Gen H1 2026 Threat Report Highlights Rising Scams and Abuse of Trusted Workflows
Original report publisher: not verified. Report discovered via Awesome Annual Security Reports.
Summary
Gen’s H1 2026 report describes attackers exploiting trusted sessions, payment flows, developer tools and AI agents. Its telemetry recorded 114.2 million blocked fake-store attacks and a 212% rise in web-skimming blocks.
Key points
- Gen blocked 114.2 million fake-store attacks, up 109% from H2 2025, and 5.26 million fake-tutorial attacks that tricked users into following malicious instructions, up 193%.
- Tech support scam blocks reached 20.3 million, up 61.6%; Gen says expanded detection coverage contributed to the increase, alongside campaigns abusing familiar cloud infrastructure and brands.
- An analysis of 14.57 million ads across the EU and UK over 23 days identified 4.51 million scam-related ads, generating 304.11 million impressions.
- GhostPairing tricks WhatsApp users into approving an attacker-controlled linked device, creating an authorized session that can expose conversations and remain active until removed.
- Supply-chain activity included hijacked package-maintainer accounts and publishing tokens; cracked macOS software campaigns also encouraged users to disable protections or grant extensive permissions.
- The report identifies AI-agent tool execution, excessive permissions, indirect prompt injection and persistent memory as security risks, emphasizing runtime controls at access and execution boundaries.
- Gen recorded 18,618 breach events affecting its customers and sent 3.3 million source-attributed breach alerts. It cautions that previously stolen, repackaged data can drive later telemetry spikes.
Article Details
- Publisher
- Gen
- Report Period
- 2026-01-01 to 2026-06-30
- Scope
- Gen consumer threat, privacy, identity and financial-alert telemetry, supplemented by threat research and third-party reporting.
- Sample Size
- Overall telemetry population not disclosed. Scam Ad Machine research analyzed 14.57 million ads representing 10.76 billion impressions across the EU and UK over 23 days.
- Key Statistics
- Gen blocked 20.3 million tech support scam attacks in H1 2026, up 61.6% from H2 2025; expanded URL-filtering coverage contributed to the increase.
- Gen blocked 114.2 million e-shop scam attacks in H1 2026, up 109% from H2 2025.
- Scam Ad Machine research identified 4.51 million scam-related ads, nearly one-third of the analyzed dataset, generating 304.11 million impressions across the EU and UK over 23 days.
- Gen identified 18,618 breach events affecting customers in H1 2026, up 94.5% from H2 2025. Figures may change as breach intelligence is enriched or reclassified.
- Gen blocked 996.3K web skimming attacks in H1 2026, up 212% from H2 2025.
- Recommendations
- Make persistent account access visible and revocable, including auditing and removing unauthorized linked messaging devices.
- Scope AI agent permissions and enforce runtime decisions before tool use, package installation, file access, execution and network connections.
- Verify AI agents and skills before use and monitor their behavior during execution.
- Continue financial fraud monitoring after onboarding through behavioral intelligence, account ownership checks and detection of coordinated activity.
- Monitor exposed identity data and downstream credit inquiries, account activity and new applications rather than waiting for confirmed financial loss.
MITRE ATT&CK
T1027 · Obfuscated Files or InformationRAT delivery used per-victim JavaScript obfuscation, and the coinminer investigation identified reused obfuscation techniques.T1027.003 · SteganographyFake invoices in Poland delivered a steganographic .NET loader that installed Remcos RAT.T1059.001 · PowerShellLocalized RAT infection chains used multi-stage PowerShell to deliver malware, including Babylon RAT.T1059.007 · JavaScriptCampaigns used JavaScript droppers and obfuscated JavaScript delivery stages.T1078 · Valid AccountsCompromised corporate mailboxes, developer accounts and GitHub accounts were used to distribute malicious content through trusted channels.T1115 · Clipboard DataA Rust-compiled cryptocurrency clipboard hijacker monitored copied wallet addresses before replacing them with attacker-controlled addresses.T1176 · Software ExtensionsBanking trojan payloads installed browser add-ons to intercept banking sessions.T1195.001 · Compromise Software Dependencies and Development ToolsAttackers compromised npm and PyPI packages and hijacked maintainer accounts to publish malicious versions.T1204.002 · Malicious FileMalware distribution relied on users opening malicious documents or launching repacked applications presented as cracked software.T1496 · Resource HijackingThe January telemetry spike was ultimately classified as coinminer activity; cracked macOS applications also distributed cryptominers.T1539 · Steal Web Session CookieRemus collected browser cookies and session-related data that could support account takeover.T1555.003 · Credentials from Web BrowsersRemus targeted browser passwords and other stored browser secrets.T1562.001 · Disable or Modify ToolsCracked-software installation guides instructed macOS users to disable Gatekeeper and turn off System Integrity Protection.T1566.001 · Spearphishing AttachmentBanking trojan and RAT delivery used email attachments, including JavaScript droppers and fake invoice PDFs.
People
Malware
Babylon RATIn Czechia, a compromised third-party domain served multi-stage PowerShell that installed Babylon RAT.Lumma StealerGen Threat Labs identified Remus, a new 64-bit infostealer attributed to the Lumma Stealer family.Remcos RATIn Poland, fake invoices delivered a steganographic .NET loader that installed Remcos RAT.RemusGen Threat Labs identified Remus, a new 64-bit infostealer attributed to the Lumma Stealer family.
Vendors
AnthropicThe Anthropic Mythos and Fable 5 sequence exposed a simple problem: once a frontier model can materially improve cyber work, the model itself becomes part of the security problem.GenOrganization: GenMicrosoftOther attacks used fake PayPal receipt PDFs, including Microsoft 365-themed lures, to create a billing dispute that victims were told to resolve by calling fraudulent support.Vercelinvoice PDFs, including Booking.com-themed lures, showed realistic Italian error dialogs, then delivered Vercel-hosted scripts with per-victim JavaScript obfuscation before moving to Blogspot-hosted PowerShell stages.xAIGen’s partnerships with xAI and Microsoft point to the same conclusion from another angle: trusted AI is becoming part of mainstream consumer workflows, not a lab problem.
Products
Agent Trust HubThe Agent Trust Hub addresses the pre-use layer, helping establish trust before an agent or skill is used through static identity and verification checks.AndroidThese campaigns mostly reached Android users through SMS, and one SMS campaign accounted for 39% of all blocked family scam attacks in the period.Claude Mythos PreviewAnthropic said Mythos Preview could identify and exploit vulnerabilities in major operating systems and browsers when directed to do so.Cline CLIAI developer tooling was also affected: in one case, a compromised npm publishing token was used to push an unauthorized Cline CLI update that installed OpenClaw on developer systems during an eight-hour window.Fable 5The Anthropic Mythos and Fable 5 sequence exposed a simple problem: once a frontier model can materially improve cyber work, the model itself becomes part of the security problem.macOSGen Threat Labs saw a related version of the same trust problem on macOS.Microsoft WindowsWindows users accounted for 92% of blocked tech support scam attacks, showing that this platform is particularly targeted.Mythos 5After launch, Anthropic said it received a US export-control directive requiring it to suspend access to Fable 5 and Mythos 5 for foreign nationals, including foreign-national employees inside the United States.Norton 360Gen announced VPN for Agents through the Agent Trust Hub and expanded Norton AI Agent Protection inside Norton 360, including protections for plugins, skills and tools, prompt injection defense, and code and fileNorton AI Agent ProtectionGen announced VPN for Agents through the Agent Trust Hub and expanded Norton AI Agent Protection inside Norton 360, including protections for plugins, skills and tools, prompt injection defense, and code and fileOpenClawAI developer tooling was also affected: in one case, a compromised npm publishing token was used to push an unauthorized Cline CLI update that installed OpenClaw on developer systems during an eight-hour window.WhatsAppA WhatsApp account takeover worked by turning the attacker’s browser into an approved linked device.
Tools
Countries
AustraliaIn Australia and the US, some campaigns reached victims through Facebook advertising.BrazilIn Brazil, fake payment PDFs impersonated ecommerce and banking brands to target the Boleto payment ecosystem.CzechiaBanking trojan campaigns in Czechia, Slovakia and Poland used email-based JavaScript droppers that impersonated normal business correspondence, such as shipment notices, scanned document notifications or invoice-relatedFranceFrance and Germany together slightly surpassed the US in raw attack count, with 2.54 million and 2.3 million blocked attacks, respectively.GermanyFrance and Germany together slightly surpassed the US in raw attack count, with 2.54 million and 2.3 million blocked attacks, respectively.IrelandThe activity was concentrated in Western Europe, led by the Netherlands, France, Ireland and Germany.ItalyWestern Europe was also heavily targeted, with the UK, Germany, France, Italy and Spain together accounting for 30.9 million blocked attacks.JapanJapan also stood out, with nearly 2.2 million blocked attacks.NetherlandsThe activity was concentrated in Western Europe, led by the Netherlands, France, Ireland and Germany.PolandBanking trojan campaigns in Czechia, Slovakia and Poland used email-based JavaScript droppers that impersonated normal business correspondence, such as shipment notices, scanned document notifications or invoice-relatedSlovakiaBanking trojan campaigns in Czechia, Slovakia and Poland used email-based JavaScript droppers that impersonated normal business correspondence, such as shipment notices, scanned document notifications or invoice-relatedSpainWestern Europe was also heavily targeted, with the UK, Germany, France, Italy and Spain together accounting for 30.9 million blocked attacks.United KingdomGen’s Scam Ad Machine research found that scam-related ads made up nearly 1 in 3 ads analyzed from Meta’s EU and UK ad dataset, generating more than 304 million impressions in less than a month.United StatesThe United States remained the largest single target, with 4.67 million blocked attacks.