LSN โ€” Local Sovereign Network

Microsoft carries more malicious traffic than any bulletproof host. Google enables more attacks than any darknet marketplace. These aren't accusations โ€” they're routing tables.

The roads the attackers travel were built by the same companies that sell you antivirus.

The Transit Hierarchy

9,972 tracked malicious IPs. 11,064 shared BGP upstream relationships. Every attack packet travels a path โ€” and those paths converge on a surprisingly small number of transit providers.

Top 15 Autonomous Systems Hosting Malicious Infrastructure

ASNOrganizationMalicious IPsAvg ThreatType
AS8075Microsoft Corporation49741.5Hyperscaler
AS14061DigitalOcean35938.5VPS
AS396982Google Cloud Platform21839.7Hyperscaler
AS132203Tencent Cloud21755.1Chinese Cloud
AS60729TorServers.net14248.6Anonymization
AS135377UCloud HK14032.1Chinese Cloud
AS16276OVH SAS13355.6Hosting
AS401401Unredacted/NOISENET12322.2Anonymization
AS53667FranTech/PONYNET12041.2Bulletproof-tolerant
AS4766Korea Telecom11938.3National ISP
AS396507Emerald Onion9728.8Anonymization
AS4134ChinaNet Backbone9627.4National Backbone
AS24940Hetzner Online9340.7Hosting
AS4818DiGi Telecommunications9252.4Mobile ISP
AS51167Contabo8842.2Budget VPS

The top three โ€” Microsoft, DigitalOcean, Google โ€” account for 1,074 malicious IPs. More than 10% of all tracked threats route through infrastructure owned by companies with dedicated security teams, abuse departments, and public commitments to internet safety.

The Legitimacy Paradox

Microsoft's AS8075 hosts 497 malicious IPs with an average threat score of 41.5. This isn't Microsoft being complicit โ€” it's Microsoft being useful. Azure's global infrastructure, instant provisioning, and pay-as-you-go model serves legitimate customers and attackers with equal efficiency.

The economics are clear:

  • Hyperscalers (Microsoft, Google, DigitalOcean): 1,074 IPs. Low-to-moderate threat scores (38-42). Attackers use these for scanning, C2, and credential operations. Accounts are disposable โ€” created with stolen cards, used briefly, abandoned.
  • Chinese Clouds (Tencent, UCloud, ChinaNet): 553 IPs. Higher threat on Tencent (55.1). Different enforcement jurisdiction. Abuse reports go unanswered across national boundaries.
  • Anonymization Networks (TorServers, Unredacted, Emerald Onion): 362 IPs. Lower average threat (22-48). These carry everything โ€” legitimate privacy traffic alongside attacks. By design, they cannot distinguish.
  • Bulletproof-tolerant (FranTech, ColoCrossing): 197 IPs. Moderate threat. Known for slow abuse response. Not technically bulletproof, but the next best thing for an attacker on a budget.

The Geography of Transit

Countries by Malicious IP Concentration

CountryMalicious IPsAvg ThreatCharacter
United States2,10239.6Volume leader โ€” hyperscaler home
China58121.6Low threat per IP โ€” residential/IoT
Germany55043.5Hosting hub (Hetzner, Contabo)
Netherlands44250.6Highest avg threat โ€” hosting concentration
Singapore28547.5Asian cloud hub (BytePlus, UCloud)
France28248.8OVH headquarters
United Kingdom26033.2Mixed ISP/cloud
India25639.1Residential compromise
Vietnam25540.6National ISP (Viettel)
Hong Kong23335.9Chinese cloud overflow

The United States hosts 21% of all malicious IPs โ€” not because Americans are more criminal, but because American infrastructure is more accessible. Azure, AWS, GCP, DigitalOcean โ€” the tools that democratized cloud computing also democratized cloud-based attack infrastructure.

The Netherlands punches above its weight: 442 IPs with the highest average threat score (50.6) of any major country. Amsterdam is Europe's peering hub โ€” data center density creates hosting options that attract both CDN customers and botnet operators.

China's paradox: 581 IPs but the lowest average threat (21.6). These are overwhelmingly compromised residential and IoT devices, not rented infrastructure. They participate in botnets unknowingly. The Great Firewall keeps abuse reports out and prevents rapid takedown โ€” creating persistent, low-threat zombie infrastructure.

The BGP Upstream Convergence

11,064 shared BGP upstream relationships map the transit layer โ€” who carries packets for whom. The top upstream clusters (20+ IPs sharing the same transit provider):

Upstream ASNKnown IdentityCluster SizeRole
AS2914NTT Communications20Tier-1 global transit
AS3257GTT Communications20Tier-1 global transit
AS9009M247 Ltd20European transit/hosting
AS2497IIJ (Internet Initiative Japan)20Japanese backbone
AS34019Hivane20French transit
AS15802PacketHub / DuoDecad20Transit reseller
AS50629LWN (Leibnitz)20Research network
AS39351Esab / 31173 Services20Dutch hosting

Tier-1 providers (NTT, GTT) appear because they carry everything โ€” malicious and legitimate alike. Their presence in the graph is structural, not indicative. But below Tier-1, the clustering becomes informative: M247, Hivane, PacketHub โ€” these are transit resellers that serve hosting companies which serve attackers. The supply chain of connectivity mirrors the supply chain of malware.

The Abuse Contact Theater

11,324 shared abuse phone relationships. When thousands of malicious IPs share the same contact phone number, the contact exists for compliance theater rather than actual abuse handling:

Largest Abuse Contact Clusters

PhoneOrganizationCluster SizeASN Spread
+603-57216174DiGi Telecommunications (MY)391
+971 45546487RIPE NCC361
+1-877-843-7627Internap / SingleHop361
+1 510 473 2303Zwiebelfreunde (Tor)351
+1-206-739-3390Emerald Onion (Tor)341
+33972101007OVH SAS341
+31 20 535 4444RIPE NCC (registry!)3012
+1-626-549-2801Psychz Networks311
+1-702-728-8933BuyVM / FranTech291
tel:+248-4-610-795Cloud Innovation (Seychelles)2517

Two entries stand out:

RIPE NCC (+31 20 535 4444): 30 malicious IPs, spanning 12 different ASNs. This is RIPE's general registry number โ€” not an abuse desk. When attackers register IP space and list the regional internet registry as their abuse contact, they're exploiting a system designed for good faith.

Cloud Innovation, Seychelles (+248-4-610-795): 25 IPs spanning 17 ASNs. A Seychelles-registered entity whose phone number appears as abuse contact across seventeen different autonomous systems. This is bulletproof hosting hiding behind a tropical jurisdiction โ€” previously documented in TI-2026-057F as a registration fraud pattern.

The RDAP Registration Layer

11,411 shared RDAP organization links reveal who registers the IP space:

RDAP OrganizationClusterCharacter
IRT-VNNIC-AP (Vietnam NIC)40National registry (Vietnam)
DiGi Telecommunications39Malaysian mobile ISP
RIPE-NCC-END-MNT34European registry marker
Psychz Networks27LA-based hosting
ZWIEBELFREUNDE27German Tor nonprofit
"Private Customer"26Anonymous registration
Internap Holding LLC25US hosting/colo
The Shadowserver Foundation23Security research (sinkholed)
ACEVILLE PTE LTD22Singapore shell company
Cloud Innovation Support21Seychelles bulletproof

"Private Customer" โ€” 26 malicious IPs registered to a completely anonymous entity. No organization name, no contact beyond the mandatory minimum. This is the WHOIS privacy equivalent for IP space โ€” technically compliant, operationally opaque.

Shadowserver Foundation (23 IPs): Not malicious โ€” these are sinkholed domains and honeypot infrastructure operated by the security research community. Their presence in the malicious graph is a measurement artifact: they attract traffic from infected machines.

ACEVILLE PTE LTD (22 IPs): A Singapore-registered entity operating across multiple Chinese cloud providers. Previously identified in the cross-provider convergence documented in TI-2026-031H.

Read Between the Lines

The roads aren't criminal. The roads are infrastructure. NTT doesn't know which packets carry botnet C2 commands. Microsoft can't distinguish a legitimate Azure VM from one running XMRig. DigitalOcean processes stolen credit card signups alongside genuine startups.

The real finding isn't that legitimate providers carry malicious traffic โ€” that's obvious and unavoidable. The finding is the concentration: three companies (Microsoft, DigitalOcean, Google) carry 10% of observed malicious infrastructure. Fifteen ASNs carry over 3,000 IPs. The internet's attack surface is narrower than it appears.

This creates a paradox: the same concentration that enables mass attacks also enables mass defense. If three providers implemented aggressive behavioral detection, 10% of attack infrastructure would evaporate overnight. But those providers also serve millions of legitimate customers โ€” and false positives cost money, reputation, and legal exposure.

The attackers understand this calculus better than the defenders. They hide in the noise of legitimate traffic specifically because they know the providers won't โ€” can't โ€” implement detection aggressive enough to catch them without catching innocents.

The roads were built for commerce. The criminals use them because everyone else does too.

The Thread Forward

In TI-2026-053B, we move from who carries the traffic to who registers the space. RDAP organization mapping (11,411 links) reveals the shell companies, resellers, and opacity layers between the attacker and the accountability framework that was supposed to prevent exactly this.

โš  Personal capacity. Research published independently โ€” not reflecting employer views. Derived from passive observation of attacks against personal infrastructure. Full disclaimer โ†’
โ† Previous The Transit Map โ€” 1 / 7 Next โ†’