Vice President Sales

14 tips for human and AI identity management from Ping Identity. September 28, 2026 Securing human identities has always been a mess. Then came AI agents. Now the mess has its own agents, running 24/7, operating at machine speed, and occasionally deciding the best way to complete a task is to do something you didn't intend. At the Ping YOUniverse 2026 conference in Austin, I chatted with three leaders at Ping Identity about how they're navigating today's wave of human and agentic identity. Big thanks to our sponsor, Ping Identity! At Ping Identity, we help organizations secure and manage digital identities across customers, employees, partners, and non-human entities. Whether securing millions of users, fighting fraud, simplifying third-party access, or enabling passwordless experiences, establishing trust in every digital moment shouldn't slow you down. Our enterprise identity platform is designed for scale, flexibility, and integration across cloud, hybrid, and on-prem environments. With our Runtime Identity capabilities, Ping enables organizations to adopt AI and automation by continuously verifying identity, context, and intent at every interaction, helping secure and govern AI agents in real time. 1. Healthy paranoia is the new best practice. "Healthy paranoia" is probably exactly where organizations should be, said Andre Durand, CEO and founder of Ping Identity, acknowledging that agents act much faster than humans. They don't share our motives, and they've been documented as being "actively deceptive" when threatened with shutdown. You're not being paranoid if you're right. 2. You can secure what you can identify. The opposite is also true. "If you can't identify what sits between you and another actor in a digital transaction, and you don't track what it does or what it could do, you don't have an inherently secure environment," said Durand. This is akin to the oft-repeated "you can't secure what you don't know" mantra that built the entire asset management industry. As we saw at Black Hat 2026, the cyber marketplace is awash with vendors offering services to identify and audit agentic behavior. Nothing is perfect yet, but that's not an excuse not to get started. Discovery is the baseline that lets you set policy and investigate incidents. 3. Your next action is the new gate, then the next action, and... We've spent decades securing the moment of login. A little 2FA and a firewall were going to be enough to protect us. Remember the "hard shell with the squishy center" candy bar metaphor? Unfortunately, that model broke years ago when it became easy to compromise single-factor and then two-factor credentials. The industry's first response to this security flaw was microsegmentation, or dividing up your entire environment into a series of mini-firewalls. Ping Identity's philosophy is to crank this up yet another notch. Validate each action, for humans and AI agents. It comes off as an extreme version of microsegmentation. "Authentication gates access. Tomorrow, authorization will gate action," said Durand, who noted that AI acts continuously, context shifts mid-task, and risk evolves as the agent changes behavior midstream. Authorization must become a continuous evaluation of trust. 4. Even a single session can spin out of control. Giving an autonomous agent a 20-minute session and saying "have at it" is inviting both bounded predictability and unbounded actions. Peter Barker, chief product officer at Ping Identity, offered the metaphor of selecting a route on Google Maps. You start by selecting a destination. At first glance, a human may see four different possible routes ("bounded predictability"), but since it's a machine that can take many more factors into account, such as other goals and other agents (cue Hugging Face), it will choose unforeseen routes that satisfy its needs, not necessarily yours. 5. Think of every AI agent as a known adversary you hired on purpose. Here's a mental exercise: Imagine you knowingly hired an adversary, gave them a computer and a badge, and let them sit down and start working. If you can control what is "acceptable work versus unacceptable work," said Durand, "you probably have the right mental construct to secure AI." Agents will probably do what you expect nine out of ten times. The issue is that tenth time. The non-deterministic nature of agents means they'll eventually come off the rails. You have no choice but to plan for that. 6. Access the specific asset in the timeframe needed, and then you're done. Think of agent permissions in two dimensions. The x-axis is your surface area and the y-axis is access time. The area between those two is your "blast zone." If you provide "just enough access it can reduce the attack surface, and just in time can reduce the blast velocity," said Durand. To pull this off, you need zero trust architecture that assigns and denies permissions quickly. 7. Your agentic speed is not necessarily the same as your neighbor's. Everyone wants to move faster. The problem is that speed requires a trust brake. Blindly trusting AI agents to just "do what it's told" is a "recipe for disaster," said Durand. His maxim: "Move at the speed of trust, no faster." As AI improves and we improve with it, our speed should increase over time. 8. The FBI's tips for spotting a deepfake were also seen by the attackers. State-sponsored actors have been using deepfakes to apply for jobs at more than 300 U.S. technology companies. This was the warning John Cannava, CIO of Ping Identity, received from the FBI. The "heads up" was partnered with some helpful tips such as asking the person on the video call to show you the electrical socket (is it U.S. based?), verify their weather (does it coincide with where they claim to be calling from?), and ask them to wave a hand in front of their face (AI often leaves artifacts). For a while, those techniques worked. But the attackers started listening to the FBI as well. Now, they can easily circumvent all these tips. You can now ignore the FBI's advice. 9. Humans should be re-verified as well. With all this talk of reverification, shouldn't we do the same with our human employees? When onboarding, the standard practice is to contract a third party to run a background check and then copy their driver's license. That's it. "That creates some risk when we trust that person for their entire lifecycle," noted Cannava. For Ping's own workforce rollout across 34 countries, Cannava led his team to a non-disruptive self-service reverification process that took under two minutes per employee and had a 97% success rate. He advised all companies with 300+ employees to implement a reverification process, since their size makes them a target for attackers. 10. Let "failure to verify" be a normal part of the process. While the rollout went better than planned, Cannava knew perfection wouldn't be possible. "Don't treat failures as exceptions. They should be normalized," said Cannava. Once Cannava's team understood what could go wrong, they built an auto-routing mechanism. For example, technical failures went to IT, name mismatches (a common problem) went to HR, and security red flags immediately locked the account and triggered an investigation. "You don't want any time between the individual seeing the failure and the security organization taking action," said Cannava. 11. Don't wait for a breach to start re-verifying your workforce. Ping didn't verify its workforce because it had found a bad actor. It verified because it hadn't yet. "I suggest, just like us, be proactive about this," Cannava said. Proactive verification lets you plan appropriately, maintain business continuity, run it on your terms, and find the problems before they find you. 12. Don't let your help desk be your verification Achilles' Heel. A multi-tiered perpetual verification system can easily be bypassed at the help desk. That worker, often a third-party contractor operating on different incentives, will easily fold when faced with a savvy, high-pressure caller. Barker noted that even after massive training campaigns, "Five percent of the help desk is still compromised." The best solution is to remove human judgment entirely from the account reset workflow. 13. Humans cannot monitor AI agents at machine speed. Barker acknowledged that managing agent accountability "is going to require AI watching AI." Don't worry, it's not an all-or-nothing proposition, he said. You can do this incrementally. Don't expect AI-assisted security operations to be built overnight. It will take a series of small wins to get comfortable. 14. Continuous authorization is a business lifestyle choice. "Authorization is not an event. It needs to become, like authentication, a continuous evaluation of trust," said Durand, who is very bullish on intent-based access control. An agent tells you what it's going to do before it does it. If your rules deem it appropriate, then authorization for that action is granted. It might start simply and non-intrusively. For example, an agent wants to read and summarize a Jira ticket. That's fine. Or maybe it wants to send a Slack message. That's ok too. But now it wants to delete a repository or read and copy HR records. No, I'm not comfortable with that. But that's today. Tomorrow, our comfort levels may change. CONCLUSION: The goal isn't to slow AI down. It's to trust it at a speed you can manage. "We don't live in a world where everyone is trustworthy, where words and actions always align," said Durand, who admits that problem is exponentially harder when the actors are autonomous, probabilistic, and occasionally deceptive. The answer isn't to avoid agentic AI. It's to build the infrastructure that lets organizations trust it safely and quickly. The companies that figure out how to break the speed-versus-trust paradox, said Durand, will be the ones that thrive.

Ping Identity launches controls for personal AI agents. Wed, 2nd Sep 2026 (Today) Ping Identity has launched Enterprise Personal Agent Access, aimed at companies using personal AI agents such as Claude in workplace systems. Available through PingOne Privilege, the offering is already being piloted with global enterprise customers. It is designed to identify personal AI agents in supported environments, link their sessions to a user and device, and control what those agents can access and do. The launch comes as businesses grapple with the spread of AI agents beyond centrally approved software. Employees and developers are increasingly using desktop assistants and coding agents in day-to-day work, sometimes with access to internal services, repositories, databases and cloud infrastructure. That creates a security problem for organisations that built identity controls around human users rather than software agents acting on a person's behalf. Ping argues that these tools can move across systems at machine speed, making it harder for companies to track who initiated an action and whether it should have been allowed. Research from Gravitee, cited by Ping, found that 48% of production AI agents are running unsecured. Against that backdrop, identity and cyber security suppliers are focusing more closely on how enterprises can discover agent use, restrict access and retain an audit trail. Runtime controls Ping said its system detects an AI agent when it is initiated in supported environments and associates the session with the person and device behind it. It then applies policy enforcement in front of managed resources to decide whether actions should be allowed, denied or logged, whether a sensitive task should require human approval, or whether access should be revoked in real time. The approach extends beyond AI applications themselves. Personal AI agents may connect to MCP servers, code repositories, internal services and APIs, Kubernetes clusters, databases and cloud systems, potentially giving them access to some of the most sensitive parts of a company's technology estate. For software development teams, Ping said the product allows agents to commit code and reach approved resources without storing long-lived credentials. It also records whether an action was taken by the developer or by the agent, giving security teams a clearer record of activity. The launch also reflects a broader shift in enterprise software towards governing a mix of AI models and agent tools from different providers. Ping said its controls are intended to work across multiple agent environments, with Claude and Claude Code highlighted as supported examples. Andre Durand outlined the company's view of the challenge facing large organisations as agent use broadens across departments and workflows. "Enterprises are deploying a number of different AI models, agents and platforms," said Andre Durand, Founder and CEO, Ping Identity. "The security challenge is establishing a common way to see and govern all of them. The question isn't whether an agent is intelligent enough to act, but whether the enterprise can see and control its action when it does." Growing concern Security reviews have become a sticking point for many AI deployments, especially when an agent can take actions rather than simply generate text or recommendations. In those cases, companies need to know what agents are running, what systems they can reach, what permissions they hold, and whether responsibility for a mistake lies with the user or the software agent. Identity management vendors have been adapting access controls, authentication and authorisation systems for this new layer of software activity. The aim is to preserve the accountability companies expect from human users while recognising that AI agents can execute tasks autonomously and at much greater speed. Ping said Enterprise Personal Agent Access is part of that effort, with a focus on visibility from discovery through to enforcement at the moment an action is attempted. It also said it took part in Anthropic's Project Glasswing to evaluate advanced AI in the security and resilience of its own codebase and internal systems. The latest product is likely to be watched closely by companies trying to balance AI adoption with internal controls, especially where personal AI agents are already appearing in developer workflows and on employee desktops before formal governance has caught up. According to Ping, every recorded action can be attributed to the developer or the agent involved in the session, giving security teams a record of what the agent did, when it happened and the user associated with the session.

Ping Identity has launched Enterprise Personal Agent Access, a solution designed to provide visibility and control over personal AI agents like Claude within enterprise environments. The platform addresses security concerns as 48% of production AI agents run unsecured, according to the Gravitee State of AI Agent Security report. The solution, delivered through PingOne Privilege, discovers AI agents including shadow AI, associates each session with its user and device, and governs access at runtime. It enables enterprises to monitor which agents are operating, identify the users behind them, and enforce access controls without impeding AI adoption. Enterprise Personal Agent Access ties agent sessions to specific users and devices, applies identity and authorization policies, and maintains accountability by logging all agent actions. This allows security teams to track what agents do whilst enabling developers to work without long-lived credentials. The solution is currently available and being piloted with global enterprises.