A Frost & Sullivan analyst perspective on securing the data foundation for enterprise AI.
Across Asia Pacific, enterprises are racing to deploy AI copilots, agents, and large language models into everyday workflows, often faster than they are reassessing the data those systems will touch. Frost & Sullivan’s own analysis points to two issues unfolding in sequence.
First, enterprises are pushing ahead with AI adoption faster than they are building the governance to support it: as Frost & Sullivan’s analysis of the AI platforms market puts it, “data quality issues are affecting trust in outputs”, and “the real challenge lies in enterprise AI governance, economics, trust, and execution,” not merely access to models.
Second, and regionally, data security is being pushed up the boardroom agenda, with Frost & Sullivan’s Frost Radar: Data Security Posture Management report pointing to mixed data security maturity across Asia Pacific as new regulations in India, Singapore, and Australia raise the stakes.
Taken together, these two issues — AI adoption outpacing governance, and data security climbing the regional agenda — raise three questions worth sitting with before going further with deployment.
1. Do You Know Who — or What — Can Access Your Data?
A copilot or an autonomous agent does not pause to ask whether a file was over-shared, over-retained, or never classified in the first place — it simply reaches whatever its permissions allow. Years of accumulated access sprawl often become visible only once an AI system is given a seat at the table.
Frost & Sullivan’s Frost Radar: Data Security Posture Management report frames the stakes: data security posture management “provides visibility into where sensitive data resides, who has access to it and how it is being used, and whether it is adequately protected and in compliance with regulations.”
Without that visibility, how confident can any organization really be that its AI deployment isn’t handing agents the keys to doors nobody meant to leave open?
2. What Happens When the Breach Leaves No Trace?
Much of enterprise resilience planning still centers on recovery from an encryption event: a ransomware actor locks systems, and the organization restores from backup.
But industry threat trackers suggest this picture may be shifting, with a growing share of extortion attacks skipping encryption entirely and relying instead on quiet data exfiltration. When there is no encryption event to trigger a recovery workflow, the first and most urgent task is no longer restoration — it becomes detection: establishing what data was touched, and assessing how exposed it already is, often well before any question of recovery arises.
If an organization’s detection and backup functions still operate as separate tools answering separate questions, how confident can it really be in assessing what was touched and how exposed it already is, when there’s no encryption event to signal that something happened?
3. Could Small Gaps Be Adding Up to a Big Risk?
Individually, many data gaps seem manageable: a folder with broader access than intended, a dataset retained past its useful life, a regulated record shared externally.
The real exposure emerges when these conditions intersect — what practitioners are starting to call “toxic combinations.” Payroll data accessible to an entire organization, or regulated records exposed to an AI model beyond its intended scope, are rarely the result of one failure — they’re the accumulation of small, individually rational decisions.
Have you ever mapped your data estate against identity, access, and AI usage closely enough to find out how many of these combinations already exist?
Questions Worth Sitting With
None of these questions have easy answers, and that’s rather the point. The visibility problem raised earlier is only one symptom of a wider pattern: Frost & Sullivan’s Global AI Maturity Assessment, 2025 finds that enterprise AI maturity remains low overall, with many organizations further along in rolling out AI tools than in building the data foundation to support them — and the gaps won’t surface evenly: some in an access review, others only after an AI agent acts on data nobody realized it could reach.
As AI adoption continues to outpace governance, and as regulation pushes data security further up the regional agenda, these questions will only grow harder to defer. A follow-up piece will dig into how they are actually being answered in practice.
