September 2026 Summaries

ClickFix is a social-engineering technique in which attackers use fake errors, CAPTCHA prompts, or meeting problems to persuade users to copy and execute malicious commands through trusted tools such as Windows Run, PowerShell, or scripting engines. CrowdStrike reports that the method exploits routine troubleshooting behavior and can bypass defenses focused primarily on malicious files, with fake CAPTCHA-related incidents rising 563% in 2025 according to its 2026 Global Threat Report. The post describes alleged ClickFix campaigns attributed to STARDUST CHOLLIMA, which reportedly used a fake video-conference site to deliver previously unknown GeniexLoader and GeniexRAT malware, and VOODOO BEAR, which allegedly used compromised Ukrainian websites and fake CAPTCHAs to induce PowerShell and VBScript execution. It argues that successful ClickFix attacks can expand from browser-based deception to endpoint compromise, credential theft, persistence, lateral movement, and cloud or SaaS access. CrowdStrike presents its Falcon platform, including browser, endpoint, identity, SIEM, threat-hunting, and managed-response capabilities, as a layered approach for disrupting attacks before execution and detecting or containing them afterward.

CrowdStrike and NVIDIA are collaborating on the NVIDIA Open Agent Safety Platform, an open-source reference design intended to secure autonomous AI agents across software, runtime, and infrastructure layers. The initiative argues that agents, which may access sensitive data, credentials, code, APIs, and critical systems, should be governed by independent controls rather than being trusted to enforce their own boundaries. NVIDIA technologies including OpenShell, Sentry, BlueField-4 DPUs, DOCA, and Vera CPUs provide runtime governance, out-of-band monitoring, identity verification, telemetry, quarantine capabilities, and Zero Trust enforcement, while CrowdStrike’s Falcon platform adds enterprise-wide context from endpoint, identity, cloud, data, and AI security signals. The collaboration builds on prior integrations and the Secure-by-Design AI Blueprint to help organizations apply principles such as least privilege, isolation, authorization, observability, and policy enforcement to AI agents operating at machine speed.

CrowdStrike reports that it was named a Leader in Forrester’s Q3 2026 evaluation of proactive security platforms, receiving the highest Strategy score and maximum scores for Innovation and Roadmap. The company argues that proactive security is shifting from basic vulnerability discovery toward contextual exposure management that identifies which risks are most likely to be exploited and prioritizes remediation. Its Falcon Exposure Management product combines exposure data, threat intelligence, exploit prediction through ExPRT.AI, attack-path analysis, runtime asset information, and compensating-control awareness within the Falcon platform. CrowdStrike also highlights Charlotte AI and AgentWorks for automating and customizing risk assessment and remediation workflows, while emphasizing that analysts retain decision-making responsibility. The post cites Intermex’s reported 98% reduction in critical DMZ vulnerabilities after using the product and presents integrated exposure management and security operations as a way to shorten the path from identifying risk to fixing it.

CrowdStrike introduces SafeMind, an agentic cybersecurity system developed by its Cyber Superintelligence Lab that pairs an offensive AI agent, Red Tempest, with a defensive agent, Blue Solano, in continuous adversarial co-evolution. Operating in simulated enterprise environments informed by Falcon platform network maps and telemetry, Red Tempest executes multi-stage attack campaigns across more than 1,000 scenarios and 155 MITRE ATT&CK techniques, while Blue Solano analyzes resulting endpoint telemetry to produce and validate detection logic against both attacks and clean data. CrowdStrike says this feedback loop hardens defenses by repeatedly challenging the offensive agent after it is given full knowledge of newly implemented protections. Based on the company’s internal benchmarks, Red Tempest reached full compromise at lower cost than comparable models, while Blue Solano improved detection accuracy by 70%, reduced detection-generation costs to about $0.03, and created validated detections in under six minutes. The initial SafeMind-related offering is the Frontier AI Readiness and Resilience service, which uses Red Tempest for tasks such as identifying vulnerabilities in custom applications and software dependencies.

CrowdStrike reports that it was named a Leader in Forrester’s Q3 2026 evaluation of external threat intelligence service providers, earning the highest scores in the Current Offering and Strategy categories. The company attributes the recognition to its platform-native intelligence, which combines proprietary endpoint telemetry with adversary research, threat hunting, incident response insights, malware analysis, vulnerability intelligence, and deep and dark web monitoring to connect external threats with customer environments. Forrester’s assessment reportedly gave CrowdStrike top marks in 15 criteria, including threat hunting and vulnerability intelligence, where CrowdStrike says it was the only evaluated vendor to receive perfect scores. The post argues that AI-enabled adversaries are shortening response windows and that security teams need integrated intelligence rather than fragmented tools and data sources. CrowdStrike outlines a strategy in which Threat AI and purpose-built agents support an agentic security operations center by investigating threats, prioritizing exposures, connecting context, and accelerating response actions, while noting that Forrester’s ratings are opinions based on information available at the time of publication.

CrowdStrike describes a new Falcon Data Security capability that uses on-device language models and Intel neural processing units to classify sensitive structured and unstructured data in real time, including credentials, documents, chat messages, AI prompts, and medical records. The approach aims to improve on pattern-based detection by interpreting contextual meaning while avoiding cloud-inference latency and privacy concerns associated with transmitting customer data. Developed with Intel for Core Ultra AI PCs, the system uses NPU-optimized models through ONNX Runtime and OpenVINO, selecting NPUs over GPUs for sustained, low-power background processing despite GPUs’ faster raw inference. CrowdStrike trained a large model and distilled it into a smaller FP16-quantized endpoint model that uses under 5% of the teacher model’s parameters while retaining similar classification effectiveness. The underlying portable ONNX architecture is intended to support different hardware environments, including NVIDIA GPU cloud deployments and Apple Neural Engine-based macOS devices, allowing the company to extend context-aware data protection across endpoints and cloud platforms.

CrowdStrike reports that a financially motivated, self-described bug bounty hunter allegedly created and distributed PhantomRaven, a JavaScript-based information stealer delivered through typosquatted npm packages and dependency-confusion techniques. The malware fetches a remote malicious dependency during installation and uses a preinstall script to collect system details, Git and npm configuration data, public IP addresses, and CI/CD environment variables that may contain credentials or tokens, then sends the information to attacker-controlled infrastructure through HTTP requests. Researchers assess with high confidence that the malware was likely generated with a large language model, citing unusually verbose comments, placeholder code, and token-analysis patterns, while characterizing the operator’s technical sophistication as relatively low. The activity appears intended to identify compromised company assets that could be used to support bug bounty claims rather than to sell stolen logs, although CrowdStrike notes that AI-generated tools may lower barriers to cybercrime. Recommended defenses include using private package registries, disabling npm scripts by default, updating to npm version 12 or later, reviewing dependencies carefully, educating developers about dependency confusion, and running npm audit.

CrowdStrike’s overview of Microsoft’s September 2026 Patch Tuesday reports a record 972 vulnerabilities, including 113 rated Critical and two actively exploited elevation-of-privilege zero-days in the Windows Update Stack and Windows ALPC, both of which can enable attackers with local access to obtain SYSTEM privileges. The release is dominated by elevation-of-privilege, remote-code-execution, and information-disclosure flaws, with Windows, Extended Security Updates, and Microsoft Office receiving the largest number of patches. Particularly serious issues include unauthenticated network-accessible remote-code-execution vulnerabilities affecting Netlogon, DNS Server, DHCP Server, MSMQ, Services for NFS, SSTP VPN, and other infrastructure components, several with CVSS scores of 9.8, as well as flaws in Kerberos, Hyper-V, Remote Desktop, SQL Server, and Windows graphics and media components. Twenty-two Critical Office-related vulnerabilities include numerous Outlook Reading Pane and Explorer Preview Pane attack paths that may allow code execution simply by previewing malicious content, increasing phishing-related risk. The report also notes a newly disclosed, unpatched proof-of-concept Microsoft Defender zero-day called ShieldCrash, whose claims remain under review, and recommends organizations prioritize exposed and high-impact systems while maintaining broader mitigation and vulnerability-management strategies when patches are unavailable.

CrowdStrike announced its Agentic Identity Provider at Fal.Con 2026 as part of Falcon Next-Gen Identity Security, aiming to address identity governance challenges created by autonomous AI agents that access systems, data, and tools on behalf of people and workloads. The offering discovers and registers agents, assigns cryptographically verifiable identities, links their actions to responsible humans or workloads, and provides short-lived, least-privilege access rather than persistent credentials. Combined with CrowdStrike’s previously introduced Continuous Identity capabilities, it continuously evaluates access using real-time identity, device, threat, and business context and can revoke permissions when conditions change. CrowdStrike also expanded just-in-time privileged access to SaaS applications, endpoints, servers, private resources, and AWS environments, while extending Falcon Complete MDR services to Falcon Shield for 24/7 monitoring and response to SaaS threats.

CrowdStrike outlines an expanded agentic security operations center (SOC) model for its Falcon platform, designed to help analysts and AI agents investigate and respond to threats collaboratively as adversaries accelerate attacks through AI. The company argues that fragmented data, siloed agents, and ungoverned automation limit many organizations’ ability to adopt agentic security operations, and proposes a unified platform combining telemetry, detection, investigation, orchestration, and governance. Announced at Fal.Con 2026, new and planned capabilities include certified third-party data pipelines, in-pipeline detection, coordinated specialist agents for cross-domain investigations and threat reconnaissance, and a unified Charlotte Agentic SOAR workspace for building and governing rule-based and agentic workflows. The platform also supports customer-selected OpenAI and Anthropic models, integrations with third-party tools and agents through MCP, and an API-first version of Hybrid Analysis for malware analysis workflows. CrowdStrike states that these capabilities use shared organizational context and global threat intelligence to improve agent accuracy over time, while allowing teams to automate approved actions and maintain oversight; several features remain in public preview or are subject to change.

CrowdStrike announced Real-Time Supply Chain Attack Protection, a Falcon sensor capability intended to detect and quarantine malicious open-source packages during download before embedded code executes on enterprise endpoints. The company cites growing supply chain risk from poisoned packages in public registries, particularly as AI-assisted development and agentic applications expand package downloads beyond traditional developer workstations. The capability monitors npm and PyPI package activity on Windows, macOS, and Linux, compares downloaded files with Falcon Adversary Intelligence, and can trigger fleet-wide lookback searches and automated containment for historical exposure. CrowdStrike also plans global package inventory visibility through Falcon Exposure Management in Q3 and policy controls in Q4 that could require package-age cooldown periods, limit public packages, or redirect users to approved versions. The feature is delivered through the existing Falcon sensor without separate tools or workflow changes, while planned features remain subject to change.

CrowdStrike announced Falcon Guardian, an expansion of its AI Detection and Response offering designed to secure autonomous AI agents at runtime across endpoint, cloud, and SaaS environments. The platform combines agent discovery, governance, data protection, threat detection, investigation, and response, linking prompts, tool calls, identities, and agent activity to downstream operating-system behavior so teams can assess impact and contain malicious actions. New features include controls to identify shadow AI agents, restrict unauthorized agent types on managed Windows, macOS, and Linux endpoints, reconstruct agent sessions for investigations, and a planned AI gateway for centralized monitoring and policy enforcement of AI traffic. CrowdStrike is also extending its Adversary OverWatch threat hunting and Falcon Complete managed detection and response services to AI agents, while integrating agent telemetry with Falcon Next-Gen SIEM to support cross-domain analysis and potentially reduce third-party data-ingestion costs.

CrowdStrike reported that, on August 31, 2026, it worked with U.S. and international law enforcement agencies and industry partners to disrupt Sality, a peer-to-peer botnet that had operated for more than two decades and distributed malware to over 15,000 infected systems worldwide. First identified in 2003, Sality spread by infecting executable files and relied on decentralized peer communications rather than centralized command-and-control servers, making it unusually persistent; its recent primary payload, EggJagger, replaced cryptocurrency wallet addresses copied to victims’ clipboards, with CrowdStrike estimating at least 12.1 million rubles in stolen cryptocurrency. The operation manipulated the botnet’s peer lists to remove legitimate nodes and introduce sinkholes, cutting the operator off from infected machines and preventing delivery of new payload instructions, while authorities also took down active payload-hosting URLs. CrowdStrike notes that the disruption does not remove existing malware from compromised devices and advises organizations to use the published network indicators, URLs, and YARA detection rules to identify and remediate remaining infections.