RSAC: AI Outpacing Defenses
Security leaders at RSA 2026 warned that AI-driven attacks are outpacing enterprise readiness—half of organizations admit they're not ready, and defenders must rethink controls around AI and GPU-accelerated infrastructure. The consensus: legacy controls aren't enough, and identity/zero-trust strategies are now central to defending AI factories. (zdnet.com)
Microsoft’s Entra team reported that 97% of organizations saw an identity or network‑access incident in the past year and that 70% of respondents linked incidents to AI‑related activity. (techcommunity.microsoft.com) EY’s recent study found 96% of senior security leaders view AI‑enabled attacks as a significant threat and about 48% estimate that at least one quarter of their incidents last year were AI‑enabled. (ey.com) Multiple RSAC sessions flagged that conventional EDRs and telemetry focus on CPUs and OS activity while leaving GPUs largely unmonitored, creating an attack surface in GPU‑accelerated AI infrastructure. (businessleadersreview.com) Conference panels and vendors warned those GPU blind spots leave “AI factory” runtimes without runtime controls, prompting calls for new GPU‑level telemetry, workload segmentation, and supply‑chain checks. (techrepublic.com) Cisco told attendees it is extending zero‑trust access capabilities to cover AI agents as a way to bind identities to agent behavior, while Google Cloud presented agentic‑AI defense integrations tied to frontline threat intelligence. (crn.com) Cybersecurity Insiders fielded an AI Defense Gap survey of more than 150 CISOs at RSAC to quantify which controls hold and where gaps remain in AI deployments. (securityledger.com) Vendors demonstrated offensive testing tools ahead of RSAC, including Votal AI’s RLHF‑trained adversarial attacker model and an open‑source attack catalog shown before the March 23–26, 2026 conference. (marketwatch.com) Exhibitors and analysts positioned those red‑team resources as validation tools for zero‑trust policies, identity controls, and GPU‑aware detection in production labs and MLOps pipelines. (pointguardai.com)
