Security | Threat Detection | Cyberattacks | DevSecOps | Compliance

Public benchmarks in AI provide important signals and allow for regression testing, directional validation of model updates, and public discussion of capabilities and limitations. But the more attention a benchmark receives, the stronger the incentive to optimize for it. Once a score becomes the goal, teams start benchmaxxing: optimizing for the benchmark rather than the capability it is meant to measure. This is a familiar problem in the AI space.

VMware ESX systems are a recurring target in ransomware campaigns. Threat groups including SCATTERED SPIDER, BlackBasta, Royal (aka BlackSuit), Akira, and the ESX-focused ransomware as a service (RaaS) platform shinysp1d3r have demonstrated that once an adversary reaches the hypervisor layer, they can rapidly encrypt virtual machines, disable logging, and cripple an entire data center.

At CrowdStrike, we conduct extensive red-team testing of agentic systems using diverse models, tools, and adversarial evaluation harnesses designed to probe for containment failures. To date, none of our offensive agents have escaped their intended sandbox boundaries.

The CrowdStrike 2026 Threat Hunting Report illustrates the next evolution in trust abuse. Adversaries are targeting trusted users and tools across identity systems, cloud environments, SaaS applications, AI services, software supply chains, and developer workflows to blend into legitimate business activity and reach critical assets before defenders can detect them. Our frontline intelligence in this year’s report underscores this shift.

Employees are already using AI at work. They build agents in Microsoft Copilot Studio, write code with Claude Code, and paste sensitive data into chatbots in the browser. Each of these actions can expose sensitive information outside of approved workflows, and most of it happens where traditional endpoint, network, and data loss prevention (DLP) security controls can't see the prompt or the tool call.

Organizations face a relentless stream of emerging threats, from zero-day exploits to rapidly evolving adversary tactics. They need protection that keeps pace with this changing landscape without adding operational complexity. The key challenge here is turning threat intelligence into production-ready detections before attackers can gain an advantage.

Every change in a cloud environment creates new security decisions. A new infrastructure as code (IaC) template needs to be validated. Cloud permissions need to be reviewed. An application release introduces new cloud interactions. A Kubernetes cluster needs protection before it goes into production. Individually, these are routine tasks. Together, they create growing operational friction that makes cloud security harder to scale.

AI is changing the speed and scale of cyber defense, and the speed and scale of the adversary. As AI becomes embedded across government, critical infrastructure, and enterprise environments, defenders need the ability to inspect, test, adapt, and secure the systems they depend on.

On June 10, 2026, the Cybersecurity and Infrastructure Security Agency (CISA) issued Binding Operational Directive 26-04, which transforms federal vulnerability management by shifting agencies from static CVSS-based patching to a dynamic, risk-based model. This supersedes BOD 19-02 and BOD 22-01. Agencies must now prioritize remediation using four key factors: public asset exposure, KEV catalog status, exploit automatability, and technical impact (partial vs. total control).

In February 2026, Socket.dev published research on a multi-stage npm supply chain worm operating under the internal flag SANDWORM_MODE. The campaign spanned 19 malicious packages in total across two unique publisher aliases and demonstrated a new class of supply chain attacks that targeted AI-augmented development workflows.