Intentionally desktop-first — best experienced on a workstation
Portfolio
Threat Analysis · Trusted Infrastructure Phishing · Platform Abuse

The Trust Tax —
How Attackers Stopped Building Infrastructure and Started Borrowing Ours

Author
Yana Ivanov
Published
August 2026
Classification
Public — Educational
Attack Class
Trusted Infrastructure Phishing (TIP)
Threat Actors
Multiple · Nation-State · Criminal · PhaaS
Framework
Trusted Infrastructure Phishing (TIP)
8 trusted platforms weaponized · github · azure · framer · figma · jira · hugging face · screenconnect · cloudflare workers
Section 01

The Infrastructure Problem Nobody Is Naming

Building malicious infrastructure is expensive and risky. You register a domain — it gets flagged. You stand up a server — it gets blacklisted. You buy an SSL certificate — it gets revoked. Every piece of custom infrastructure you build as an attacker leaves a trail and has a shelf life measured in days.

Attackers solved this problem. They stopped building their own infrastructure. They started borrowing ours.

The shift is structural, not tactical. Across eight different platforms in the last eighteen months, the same pattern appears: take a legitimate, widely trusted service that developers, designers, or business users depend on daily, and use it exactly as designed — to host files, send notifications, deliver invitations, or provide remote access. The platform's trust becomes the attack surface. The victim never lands on a suspicious domain. The email never comes from an unknown sender. The link never goes somewhere flagged by reputation tools. Everything looks clean because it is clean — the infrastructure is legitimate. Only the intent is not.

This analysis documents eight confirmed instances of what I'm calling Trusted Infrastructure Phishing — a framework that names what is actually happening across campaigns that are currently being treated as unrelated incidents. They are not unrelated. They are the same attack, executed across different platforms.

The Trust Tax: Organizations pay a trust tax every time they allow a third-party platform — a file host, a project management tool, a design platform, a remote access tool — into their environment. That trust is earned by the platform through years of legitimate use. Attackers borrow it for free. The tax is paid by defenders who cannot block the platform without breaking legitimate workflows, and by victims who have no way to distinguish a legitimate notification from a weaponized one.

8
Trusted Platforms
Confirmed Weaponized
GitHub, Azure Blob, Framer, Figma, Jira, Hugging Face, ScreenConnect, Cloudflare Workers. Each documented with confirmed campaigns in 2025-2026.
$0
Infrastructure Cost
to Attacker
Every platform on this list offers a free tier. Free accounts, free hosting, free email delivery. The attacker's infrastructure cost is zero.
100%
Pass Rate
Reputation Checks
Links to github.com, blob.core.windows.net, and framer.com pass every URL reputation check. The domains are trusted by every security tool.
Defender's
Dilemma
You cannot block GitHub, Microsoft Azure, or Figma without breaking your organization's legitimate workflows. The platform's value is the attacker's shield.
Section 02

The Eight Platforms — A Taxonomy

These are not eight separate campaigns. They are eight instances of the same attack philosophy applied to different platforms at different layers of the attack chain. Some platforms are used to send the email. Some are used to host the phishing page. Some are used to stage the malware payload. Some are used as command and control infrastructure. The platform layer varies. The trust exploitation does not.

Platform · LayerHow Trust Is ExploitedConfirmed CampaignDetection
GitHub
Payload staging · C2
Malware hosted in GitHub releases. Attackers create repos impersonating trusted brands. Domain reputation: trusted. HTTPS: valid. Content: malware. kaswareteam org (17 repos, ScreenConnect delivery). Shapecluneedle/jubilant-engine (Vidar + SheetRAT). Kimsuky LNK delivery via raw.githubusercontent.com. Partial
Azure Blob Storage
Phishing page hosting
HTML credential harvesting pages uploaded to public blob containers. Sits on blob.core.windows.net — a Microsoft domain with a Microsoft SSL certificate. Security tools see legitimate Microsoft infrastructure. Ongoing campaigns impersonating Microsoft 365, OneDrive, Entra ID. Confirmed by Microsoft Threat Intelligence, Nopal Cyber, CYFIRMA (2025-2026). Gap
Cloudflare Workers
PhaaS infrastructure · Evasion
EvilTokens PhaaS serves AES-GCM encrypted phishing pages via Cloudflare Workers. Keys rotate on every request, defeating sandbox replay. Cloudflare domain passes every perimeter control. EvilTokens — M365 device code phishing, OAuth token theft. Confirmed by Sekoia TDR, Abnormal AI (2026). Partial
Hugging Face
Payload staging · Malware hosting
Malicious model repos published on the AI community's primary hub. Trusted implicitly by developers loading models as part of normal workflow. 244K downloads of one malicious repo in 18 hours. Open-OSS/privacy-filter typosquatting (May 2026). North Korea Lazarus Group payload staging. Android RAT via TrustBastion (Jan 2026). FaceHugger CVEs (Aug 2026). Partial
Framer
Phishing page hosting
Credential harvesting pages built on Framer's free platform. Served from framer.com — SPF, DKIM, and DMARC all pass. Legitimate sending domain, malicious content. Active campaigns using Framer-hosted phishing pages targeting Microsoft 365 credentials. Confirmed in Sublime Security telemetry (2026). Partial
Figma
Email delivery · Invite abuse
Figma's invite notification system delivers phishing email from no-reply@email.figma.com. Passes SPF, DKIM, and DMARC. Invite links to a malicious Figma file. Figma invite phishing targeting first-contact recipients. Figma plugin ecosystem abuse via CVE-2025-53967 (MCP RCE). Documented in field notes (2026). Partial
Jira Service Management
Email delivery
Free Jira projects with phishing content in the welcome message. Invite Customers delivers email from Atlassian's own infrastructure. Passes every authentication check. Documented by Cisco Talos (April 2026). Billing lure, invoice phishing, callback scam content delivered via Atlassian-signed email. Partial
ScreenConnect
C2 · Remote access
Free *.screenconnect.com subdomains with no verification. Attackers register them as C2 infrastructure. Domain passes every reputation check — it's a legitimate enterprise platform. TELEPUZ ClickFix campaign. SMOKE#SCREEN fake Adobe/Zoom updates. kaswareteam GitHub delivery. Multiple active subdomains confirmed via URLhaus (Aug 2026). Partial
Section 03

Why It Works — The Three Laws of Trusted Infrastructure Phishing

These eight campaigns share three structural properties that explain why the technique works and why it is spreading. Each property is a feature of legitimate platform design that becomes a vulnerability when exploited.

1
Law 1 — The Domain Is the Defense
Email security tools, web proxies, and endpoint controls make decisions based on domain reputation. github.com, blob.core.windows.net, framer.com, and figma.com all have excellent reputation scores built over years of legitimate use. When an attacker hosts malicious content on these platforms, the domain reputation protects them the same way it protects legitimate users. There is no technical distinction between a legitimate GitHub release and a malicious one at the domain reputation layer — both resolve to github.com. The security tool's domain-based decision cannot catch the difference.
CRITICAL
2
Law 2 — Free Tiers Fund the Attack
Every platform on this list offers a free tier with enough capability to run a phishing campaign. A free GitHub account can host unlimited public repositories. A free Framer account can publish a phishing page. A free ConnectWise account gets a *.screenconnect.com subdomain. A free Jira Service Management project can invite unlimited customers. A free Hugging Face account can publish unlimited models. The attacker's infrastructure cost is zero — and the platform's economics require offering free tiers to grow their user base, creating an inherent tension between growth and security. Restricting free tiers would disadvantage the platform competitively. So the free tier stays, and attackers use it.
CRITICAL
3
Law 3 — Blocking Is Breaking
A defender who identifies GitHub as a malware delivery vector cannot block github.com without breaking every developer's workflow. A defender who identifies Figma as a phishing vector cannot block figma.com without breaking design handoffs. A defender who identifies Azure Blob Storage as a phishing host cannot block blob.core.windows.net without breaking Microsoft's own services. The attacker has positioned themselves inside the defender's blast radius — any block that hits the attacker also hits the organization's legitimate users. This is the Trust Tax. The defender must pay it in the form of precision detection work rather than simple domain blocking, because the simple answer breaks the business.
HIGH
Section 04

The Detection Gap — What Exists and What Doesn't

Detection for Trusted Infrastructure Phishing cannot be domain-based. The domain is the problem — it's trusted. Detection has to be behavioral: who is sending the email, what is the relationship between sender and recipient, what signals inside the email suggest the platform is being abused rather than legitimately used.

For each platform in the taxonomy, I have either written a detection rule for Sublime Security's open-source ruleset or identified the gap. The common pattern across all eight: the existing rules are too narrow, catching only the most obvious abuse while missing the wider campaign.

What Detection Exists

GitHub: LNK/ZIP GitHub raw C2 rule — targets Kimsuky-specific delivery pattern via raw.githubusercontent.com

Cloudflare: Six rules covering Cloudflare Workers, tunnels, and service abuse

Firebase: Two rules covering Firebase sender abuse and redirect patterns

Azure Blob: One spam rule — new domains only, single non-org recipient

Framer, Figma, Jira, Hugging Face, ScreenConnect: Rules written and submitted or in pipeline

What's Missing

Azure Blob — body link: No rule for blob.core.windows.net links in email body from unsolicited senders with credential theft signals

AWS S3 — body link: S3 link in PDF attachment covered; email body link not covered

Firebase Storage: firebasestorage.googleapis.com not covered separately from Firebase sender rules

Cross-platform TIP detection: No rule that identifies the pattern across platforms — a rule that catches "trusted platform + first-contact sender + credential theft signal" regardless of which platform

The hardest gap to fill: A cross-platform TIP detection rule — one that catches the pattern regardless of which trusted platform is being abused — would require matching a large list of trusted hosting domains against first-contact sender and credential theft signals. The FP risk is high because legitimate use of these platforms is common. But the signal combination of trusted-platform-hosted link + first-contact sender + NLU credential theft intent is genuinely distinctive. This is the detection architecture that would close the gap across all eight platforms simultaneously.

Section 05

Key Findings

1
Trusted Infrastructure Phishing Is a Category, Not a Campaign
The eight platforms documented here are treated in security reporting as eight separate issues. They are not. They share the same attack philosophy, the same structural properties, and the same detection gap. Naming the category matters because it changes how defenders think about the problem — from "block this specific domain" to "build behavioral detection that works regardless of which trusted platform the attacker chooses next." The attacker's next platform is already out there. It's trusted, it's free, and it accepts signups without identity verification.
CRITICAL
2
The Free Tier Is the Attack Vector
Every platform on this list offers a free tier with enough capability to run a phishing campaign at scale. Platform security teams are aware of this and run abuse detection — but the detection is reactive, not preventive. A malicious Hugging Face repo runs for 18 hours and accumulates 244,000 downloads before removal. A kaswareteam GitHub org runs 17 repos before GitHub takes it down. The abuse is detected and removed, but the damage happens during the window between deployment and takedown. Prevention requires identity verification for free account creation — a business model decision most platforms are not willing to make.
CRITICAL
3
Email Is Still the Last Reliable Interception Point
For six of the eight platforms documented here, the attack chain includes an email delivery step — an invitation, a notification, a lure, or a link. That email is the last point where the attack can be interrupted before the victim clicks. Once the victim clicks a link to a trusted platform, domain-based controls offer no protection. The email layer, with behavioral detection on sender history, recipient relationship, and content signals, is where Trusted Infrastructure Phishing can be stopped before it reaches the browser.
HIGH
4
The Next Platform Is Already Trusted
The eight platforms documented here are not a complete list — they are the confirmed cases as of September 2026. The attack philosophy requires no novel technique. Any platform that offers free hosting, free email delivery, or free remote access with a trusted domain is a candidate. Notion, Airtable, Linear, Loom, Canva, and dozens of other widely used platforms have the same structural properties. The attacker's next campaign is a free account signup away.
HIGH
Section 06

What Defenders Can Do

ActionWhyPriority
Build behavioral detection, not domain blocklists Domain blocking cannot work when the domain is legitimate. Detection must be based on sender history, recipient relationship, and content signals — not whether the link goes to GitHub or Azure. IMMEDIATE
Flag first-contact emails containing trusted-platform links The combination of a first-contact sender and a link to a trusted hosting platform (GitHub releases, blob.core.windows.net, framer.com) with credential theft language is the core TIP signal. Flag for review rather than deliver silently. IMMEDIATE
Audit your platform exposure — which trusted services does your organization use? If your organization uses Figma, attackers can deliver phishing via Figma invites. If your developers use Hugging Face, attackers can deliver malware via model sharing. Map your trusted platform dependencies and apply appropriate controls for each. HIGH
Train users on trusted-platform lures specifically Standard security awareness training focuses on suspicious links and unknown senders. TIP attacks use trusted links and often known platforms. Training needs to cover "why does this Figma invite from someone I've never worked with contain a billing alert?" HIGH
Use a password manager and treat failed auto-fill as a red flag Password managers bind credentials to specific domains. A credential harvesting page hosted on blob.core.windows.net will not auto-fill Microsoft credentials — because the domain doesn't match login.microsoftonline.com. If your password manager doesn't offer to fill on a login page, stop and verify the URL. HIGH
Report platform abuse directly to the platform GitHub, Microsoft, Atlassian, Figma, and Framer all have abuse reporting mechanisms. A reported malicious repo or page gets taken down within hours in most cases. Reporting reduces the takedown window during which victims can be compromised. MEDIUM

This analysis synthesizes original research published across this portfolio between March and September 2026, including analyses of ScreenConnect RMM abuse, Hugging Face supply chain attacks, Figma platform abuse (Parts 1 and 2), Jira Service Management PaaP delivery, and kaswareteam GitHub infrastructure. Campaign data sourced from URLhaus and MalwareBazaar via ArgusX threat intelligence platform. Third-party sources include Microsoft Security Blog, Cisco Talos, Sekoia TDR, CYFIRMA, Nopal Cyber, HiddenLayer, and Bitdefender. The "Trusted Infrastructure Phishing" framework and "Trust Tax" framing represent the author's original analytical contribution. Detection rules referenced are available in Sublime Security's open-source ruleset and at the author's GitHub. This analysis represents independent research produced for educational purposes.

YI
Yana Ivanov
Security Analyst  ·  Threat Intelligence  ·  Detection Engineering

I'm a security researcher in Connecticut. Analysis is the part I love: tracing threat actor behavior, pulling apart supply chain attacks, and following evidence even when it lands on "unknown." I co-founded ArgusX, a live threat intelligence platform. Before security I spent 15 years designing enterprise software. I contribute detection rules to Sublime Security's open-source production ruleset. Security+ in progress. Everything here is independent work, shared as a contribution to the security community.

Portfolio