyber insurance questionnaire just landed. Now what?
SOC 2, HIPAA, insurance renewals - someone has to own security strategy. Nubivance provides fractional CISO leadership without the full-time salary.
Get a security leadRegulatory guidance across APAC recognises agentic identities
Policymakers in the region are treating agentic AI as an identity problem. Japan's Ministry of Economy, Trade and Industry (METI) and the Ministry of Internal Affairs and Communications (MIC) published version 1.2 of the AI Guidelines for Business in March, introducing a definition of AI agents. Singapore's Infocomm Media Development Authority (IMDA) launched the Model AI Governance Framework for Agentic AI in January, calling for each agent to carry a traceable identity linked to an accountable human and recommending that organisations restrict agents' tools and data to the minimum necessary. The Cyber Security Agency of Singapore (CSA) finalised an Addendum on Securing Agentic AI in June. In Australia, the National AI Centre's Guidance for AI Adoption asks organisations to decide who is accountable and to maintain human control, while the Office of the Australian Information Commissioner (OAIC) recommends organisations avoid entering personal information into publicly available generative AI tools. These frameworks remain voluntary, but the source notes they set expectations that boards and auditors will apply.
Practical controls security teams are advised to adopt now
The source is clear that blocking AI tools outright is not an effective response: employees who lose access to a sanctioned tool will route around the block, making the created identities harder to see. Instead, organisations should extend identity‑first security controls to every identity AI tools create. Recommended measures include continuous discovery of every sanctioned and unsanctioned AI tool, agent and integration; real‑time monitoring and recording of privileged sessions so AI‑driven workflows can be audited as they happen; and treating NHIs with the same authentication, authorisation and lifecycle discipline as human accounts. Specific controls enumerated include least‑privilege access, automated credential rotation, deprovisioning that follows the human owner or workload, moving toward zero standing privilege (granting elevated access only for the task at hand and revoking it automatically afterwards), and applying zero trust to verify every access request, human or machine.
What this means for technologists, regulators, and affected enterprises
- Technologists and security teams: The source's survey work—Identity Security at Machine Speed—found 53 percent of APAC respondents identify AI‑related NHI management as a top gap, 10 points above the global average, and 47 percent cite a lack of visibility into AI tools employees use. For these teams the immediate priority is discovery and session monitoring to prevent unattended privileges from becoming breach vectors.
- Regulators and policymakers: APAC guidance from METI, MIC, IMDA, CSA, the National AI Centre and the OAIC frames agentic AI as an identity and accountability problem. Regulators will likely use those voluntary frameworks to set expectations for boards and auditors; organisations aligning controls early will face an easier path to compliance with any future binding rules.
- Affected enterprises and procurement leaders: The accumulation of NHIs — service accounts, API keys and tokens — often sits outside procurement and IT governance. Procurement and risk teams will need to treat AI integrations as identity events that trigger lifecycle management, approval, and recertification processes comparable to privileged administrator accounts.
Shadow AI replaces a familiar playbook. Where past shadow IT was a single unauthorised app, today’s agentic integrations create persistent non‑human identities with privileged reach. The practical remedy the source lays out is precise: bring AI agents into identity governance, instrument their sessions, and treat every machine actor as an auditable, revocable identity — not a black box. Failure to do so, the data and guidance cited here suggest, will leave organisations exposed to breaches that are already more frequent and more costly.




