Ian Glazer, head of continuous identity and product strategy at CrowdStrike, presents at RSAC 2026. (Photo Credit: Laura French)
SAN FRANCISCO – Continuous validation, securing AI agent identities and preparing for cryptography-breaking quantum computers were major identity themes presented throughout RSAC 2026 in San Francisco this week.As attacks grow faster and more sophisticated, there is a pressing need to rapidly respond to identity compromise, secure non-human identities and strengthen foundational cryptographic algorithms.These challenges often expand beyond the technical aspect, with several talks discussing how to balance security with user experience and how to communicate identity security needs to company boards.“The risk is at the business level, not just at the technology level,” said IBM Security CTO Sridhar Muppidi during the session “Quantum-Safe Readiness: Practical Steps for Identity and Data Protection” Monday morning.
Making the move from static to dynamic validation
Identity compromise is at the forefront of cyber risk, with Palo Alto Networks’ Unit 42 reporting earlier this year that attackers used identity-based techniques to gain initial access in 65% of incidents.“[There is a] dramatic shift in how quickly an actor who has been trusted and verified can become a malicious actor and the time it takes to do that has significantly reduced,” LinkedIn Engineering Manager Mitendra Mahto said during the session “Beyond Zero Trust: Continuous Validation for Modern Enterprise Security” Monday.
The speed at which these compromises now occur means that static validation of who someone is and what they have access to is no longer sufficient, speakers said. Instead, dynamic and continuous monitoring of context such as time, location, device and behavior is needed to detect anomalies in real time and revoke access faster than attackers can exploit it.“There is a realization that we can’t keep doing the same things and expect better outcomes. We cannot just IGA harder and expect something different to happen,” said Ian Glazer, head of continuous identity and product strategy at CrowdStrike, during his session “Beyond SOX: Generational Change in Identity Management” Tuesday.Glazer recommended businesses move toward just-in-time access schemes and lean into the principle of zero standing privilege, which he calls “identity’s key contribution to zero trust.”This means minimizing persistent “birthright” permissions for accounts and only granting access when it is needed, something Glazer says services like AWS and Salesforce do well with ephemeral users and session-bound privileges, respectively.At the same time, contextual signals such as behavioral risk scores, managed device status and on-duty status can be used to continuously validate an identity’s trusted status and determine when access should be revoked.Frameworks such as the OpenID Foundation’s Shared Signals Framework (SSF) can assist in applying these controls, with Glazer calling SSF’s Risk Incident Sharing and Coordination (RISC) and Continuous Access Evaluation Profile (CAEP) components a “dial tone for continuous identity.”Open-source resources also played a key role in LinkedIn’s approach to implementing continuous validation, Mahto and colleague Sanjay Singh, staff software engineer at LinkedIn, said during their session.“We really strongly believe in the open-source policies, so it was very natural for us to go to this ecosystem and leverage it and ensure that we provide a platform which is vendor-free — no vendor lock-in — and it needs to be portable across on-prem, across cloud,” said Singh.LinkedIn’s system is based on network-segmented trust zones defined across two dimensions — managed/unmanaged and trusted/untrusted — with validation required to move across zone boundaries. A central reverse proxy acts as a control zone where requests are evaluated and authenticated.For network segmentation, LinkedIn used open-source technologies including a Linux-based VLAN, Cilium, Open vSwitch and FRRouting, with Cilium NetworkPolicy and nftables enforcing access controls. Its proxy stack relied on Linux Virtual Server’s IPVS, Envoy and HAProxy, while gRPC and xDS handled control plane communications.Continuous validation becomes even more important in the age of AI, when a malicious or compromised agent can conduct a large-scale breach much faster than a human attacker ever could.“If I can maybe read five documents in a couple of minutes […] an AI agent can run through 500 documents in a few minutes,” said Vijeth Lomada, lead AI engineer at Adobe, in the Monday session “Trusted Identity Propagation for Autonomous Agents Across Cloud & SaaS.”Thus, it is crucial to not only know in real time whether an agent has been compromised — and to be able to revoke access rapidly — but also to know when an identity is an agent, in order to determine what it should have access to.
Securing the identity of our AI 'digital coworkers'
The agentic age has made securing non-human identities more important than ever, and the extensive capabilities and autonomous nature of these agents come with new challenges in trust and authentication.Agents need access in order to unlock the capabilities business now rely on them for, but can wreak havoc if this access is misused, creating a delicate balancing act between security and efficiency.“Zero trust is critical and needs to be extended to AI,” said Vasu Jakkal, corporate vice president of Microsoft Security, during Monday’s main stage keynotes. “We need to have continuous verification, least-privilege access and assume breach across identity, data and runtime behavior.”Cisco President and Chief Product Officer Jeetu Patel focused his keynote address on “Reimagining Security for the Agentic Workforce,” saying that unlike other non-human identities, AI agents should be viewed not as tools but as “digital coworkers.”This doesn’t mean treating agents like humans, Patel noted — agents can be unpredictable, and act without fear of consequences, making them even more dangerous.“With agents we’re going to need to make sure that we move our mindset from access control to action control,” said Patel.This means clearly identifying each agent in an organization’s ecosystem and ensuring every agent has an accountable human identity behind it, giving short-lived, strictly scoped, just-in-time permissions for each agentic action and dynamically measuring behavioral and threat context to intercept unwanted actions, such as connections to malicious servers or unintended file deletions.The Monday session hosted by Lomada and AWS Senior Solutions Architect Swara Gandhi went into greater depth about how this can be achieved, focusing especially on OAuth-based agent access methods. The speakers stressed the importance of tracing the delegation path between humans and agents and preventing privilege escalation between these identities.“This becomes really important, that when we look at the logs, the logs at the resource level, we can differentiate if this API call was made by a human user or an agent, or if the agent was working on behalf of the user,” said Gandhi.The presentation outlined how OpenID connect (OIDC) can be used with OAuth scoping and RFC 8693 token exchange to identify the human user in charge of the agent, what the agent is allowed to do on behalf of the user and when the agent was acting vs. when the human user was in control. This creates an auditable log trail of agentic actions while restricting an agent’s access to the human user’s permissions.Defending against AI-driven identity attacks was also an important topic at RSAC 2026, as Brian Contos, Field CISO at Mitiga, noted that 16% of phishing attacks were already being AI-driven by 2024.“Next year, this will probably be closer to 96%,” said Contos, who presented the Monday session “It's Getting Real & Hitting the Fan 2026: Real World AI(dentity) Attacks” on behalf of Mitiga Co-founder and CTO Ofer Maor, who could not attend.The presentation went over how high-profile threat actors, including Scattered Spider, ShinyHunters and North Korean state-sponsored groups like Lazarus, leverage AI-powered phishing, vishing, deepfakes and impersonation to gain access to privileged identities within target organizations.Contos also covered how AI and identity converged in the Salesloft Drift supply chain attack, where a stolen OAuth token allowed UNC6395 attackers to access victims’ Salesforce environments via the Salesloft chatbot’s identity.Leveraging AI and automation are now the only options defenders have to match the machine speed of these AI attacks, according to Contos.“You want to be as close to real time as you can,” Contos said.
Preparing for a quantum ‘existential crisis’ for identity security
While AI and agents dominated many of the identity conversations throughout RSAC this year, another major technology shift that could soon shake the very backbone of identity security also received recognition. Quantum computing will give attackers the ability to crack the cryptographic algorithms that underly most of the authentication systems used today.“The impact of all this allows attackers to impersonate, to attack, to perform man-in-the-middle attacks; every attack vector that we use crypto as a defense mechanism [for] — now we lose that defense mechanism,” said Bassam Al-Khalidi, cofounder and chief innovation officer at Axiad, in the session “Building Quantum-Ready Identity Defense with Risk Scoring & PQC Resilience” Tuesday.IBM Quantum Safe Fellow Ray Harishankar, who presented alongside Muppidi, said a cryptographically relevant quantum computer, which could crack RSA-2048, is likely to arrive by 2033, while computers that can break less powerful — but still widely used — algorithms may arrive closer to the turn of the decade.Al-Khalidi, along with co-presenter Rich Bush, cryptography security leader at State Farm, said that “Q-day” may be even sooner — or even today — depending on an organization’s cryptographic posture. However, all speakers agreed that the need to implement post-quantum cryptography (PQC) is extremely urgent, and businesses can’t afford to wait any longer.For any systems that are not quantum-ready by the time Q-day hits, Bush says businesses will be faced with decision of whether to shut those systems down or continue to use weaker algorithms — and that option two isn’t really an option.“That’s an existential crisis,” Bush said.An IBM survey of 750 CISOs identified a severe “readiness gap” within organizations where 73% of leaders are aware of the importance of quantum readiness but only 19% have taken any tangible action to achieve PQC. One of the challenges for leaders is understanding where the responsibility for PQC lies, with 63% saying it’s a “vendor problem” and few respondents agreeing on who in their organization “owns” this responsibility.“The understanding is all over the place,” says Harishankar, who emphasized that leaders need to recognize that post-quantum is a business risk, not just a technology problem.To move toward post-quantum readiness for identity — and all other areas of security touched by cryptography — organizations need to begin building a comprehensive inventory of their cryptographic assets, prioritize the highest-risk areas and leverage automation to continuously monitor cryptographic posture and risk scores. Both sessions emphasized that PQC is not a one-time event but a continuous process that will require updates well after Q-day arrives. Key to this evolution is the concept of crypto-agility — the ability to quickly rotate keys and certificates or change cryptographic algorithms without disrupting the business while maintaining an accurate inventory of the organization’s crypto ecosystem.“It’s not a project — it’s a program. It requires a level of transformation across different parts of the organization,” Muppidi said.
