CyberProof 2026 Mid-Year Research

Executive Summary

The report examines how the first half of 2026 marked a pivotal evolution in the threat landscape, characterized by adversaries increasingly bypassing traditional endpoints to target trusted enterprise technologies—including cloud identities, software supply chains, SaaS platforms, AI infrastructure, and enterprise management systems—to achieve broad, scalable access.

The first half of 2026 saw attackers increasingly target trusted enterprise technologies rather than traditional endpoints. Cloud identities, software supply chains, SaaS platforms, AI infrastructure, and enterprise management systems became preferred paths to achieve broad access. Geopolitical tensions also drove continued state-sponsored cyber activity, while AI-assisted vulnerability discovery and the rapid release of public proof-of-concept exploits reduced the time between vulnerability disclosure and exploitation. Together, these developments reinforced the growing importance of securing identities, trusted platforms, software supply chains, and AI-enabled enterprise technologies.

Major Developments in H1 2026 

The first half of 2026 was shaped by several developments that influenced the global threat landscape. The following sections summarize the most significant campaigns, vulnerabilities, and attack trends observed during the reporting period, along with the broader changes they reflect. 

Geopolitical Tensions Drive Cyber Operations 

Geopolitical tensions remained a significant driver of cyber activity throughout H1 2026. Activity associated with the Iran-Israel conflict included espionage, disruptive attacks, influence operations, website defacements, and distributed denial-of-service (DDoS) campaigns targeting government, critical infrastructure, and commercial organizations. The escalation between Israel, the United States, and Iran was accompanied by a 168% year-over-year increase in global DDoS activity during Q1 2026. More than 70 hacktivist groups were observed launching campaigns linked to the conflict, reflecting the scale of politically motivated cyber activity. [1] 

Iran-linked activity was particularly prominent during the reporting period. Handala claimed attacks against Israeli organizations, U.S. companies, healthcare providers, and water utilities, while Ababil of Minab was linked to the disruptive cyberattack against the Los Angeles Metro transportation system. The Israeli National Cyber Directorate also reported that Iran-linked actors erased data from at least 60 Israeli organizations through destructive wiper attacks. Handala further claimed responsibility for breaching California water utility systems following U.S. airstrikes, while Iranian actors attempted to compromise internet-connected surveillance cameras across the Middle East to support intelligence collection and military operations. [1][2] 

One of the most significant incidents during H1 2026 involved Stryker, a U.S.-based medical technology manufacturer. Attackers compromised an administrative Microsoft Entra ID account, created a new Global Administrator account, and leveraged Microsoft Intune to remotely wipe managed devices across the enterprise. Rather than deploying malware, the operation abused trusted identity and endpoint management infrastructure to disrupt business operations. 

Geopolitical cyber activity was not limited to the Middle East. Russian groups, including Sandworm (APT44), continued targeting critical infrastructure, particularly the energy and water sectors, while Chinese state-sponsored actors focused on internet-facing edge devices, VPN appliances, virtualization infrastructure, and operational technology to establish long-term access. These campaigns indicate that persistent access to strategically important networks and critical infrastructure remains a priority for state-sponsored actors. 

Software Supply Chain Attacks Expand into Developer and AI Ecosystems 

Software supply chain attacks accelerated during H1 2026, moving beyond isolated package compromises to coordinated campaigns targeting the broader developer ecosystem. Rather than focusing on end-user systems, attackers increasingly targeted source code repositories, CI/CD pipelines, package registries, developer credentials, and software publishing infrastructure. The attacks also expanded into AI-assisted development environments, targeting platforms now deeply integrated into enterprise software development. More than 170 compromised npm and PyPI packages with over 518 million cumulative weekly downloads were linked to these campaigns, creating the potential for large-scale downstream compromise. [3] 

The most prominent campaign during H1 2026 was attributed to TeamPCP, which targeted GitHub Actions, npm, PyPI, Docker Hub, Visual Studio Code extensions, Trivy, KICS, LiteLLM, Telnyx, and other widely used development tools. The group repeatedly abused compromised CI/CD workflows and stolen maintainer credentials to harvest secrets, compromise additional projects, and distribute malicious code through trusted software supply chains. Within just five hours, TeamPCP published more than 400 malicious package versions across 172 packages, while GitHub later confirmed that approximately 3,800 internal repositories had been compromised. Credentials stolen during these campaigns were later linked to downstream ransomware operations, indicating that software supply chain compromises increasingly provide an initial access vector rather than serving solely as a malware distribution mechanism. 

The public release of the Mini Shai-Hulud worm source code lowered the barrier further for follow-on campaigns. Subsequent attacks, including the compromise of hundreds of npm packages and the Miasma campaign targeting Red Hat’s npm ecosystem, introduced self-propagating malware, large-scale credential harvesting, abuse of trusted code-signing and provenance mechanisms, and techniques designed to bypass multi-factor authentication during automated publishing. By June, malware also targeted AI coding assistants such as Claude Code, Cursor, Gemini CLI, and Visual Studio Code by modifying local project configuration files used by these tools. [3] 

Unlike earlier software supply chain attacks that primarily focused on distributing malicious packages, the campaigns observed during H1 2026 increasingly targeted developer identities and software publishing infrastructure. GitHub personal access tokens (PATs), npm and PyPI publishing tokens, cloud credentials, CI/CD secrets, and code-signing infrastructure became primary objectives, allowing attackers to compromise trusted release pipelines without exploiting software vulnerabilities. Analysis of more than 130,000 GitHub Actions pipelines identified over 250,000 potential CI/CD security misconfigurations, with many affecting permissions, version pinning, and other security controls within software development pipelines.[4] 

Identity-Based Attacks Drive Data Extortion and SaaS Compromise 

Data extortion continued to evolve throughout H1 2026, with attackers increasingly prioritizing identity compromise over traditional malware deployment. Rather than encrypting systems, many operations focused on obtaining legitimate access to cloud identities, SaaS platforms, and enterprise applications before stealing sensitive data and demanding payment. [5][6] 

Identity infrastructure became a primary target throughout H1 2026. Compromise of Microsoft Entra, Active Directory, OAuth tokens, API keys, developer credentials, and SaaS integrations enabled attackers to gain legitimate access to enterprise environments, allowing them to operate with minimal reliance on malware while making malicious activity more difficult to distinguish from normal enterprise operations. According to industry reporting, 83% of ransomware attacks involved compromise of identity infrastructure.[7] 

Several high-profile campaigns relied on these techniques. ShinyHunters exploited the Oracle PeopleSoft zero-day (CVE-2026-35273) to compromise more than 300 PeopleSoft instances across over 100 organizations, with universities accounting for approximately 68% of identified victims. The Klue compromise showed how a single trusted SaaS integration could be abused to harvest Salesforce OAuth tokens and access customer environments without generating traditional authentication alerts. Meanwhile, the FortiBleed credential exposure highlighted the growing value of valid identities, exposing approximately 86,000 Fortinet administrative and VPN credentials across 194 countries through credential theft and password reuse rather than software exploitation. [8] 

Cloud identities, SaaS platforms, and third-party integrations increasingly enable attackers to access enterprise environments through a single compromise. The campaigns observed during H1 2026 showed that trusted identities and inherited trust relationships now offer many of the same opportunities previously achieved through endpoint compromise, often without deploying malware. 

Privilege Escalation Vulnerabilities Accelerate Across Linux and Windows 

Privilege escalation became one of the most active areas of vulnerability disclosure during H1 2026, particularly across Linux and Windows. Multiple local privilege escalation (LPE) vulnerabilities were disclosed in rapid succession, accompanied by public proof-of-concept exploits and detailed technical analysis. Within approximately two weeks, four major Linux privilege escalation vulnerabilities were disclosed, while the broader page-cache vulnerability family expanded to at least five related flaws.[9][10] 

The Linux kernel vulnerabilities Copy Fail, DirtyFrag, Fragnesia, PinTheft, and DirtyClone were disclosed within weeks of one another. Copy Fail alone generated more than 140 public exploit variants, many derived from AI-assisted modifications of the original proof of concept. Public proof-of-concept releases also appeared within hours of upstream code changes, significantly reducing the time between vulnerability disclosure and exploit availability.[9][10] 

Windows also saw a series of publicly disclosed privilege escalation vulnerabilities affecting Microsoft Defender and other security components. Nightmare Eclipse released YellowKey, GreenPlasma, MiniPlasma, and RoguePlanet, with RoguePlanet becoming the seventh public zero-day released by the actor since early April 2026. Several earlier disclosures were later incorporated into real-world attack chains, reducing the time available to assess exposure and prioritize mitigations.[11]  

AI-assisted vulnerability discovery also became more prominent during H1 2026. Copy Fail and DirtyFrag were identified with AI assistance, while public proof-of-concept releases quickly generated hundreds of derivative exploits. As a result, organizations increasingly faced the challenge of distinguishing meaningful developments from derivative code and prioritizing the vulnerabilities most likely to be exploited in real-world attacks.[10] 

AI Adoption Expands the Enterprise Attack Surface 

Enterprise adoption of AI accelerated throughout H1 2026, introducing new attack surfaces across AI infrastructure, coding assistants, enterprise copilots, and model-serving platforms. As organizations integrated large language models into business operations and software development, vulnerabilities affecting AI frameworks, gateways, and agent platforms emerged before security controls had matured. 

AI infrastructure became a frequent target during H1 2026. Vulnerabilities affecting LiteLLM, OpenClaw, and multiple Model Context Protocol (MCP) implementations enabled authentication bypass, remote code execution, and unauthorized access to AI agents and connected enterprise resources. One exposed MCP infrastructure used in AI-assisted intrusion activity contained more than 1,000 operational files, including firewall configurations, credential dumps, vulnerability scanning templates, and attack planning data. The ClawJacked attack chain showed how trust relationships between AI agents, local services, and connected enterprise resources could be abused to gain unauthorized access.[12] 

AI assistants also became a target throughout H1 2026. Vulnerabilities affecting Microsoft 365 Copilot, Amazon Q, Gemini CLI, Cursor, and other AI coding assistants enabled remote code execution, unauthorized data access, and theft of cloud credentials, API keys, authentication tokens, and developer secrets. Many of these attacks abused automatically- trusted workspace configurations, inherited permissions, and AI tool execution rather than traditional software vulnerabilities.[13] 

AI also changed vulnerability discovery during H1 2026. Project Glasswing expanded access to Claude Mythos to approximately 150 organizations across more than 15 countries, with the platform identifying more than 10,000 high- and critical-severity vulnerabilities during early deployments, including previously unknown vulnerabilities across major operating systems, browsers, and open-source software. Although access remains limited, AI-assisted vulnerability discovery is expected to significantly reduce the time required to identify vulnerable code, analyze large codebases, and develop proof-of-concept exploits.[14] 

H1 2026 showed that AI is no longer only assisting defenders or improving attacker efficiency. AI platforms, enterprise assistants, coding tools, and model infrastructure have become enterprise assets that require the same security controls, monitoring, and vulnerability management as traditional business applications and cloud services. 

Key Trends Observed 

While the events described above span different technologies and threat actors, several common themes emerged throughout H1 2026. Together, they point to broader shifts in attacker priorities and enterprise risk. 

Vulnerability Exploitation Continues to Drive Initial Access 

Vulnerability exploitation remained one of the primary methods for gaining initial access throughout H1 2026. Campaigns targeting Oracle PeopleSoft, LiteLLM, and Linux kernel vulnerabilities such as Copy Fail and DirtyFrag showed how quickly newly disclosed vulnerabilities progressed from disclosure to operational exploitation. Public proof-of-concept releases, combined with AI-assisted vulnerability discovery, further reduced the time available for organizations to assess exposure and deploy mitigations. 

Identity Replaces the Traditional Network Perimeter 

Identity compromise emerged as the primary path into enterprise environments, replacing traditional endpoint-focused attacks. Campaigns involving ShinyHunters, TeamPCP, FortiBleed, and the Stryker incident relied on compromised identities, administrative accounts, API keys, and authentication tokens rather than malware deployed directly to endpoints. As organizations continued moving critical services to cloud platforms, valid identities increasingly provided access to enterprise applications, sensitive data, and business operations, reducing the need for traditional malware-based intrusion techniques. 

Trusted Platforms Become High-Value Targets 

Trusted platforms increasingly became a preferred target during H1 2026. Software repositories, CI/CD pipelines, developer identities, SaaS platforms, AI agents, coding assistants, and enterprise management systems were repeatedly targeted to obtain downstream access to customers, partners, and enterprise environments. Rather than compromising organizations individually, many campaigns focused on platforms capable of scaling attacks across multiple victims. 

AI Expands the Enterprise Attack Surface 

AI expanded both the enterprise attack surface and the pace of vulnerability discovery during H1 2026. AI infrastructure, coding assistants, enterprise copilots, and model-serving platforms introduced new opportunities for compromise, while AI-assisted vulnerability discovery accelerated exploit development. As organizations integrated AI into enterprise environments, these platforms became part of the same attack surface as cloud services, software supply chains, and enterprise identities. 

Top Attack Indicators Identified  

Top attack indicators are specific pieces of data or observations that strongly point to potential or confirmed malicious activity within a system or network. These can include unusual network traffic patterns, suspicious file executions, unauthorized access attempts, or connection to known malicious IP addresses or domains. 

Here are some of the top threat signals that the CyberProof Threat Research Team observed in breaches and attacks in the first half of 2026:  

Top Attack Vectors: Exploitation of Edge Devices & Public-Facing Vulnerabilities  ClickFix / Fake CAPTCHA (including FileFix & CrashFix)  Identity & Credential-Based Attacks (AiTM, Session Hijacking, Token Theft)  Voice Phishing (Vishing) & Helpdesk Social Engineering  Software Supply Chain Compromise (CI/CD, npm, PyPI, GitHub)	Top Active Ransomware Groups:  Qilin  The Gentlemen  Akira  LockBit (5.0)  DragonForce
Top Malwares:  LummaC2  ACRStealer  Vidar  StealC  XWorm	Top LOLBINs:   PowerShell  Mshta  rundll32  certutil  msiexec
Top RMMs Abused:  ScreenConnect  AnyDesk  NetSupport  Atera  PDQ Connect	Top Attack Tools:  Impacket  PsExec  Mimikatz  Rclone  Cobalt Strike
Top Malicious File Extensions:  .zip  .lnk  .js  .iso  .svg	Top Malicious TLDs (Top Level Domains):  .top  .ru  .xyz . sbs  .es
Top Abused Third-Party Platforms: GitHub  Telegram  Cloudflare (Tunnels / Workers / R2)  Discord  Vercel	Top MITRE IDs:  T1190 (Exploit Public-Facing Application)  T1204.004 (User Execution: Malicious Copy and Paste – ClickFix technique)  T1219 (Remote Access Tools)  T1555.003 (Credentials from Web Browsers)  T1567.002 (Exfiltration to Cloud Storage)
Most Active Intrusion Sets:  ShinyHunters  TeamPCP  Handala  MuddyWater  Scattered Spider

2025 Predictions vs. H1 2026 Threat Landscape Realities 

1. Cybercriminals Will Have a Tactical Advantage with AI: CONFIRMED AND EXCEEDED 

The prediction that cybercriminals would gain a decisive tactical advantage through AI-driven velocity and iteration was confirmed, with the real-world impact outpacing initial expectations. Throughout H1 2026, AI-assisted vulnerability discovery compressed defender reaction windows significantly. For instance, the disclosure of the Copy Fail Linux kernel vulnerability generated over 140 public exploit variants within weeks, many directly derived from AI-assisted modifications. Simultaneously, automated research engines like Project Glasswing / Claude Mythos identified over 10,000 high- and critical-severity vulnerabilities across major operating systems during initial deployments, confirming the immense speed AI brings to attack preparation. However, the threat landscape evolved beyond offensive scripting; AI infrastructure itself—such as Model Context Protocol (MCP) servers and LLM gateways like LiteLLM and OpenClaw—rapidly emerged as a direct, unmonitored attack surface targeted by adversaries to harvest credentials and execute arbitrary code. 

2. The Rise of Vishing, Deepfakes, and Identity Deception: STRONGLY CONFIRMED 

The prediction that identity manipulation would become a major feature of the 2026 threat landscape was strongly reflected in H1 activity. This trend was driven in part by the continued growth of social engineering, including deepfake-assisted vishing and impersonation, alongside credential theft and authentication abuse. Threat actors increasingly targeted valid identities and trusted access mechanisms to gain initial access, reducing their reliance on traditional malware-based intrusion. 

Several high-profile campaigns demonstrated the broader shift toward identity-focused attacks. ShinyHunters exploited Oracle PeopleSoft vulnerabilities, including CVE-2026-35273, in activity affecting more than 100 organizations, while the Klue SaaS compromise involved abuse of OAuth access and the FortiBleed campaign exposed approximately 86,000 administrative credentials. Together, these incidents highlight how compromised identities, credentials and trusted access paths have become increasingly central to intrusion activity. 

3. Cloud Misconfiguration Settings Will Persist: CONFIRMED 

The prediction that multi-cloud complexity and human configuration errors would remain top entry points for enterprise breaches was confirmed. Systemic configuration drift migrated beyond standard cloud storage buckets and virtual machines directly into developer build pipelines and AI integrations. Analysis of over 130,000 GitHub Actions workflows during H1 2026 uncovered more than 250,000 security misconfigurations tied to elevated permissions, missing version pinning, and exposed secrets. Furthermore, unmonitored and exposed Model Context Protocol (MCP) servers alongside exposed administrative endpoints (such as those in the FortiBleed campaign) allowed threat actors to perform silent, large-scale data exfiltration without generating traditional authentication anomalies. These misconfigured development environments were routinely weaponized as distribution engines for self-propagating supply chain malware. 

4. Increasing Use of Regulatory Exposure as Deliberate Leverage: PARTIALLY CONFIRMED 

The prediction that extortion groups would weaponize strict regulatory reporting timelines and legal liability was partially confirmed. Adversaries continued prioritizing pure data exfiltration and SaaS/cloud platform compromise over system encryption to maximize extortion leverage, as seen in ShinyHunters’ compromise of over 300 Oracle PeopleSoft instances specifically to harvest high-value, regulated records. However, pure regulatory coercion was frequently overshadowed by a parallel surge in destructive geopolitical operations. In H1 2026, state-sponsored actors and hacktivist groups regularly bypassed financial extortion altogether, favoring destructive wiper malware—such as Handala erasing data across at least 60 Israeli organizations—or abusing enterprise management software to execute remote device wipes across targeted infrastructure. 

5. The Blurring Lines: Abuse of Legitimate Software: STRONGLY CONFIRMED 

The prediction regarding the widespread weaponization of administrative software and legitimate tooling was strongly confirmed. H1 2026 demonstrated a significant shift from relying on third-party Remote Management and Monitoring (RMM) utilities toward abusing core, built-in enterprise management platforms. In a major incident involving medical technology manufacturer Stryker, attackers compromised a Microsoft Entra ID account and leveraged Microsoft Intune—a trusted enterprise endpoint management tool—to remotely wipe devices across the global enterprise without deploying custom malware. Concurrently, attackers in the TeamPCP and Miasma campaigns systematically subverted software supply chain provenance by abusing trusted code-signing mechanisms, publishing workflows, and maintainer tokens across npm, PyPI, and Visual Studio Code extensions. This tradecraft expanded the definition of software abuse to encompass cloud identity governance, mobile device management, and software release pipelines. 

6. Shadow AI Will Emerge as the Next Unmanaged Risk Surface: CONFIRMED 

The prediction that ungoverned internal AI adoption would introduce critical unmanaged exposure was confirmed throughout H1 2026. Rapid enterprise adoption of AI coding assistants—such as Claude Code, Cursor, and Gemini CLI—created immediate operational risks within development environments. By June 2026, threat actors had transitioned from passive data leakage concerns to deploying active malware that modified local workspace configuration files used by these AI tools to inject malicious commands and harvest secrets. Additionally, security research into exposed Model Context Protocol (MCP) infrastructure uncovered unmonitored clusters containing sensitive enterprise files, firewall configurations, and attack-planning data. Shadow AI rapidly evolved from employee data governance policy failures into direct, active attack chains against AI agents and tool integrations (such as the ClawJacked attack sequence) that exploited inherited trust permissions between local AI agents and enterprise backends. 

What to Watch in H2 2026 

The first half of 2026 showed attackers prioritizing trusted enterprise technologies, cloud identities, software supply chains, and AI platforms to gain scalable access to enterprise environments. As organizations continue expanding these technologies, the following areas are expected to shape the threat landscape during the second half of the year. 

AI Security Will Become an Enterprise Priority 

Enterprise adoption of AI is expected to continue accelerating, expanding the use of AI agents, MCP servers, coding assistants, enterprise copilots, and model-serving platforms. Organizations should expect additional vulnerabilities affecting AI infrastructure, increasing scrutiny of AI supply chains, and growing efforts to abuse trusted AI services and integrations. AI-assisted vulnerability discovery is also expected to further reduce the time between identifying vulnerabilities and developing working exploits. 

Identity Will Continue to Drive Enterprise Risk 

Cloud identities, administrative accounts, API keys, authentication tokens, and developer credentials are expected to remain primary targets for financially motivated and state-sponsored threat actors. As organizations continue migrating critical services to cloud and SaaS platforms, protecting identities and privileged access will remain central to limiting unauthorized access and reducing the impact of credential compromise. 

Continued geopolitical tensions are also expected to sustain state-sponsored and hacktivist activity, with organizations outside the immediate conflict regions increasingly exposed through supply chains, shared technology providers, and regional business relationships. 

Trusted Platforms Will Remain Attractive Targets 

Software repositories, CI/CD pipelines, enterprise management platforms, SaaS applications, and AI development environments have become high-value targets because they provide opportunities to compromise multiple downstream organizations through a single intrusion. Organizations should expect continued attempts to abuse trusted software, third-party integrations, and enterprise platforms as a means of scaling attacks. 

Faster Exploitation Will Continue to Compress Response Timelines 

Public proof-of-concept releases, AI-assisted vulnerability discovery, and rapid sharing of exploitation techniques continue to reduce the time between vulnerability disclosure and operational exploitation. As exploit development accelerates, organizations will have less time to assess exposure, prioritize remediation, and validate mitigations before vulnerabilities are incorporated into real-world attack activity. 

Geopolitical Cyber Activity Will Continue to Shape the Threat Landscape 

Geopolitical tensions are expected to remain a significant driver of cyber activity throughout H2 2026. State-sponsored actors and hacktivist groups are likely to continue targeting government, critical infrastructure, defense, telecommunications, and organizations perceived to support opposing interests. As demonstrated during H1 2026, organizations outside the immediate conflict regions may also face increased exposure through supply chains, shared technology providers, and regional business relationships. 

Final Analysis 

H1 2026 was defined less by entirely new attack techniques than by changes in how attackers gained access and scaled their operations. Cloud identities, software supply chains, AI platforms, and trusted enterprise technologies became high-value targets because they provided broad access to enterprise environments through a single compromise. At the same time, AI-assisted vulnerability discovery, publicly available proof-of-concept exploits, and attacks against trusted platforms further reduced the time between vulnerability disclosure and operational exploitation. 

As organizations continue expanding their use of cloud services, AI technologies, and interconnected software ecosystems, these platforms will require the same level of visibility, monitoring, and security oversight traditionally applied to endpoints and network infrastructure. The H1 2026 analysis shows that securing cloud identities, trusted platforms, software supply chains, and AI infrastructure is no longer a specialized challenge but part of day-to-day enterprise security.

About CyberProof 

CyberProof delivers threat-led, co-managed security operations with the belief that better security is achieved through the right partnerships, technology and client experiences. Our threat-led, cloud-first, and AI-powered approach to security, delivers industry-leading security services which drives real and measurable business outcomes. We believe that working closely with our clients and partners through a better security, together model, jointly empowers us to defend against the greatest of threats.  

To learn more visit www.CyberProof.com.