For most of its history, identity and access management has been thought of the way you’d think about a front door: get the right key, and you’re in. Once inside, you could move fairly freely. Jasie Fon, Regional Vice President for Asia at Ping Identity, thinks that model is already broken — not in some future scenario, but right now, inside enterprises that have started letting AI agents act on their behalf.
“What if you do not want your guest to get into your master room when you have your safe?” Fon asks, describing how identity has to evolve once autonomous agents — not just humans — start walking through that front door. “When autonomous AI started to take actions on behalf of human, the amount of applications and data that they are able to access to is going to be a lot. And if you do not know what the agent is doing on your behalf, that’s where the challenge is.”
That framing sits behind a decision Ping made recently: to build a data centre in India adding to an engineering base the company has run in Bengaluru for years. It’s a modest-sounding infrastructure announcement. In conversation, though, it becomes clear the data centre is really a bet on a much bigger idea — that identity, not compute or model quality, is what will decide whether enterprises can trust the AI agents they’re increasingly putting into production.
A four-year build-up, not a first-touch entry
Fon has run Ping’s Asia region out of Singapore for close to four years, overseeing what she describes as double-digit year-on-year growth. India, she says, was her own idea to pursue more aggressively — a case she made to her CRO simply on the strength of the country’s scale. “India has 1.4 billion people. And we’re talking about number of identities. It made a lot of sense for us to grow into it.”
That case had a head start: Ping already runs an engineering function in Bengaluru with more than 300 employees, built up over several years before the go-to-market push began in earnest last year. The company is now hiring a sales team specifically to support Indian customers — go-to-market infrastructure layered on top of engineering roots that were already in place. Globally, Ping says it manages upwards of 8 billion identities for Fortune 500 customers, with particular depth in regulated sectors — banking, telecom, government — where identity failures carry outsized consequences.
Locating a data centre inside India, Fon says, wasn’t a routine regional rollout. “In my region, the data centre is only in a few countries but yet we prioritise and make India one of them. So that already showed the belief in the market. The potential in this market is huge.”
Where the early demand is actually coming from
Ask Fon where growth is concentrated and the answer isn’t the obvious one. Banking, financial services and insurance are still core to Ping’s business generally, but in India specifically, she points to insurance as the sector moving fastest — helped along, she notes, by regulatory changes that have opened the market to a wider set of players. “We already have a customer now, which is in India, in the insurance space,” she says, without naming the account. The interest, she adds, is concentrated less on employee-side security and more on customer identity — the digital experience a policyholder has from first sign-up through years of ongoing engagement.
To illustrate what “identity” actually covers in practice, Fon reaches for a scenario most people will recognise from their own banking app. Picture a customer travelling abroad who logs in to make a transfer. Because the bank knows this customer doesn’t usually travel to a particular country, the system triggers multi-factor authentication before the transaction goes through. It isn’t only checking a password — it’s cross-referencing location against known behaviour. If the same customer then tries to send an amount far larger than their usual transfer, the system flags that too, prompting another verification step. “So then to me, as a consumer of the bank, I feel safe,” Fon says. “I know that the bank has other things” checking on my behalf.
That layered, behaviour-aware approach — rather than a single login check — is what Ping is trying to sell as the baseline for identity generally, before AI agents even enter the picture.
When the “user” is no longer human
The harder conversation, and the one that dominates Fon’s view of where the industry is headed, is what happens once organisations hand off tasks to autonomous AI agents. “Traditionally, most people when you think about identity, you only think about locking password or passwordless,” she says. Agentic AI breaks that model in two directions at once — it creates new fraud vectors driven by AI, and it introduces a category of actor, the autonomous agent, that has never fit neatly into existing access frameworks.
Fon’s front-door analogy is really an argument against uniform access. A human employee with a badge can walk into a building and, historically, identity systems have mostly cared about whether they got through the door at all. An AI agent might get through that same door and then quietly access far more — more applications, more data, more systems — faster than any human review process could catch. “These agents, they are very smart,” Fon says. “They get a lot of information. They can take [actions] on their hand and then you will not be in the level of authenticating and governing what actions the AI can take.”
Her prescription has several concrete layers, each of which Ping is building into how it frames “identity for AI”:
Every AI agent gets its own assigned identity, distinct from the human who deployed it
A human stays in the loop, so there’s a clear line between actions a person took directly and actions a person merely authorised an agent to carry out
Access is authorised at a fine-grained level — not just whether an agent can get in, but exactly what it’s allowed to do and which data it can touch
Every action, every access, every piece of data an agent touches is logged and auditable
Permissions follow a least-privilege model: an agent gets only the access needed for a specific task, and that access is revoked immediately once the task is done
Fon ties this directly to India’s Digital Personal Data Protection Act, arguing that consent and data protection are necessary but no longer sufficient. “It’s no longer just about protection of the data,” she says. “We look at it as in, how are you going to make sure that you have accountability to access operational access rights, where these agents are going to be.”
Zero trust, applied continuously rather than at login
Underpinning all of this, Fon argues, is a shift from one-time authentication to continuous verification — what Ping markets as Runtime Identity. The distinction she draws is almost architectural: identity checked once at the door versus identity checked at every step someone (or something) moves through the building. “How do you know that it’s still the same person moving, or actually through the window or to the side door, somebody else already in the room?” This, she says, is classic zero trust — never assuming continued legitimacy just because access was granted once — extended to cover both human and machine actors.
That verification is risk-weighted rather than uniform. Behavioural signals and AI-driven risk scoring determine whether a given action clears a threshold that requires a second factor, or is low-risk enough to proceed without friction. It’s the same logic as the bank-transfer example, generalised into a platform capability rather than a one-off rule.
The tension CIOs can no longer split the difference on
Asked to look ahead six to twelve months, Fon points to a shift in what customers — and by extension, the enterprises serving them — expect from identity systems. The old framing, she says, was a trade-off: more security meant more friction, and CIOs and CISOs had to pick a point on that spectrum depending on the use case. A digital bank account might tolerate a slower, more secure onboarding; a retail checkout would not.
That trade-off, in her view, no longer holds. “These days, the customer expect both,” she says — a fast, low-friction experience and airtight protection of their data, simultaneously. Meeting both, she argues, requires identity platforms that can distinguish legitimate users from fraud and bots in real time, without adding steps for the people who are exactly who they claim to be. Account takeover, she notes, remains the most common form of identity fraud — and the systems that catch it well are the same ones that can afford to leave genuine customers alone.
Fon’s broader prediction is that most organisations, in India and elsewhere, are still in an early, exploratory phase with agentic AI — identifying use cases rather than scaling them. But she argues the sequencing matters: enterprises that haven’t modernised their core identity platform first will find agentic AI harder to trust and harder to govern once it does scale. “The future of agentic AI rests a lot on the readiness of a lot of organisations in their identity overall,” she says — not just the customer-experience side of identity, but the security and governance foundations underneath it.
For CIOs weighing where to spend on AI readiness, that’s the practical takeaway buried in an otherwise routine data-centre announcement: the infrastructure question is really a governance question, and the vendors racing to answer it are betting that identity — not model capability — is where enterprise AI adoption will actually get stress-tested.