ISC2 Security Congress runs October 24 to 28 in Denver, and once again, we’ll be sharing our thoughts on this year's sessions. Last year we picked sessions on scaling security, strategic cyber investment in the AI age, and the data exposure that comes with operationalizing generative AI. 365 days later, the tension isn’t easing. Organizations are adopting enterprise AI while their security programs try to keep up with new tools, and every assistant, agent, and pipeline they add becomes another identity with access to sensitive data. Thank goodness we’re all on the way to ISC2, where we can discuss how to keep securing the risks AI introduced.
This year, Four sessions stand out: a quantum-era threat model for de-identified data called Harvest Now, Re-identify Later; the Department of Defense’s move from RMF to continuous risk management; zero trust deployments at enterprise scale; and identity governance across CI/CD pipelines. We read each one through a question that follows from the AI tension above: what can an intruder, an overprivileged service account, or an AI agent read once access is granted or taken? That is the question Confidencial works on, so here are the four sessions with our take on where data protection fits in each.
That is the question Confidencial works on, so here are the four sessions with our take on where data protection fits in each.
Harvest Now, Re-Identify Later: Quantum Computing and the Architecture of Identity
Tuesday, October 27 · 4:45 PM MT · Adams C
Harvest Now, Decrypt Later (HNDL) focuses on encrypted data that adversaries collect today and open once quantum hardware matures. This session introduces another quantum-era threat model: Harvest Now, Re-identify Later (HNRL). HNRL is when de-identified data sets shared under reasonable anonymity assumptions become identifiable after the fact through quantum-accelerated linkage and inference attacks. The session will examine the regulatory gap created, since current privacy frameworks trigger obligations based on present identifiability rather than future identifiability. The session will also make the case for differential privacy and recalibrated threat models for long-lived data sets while arguing that the stakes extend beyond data security to human autonomy and identity development.
Confidencial's POV
HNDL assumes the data was encrypted when it was taken. HNRL describes data that was only de-identified, where protection rested on assumptions about what an adversary could link or infer, and those assumptions can expire. Re-identification does not wait on quantum hardware: in 2017, University of Melbourne researchers re-identified patients in Australia’s open Medicare data, a government release published as de-identified. The session’s premise is that quantum-accelerated linkage and inference attacks could extend that risk to data sets shared under reasonable anonymity assumptions.
Confidencial’s "harvest now, read now" argument starts with a related observation: quantum is not the only threat, because too much sensitive data is already exposed in usable form inside applications, documents, and AI workflows. The argument applies here as well: where is sensitive data readable today, and how can we reduce that exposure at the content layer? For documents and files that carry identifying details, Confidencial’s answer is selective encryption, which protects the sensitive pieces within a document.
Framework to Function: Moving from Static Compliance to Dynamic Risk Management and Resilience
Monday, October 26 · 11:20 AM MT
Organizations rely on frameworks such as NIST and ISO, while their programs focus on periodic assessments, documentation, and fixed system boundaries which don't support continuous risk management. Nakita Albritton of Infortress presents a transition model to move from static compliance to continuous operational risk and resilience, using the Department of Defense's shift from RMF to the Cyber Security Risk Management Construct (CSRMC) as a working example. The session covers operationalizing existing controls through continuous monitoring, integrating supplier and service provider risk, while exploring the alignment of governance, roles, and data flows.
Confidencial's POV
A fixed system boundary is a reasonable unit for a periodic assessment, but data does not stay inside it. Files move to suppliers, service providers, cloud platforms, and AI tools, and once they do, the framework's controls describe the organization's own systems and not the data that left them. Continuous monitoring can show where data flows and who touches it, yet the controls at the other end belong to someone else. Protection bound to the data itself closes that gap. Selective encryption keeps sensitive content unreadable to anyone not authorized to access it, wherever the file sits, so the risk decision made at the point of sharing holds even after the file leaves. This approach makes supplier risk easier to monitor because the data's protection no longer depends on every supplier's controls staying current.
Implementing Zero Trust to Secure Hybrid and Multi-Cloud Enterprises
Monday, October 26 · 2:45 PM MT · Summit 8/9
As enterprises adopt hybrid and multi-cloud environments, perimeter-based security may no longer cover critical data, applications, and workloads. This session draws on real deployments across cloud and hybrid infrastructure, where zero trust at scale meant adaptive authentication, micro-segmentation, least-privilege access, and continuous monitoring. Those controls reduced attack surface, prevented lateral movement, and improved resilience, and attendees leave with guidance for building and maintaining the same architectures.
Confidencial's POV
Zero trust removes implicit trust from users, devices, and systems. At Confidencial, we approach it from the data itself: the model stays incomplete when protection isn't bound to data objects, because identity, network, and device controls don't protect an asset that becomes broadly readable once access is granted. That gap widens as AI assistants and agents ingest, summarize, and correlate data across systems, since each one adds access paths and machine identities. Extending zero trust to the data means the file stays protected inside and outside the approved workflow, and selective encryption applies that protection only to the content that needs it, so access to a document doesn't automatically grant access to everything inside it.
Identity as the Control Plane for Software Supply Chain Security
Wednesday, October 28 · 10:45 AM MT
According to the session description, people often blame supply chain breaches on code tampering, but many start with compromised identities, overprivileged service accounts, and weak access governance in CI/CD pipelines. The session, with co-speakers Vatsal Gupta of Apple and Alex Olivier of Cerbos, treats this as an identity governance problem and follows authorization decisions for people and machines through source control, builds, artifact repositories, and deployment. It covers segmenting non-human identities, enforcing least privilege in pipelines, federated access governance, and continuous monitoring, and it ends with a framework that ties IAM, DevSecOps, and enterprise risk management together to shrink blast radius and improve auditability.
Confidencial's POV
Least privilege and segmentation limit how far a compromised identity can travel, but they do not determine what that identity can read at each stop. In 2025, attackers used OAuth tokens stolen from the Salesloft Drift integration to read Salesforce data at hundreds of companies, including security vendors, and mined the exported records for AWS keys, passwords, and access tokens to other systems. In this example, identity controls failed first, and the scale of the loss depended on what the stolen identity could read. Data-level protection changes that outcome because the identity may reach the file, but the content stays unreadable unless that identity is authorized for it.
The question underneath all four
A re-identification attack, a compromised pipeline credential, and an AI agent with too much reach are different failures with the same result: someone or something the controls were supposed to stop is reading the data. The standard objection is that everything is already encrypted, and that's largely true for data in motion and data at rest. Both protections end at the moment of use, when a valid session sees plaintext, and a stolen credential or token is a valid session.
Confidencial's position is that data needs protection that holds in motion, at rest, and in use. Selective encryption seals the sensitive pieces within a document rather than the whole file, so the document stays usable while the sealed content stays unreadable to unauthorized parties, both inside an organization and after it leaves. It's crypto-agile, so moving to post-quantum algorithms is a configuration change, not a rebuild.