.png)
Between June 17 and October 15, 2026, infrastructure that Augur predicted as malicious appeared in 121 public threat reports. Those reports came from 54 publishers, including Unit 42, Microsoft, CrowdStrike, ESET, SentinelOne, Kaspersky's Securelist, and three FBI IC3 advisories.
Key Findings
- Augur predictions surfaced in public reporting about seven times a week. Across 121 reports, 243 of 1,245 published IOCs (19.5%) were predicted by Augur.
- The lead can be months, even on 2026 campaigns. Augur flagged the Command and Control (C2) infrastructure used in the GhostAction GitHub Actions credential theft 163 days before first activity, and the N26 vishing infrastructure behind the Copybara Android RAT 118 days before.
- Software supply chain attacks were the clearest win. In 14 of 15 supply chain reports with an active prediction, Augur's prediction preceded the attack, covering npm and NuGet typosquats, trojanized browser and IDE extensions, and malicious GitHub workflows.
- Attackers reuse hosting, and Augur scores hosting. A single predicted block, 91.92.240.0/22, appeared in ten unrelated reports spanning ClickFix delivery, AI agent supply chain abuse, BYOVD tooling, and crypting services.
By the Numbers
Augur issues predictions on IP addresses and network ranges, each with a validity window and an importance score. We matched those predictions against the IOCs published in public threat reporting over 17 weeks.
Lead time is measured from Augur's earliest prediction on any IOC in a report to the most recent date that report's infrastructure was first observed active. Reports whose predictions had expired before the attack are excluded from the lead figures.
Prediction rates are highest in focused reports. Augur predicts IP infrastructure, so reports that list mainly C2 and delivery servers match best; large reports add domains, hashes, and short-lived infrastructure.
The Threat Landscape Through Augur's Lens
Augur performed best against software supply chain attacks and ClickFix-style social engineering, both of which rely on small, reused sets of servers. Reports can fall under more than one theme.
Software supply chain. Attackers seeded malicious code through npm and NuGet typosquats, trojanized Open VSX and browser extensions, hijacked npm packages (keyv and cacheable), and credential-stealing GitHub Actions workflows. The C2 behind these campaigns sat on infrastructure Augur had already predicted, which is the pattern preemptive blocking is built for.
ClickFix and fake updates. ClickFix was the most frequently matched initial access technique, appearing in DriveSurge, DOUBLECUP, Macfinger, CrocoRat, NetSupport RAT delivery via WordPress, and AI-themed InstallFix lures. Half of all IOCs in these reports carried an Augur prediction, the highest rate of any theme.
AI infrastructure. Thirteen reports covered attacks on AI workloads or AI-enabled operations, including LiteLLM gateway and Ollama exploitation, OpenClaw skill marketplace abuse, AI API token theft, and agent hijacking research. Augur had predicted infrastructure later used in these attacks. Push Security's AI-themed ClickFix campaign and Wiz's AI honeypot telemetry both involved IPs Augur flagged 174 and 503 days before first activity.
Edge and enterprise applications. FortiGate, Check Point SmartConsole, Citrix NetScaler, VMware vCenter, Windchill, AhsayCBS, Telerik and Nextcloud exploitation all involved predicted infrastructure. In several of these cases, the infrastructure had been active before Augur's prediction, so the value for defenders is a ready-made block list of ranges already scored as malicious.
One Prediction, Many Campaigns
The same predicted network ranges kept reappearing in reporting on unrelated actors. One /22 alone surfaced in ten separate reports. Threat actors rent from the same hosting pools, so a single range prediction protects against campaigns not yet written about.
Individual IPs recur too. 91.92.40.13 appeared in two SANS ISC diaries on different malware, a Linux SSH cryptominer bot and the AMOS macOS stealer.
Case Studies
In each of these 2026 campaigns, Augur predicted the attacker's IP months before it was first seen in use.
GhostAction. Compromised maintainers pushed fake "security audit" workflows that harvested CI/CD secrets and cloud credentials across tens of thousands of GitHub repositories. Augur had flagged the campaign's infrastructure more than five months before it was first observed, and three separate research teams later published it.
N26 and Copybara. Operators impersonated N26 support by phone, then pushed victims to install the Copybara Android RAT for on-device fraud. Augur's prediction landed nearly four months before the infrastructure was first observed.
Silent Swap and fake VPNs. The Hacker News and Socket both reported clipboard theft on the same IP: the Silent Swap crypto clipper and fake VPN browser extensions. Augur's importance-80 prediction on that /24 also matched BI.ZONE's reporting on the Insolent Hyena cluster.
What This Means for Defenders
Public reporting tells you where attackers were. Predictive threat intelligence (PTI) tells you where they are going to be. Over the 17 weeks covered by this report, organizations blocking Augur-predicted infrastructure would have been blocking servers that later appeared in reporting on supply chain compromises, ClickFix campaigns, state-nexus intrusions, and attacks on AI infrastructure.
Here’s what we recommend:
- Block at the range, not just the IP. The same predicted ranges served unrelated actors, so range-level predictions cover campaigns that have not been published yet.
- Prioritize supply chain egress. Build systems, CI runners, and developer endpoints to reach out to the kind of C2 Augur predicted most reliably in this period.
- Use predictions to shorten triage. When a new report drops, IOCs that already carry an Augur prediction can be actioned immediately, with the prediction date and importance as context.
For questions or additional analysis, contact: research@augursecurity.com
