GreyNoise: residential proxy IPs rotate too fast for IP reputation to catch
GreyNoise studied 4B edge-attack sessions: 39% of source IPs were residential and 78% of those appeared at most twice. Why IP blocklists fail and what works.

GreyNoise analysed about 4 billion malicious sessions aimed at internet edge devices over 90 days and found that a large share of attack traffic comes from home internet connections that appear once or twice and are never seen again. For anyone defending a login page, an API or a VPN with IP blocklists and per-IP rate limits, the conclusion is uncomfortable: the IP address is becoming a weak signal.
What happened
On 2 April 2026 GreyNoise published "The Invisible Army: Why IP Reputation Fails Against the Rotation Economy". Key findings from GreyNoise's write-up:
- 39% of unique IPs targeting the edge came from residential connections, though they produced only 22% of sessions. Many addresses each sent a small amount of traffic.
- 78% of residential IPs were observed at most twice across GreyNoise's sensor network before rotating away. That is usually too briefly for reputation feeds to list them in time.
- A residential IP averaged fewer than three sessions, with a median of one.
- Only 0.1% of residential sessions carried exploitation payloads, compared with 1.0% from hosting infrastructure. Residential traffic is mostly scanning and credential activity.
- Residential traffic follows human daily cycles. Traffic from IPs in India dropped 34% between day and night, while server traffic varied by less than 3%. That fits compromised home devices and proxy-SDK apps that go offline overnight.
- 84% of SMB worm propagation came from residential space.
BleepingComputer's coverage adds that residential IPs came from 683 ISPs, that 89.7% were active for less than a month, and that 1.3% targeted enterprise VPN login pages. GreyNoise also notes that after Google disrupted the IPIDEA proxy network in January 2026, the pool shrank but other providers absorbed the demand.
Who is affected
- Teams relying on IP reputation feeds or geo/IP blocklists to protect edge devices, VPN portals, login pages and APIs.
- Applications that rate-limit only per IP, which rotating proxies avoid by design.
- Fraud and bot-detection pipelines that treat residential ISP addresses as low-risk.
What it means for developers and engineering teams
Residential proxy networks sell access to real home connections, usually from SDKs bundled into free apps, from compromised routers and IoT devices, or from malware. A buyer can send each request from a different consumer IP. The request then looks like normal customer traffic: a big ISP, a plausible country, no history of abuse.
This breaks several common defences. A blocklist only helps if the address comes back, and most don't. A limit_req rule keyed on $binary_remote_addr stops one noisy IP but not 10,000 quiet ones. Blocking whole residential ranges would block your real users.
GreyNoise's recommendation is to stop using IP reputation as the primary signal and focus on behaviour. In application terms, rate-limit by what the attacker can't rotate cheaply:
# pseudo-code: limit login attempts per account, not just per IP
key = f"login:{username.lower()}"
if redis.incr(key) > 10: # 10 attempts
deny_or_challenge()
redis.expire(key, 900) # per 15 minutes
Combine per-account limits with per-device fingerprints, breached-password checks and a step-up challenge rather than a hard block, so real users locked out by an attack still have a way in.
What to do now
- Add per-account and per-credential rate limits on login, password reset and OTP endpoints, alongside per-IP limits.
- Use IP data as context, not a verdict. Enrich requests with ASN type (residential, mobile, hosting) and proxy-detection scores, and feed them into risk scoring.
- Track signals that survive rotation: TLS fingerprints (for example JA4), device identifiers, header order and request timing patterns.
- Block protocols that make no sense from consumers. GreyNoise suggests blocking clearly illegitimate traffic such as SMB from ISP address space at the edge.
- Protect VPN and admin portals with phishing-resistant MFA so a successful password spray from a clean residential IP still fails.
- Watch for low-and-slow patterns, such as many accounts each seeing one or two failures from different IPs within the same window.
- Review your threat-intel spend. If a paid IP reputation feed is your main bot or attack defence, measure how often it actually matches traffic you later confirmed as malicious, and move budget towards behavioural detection if the hit rate is low.
Related on TechLogHub
- Caddy vs Nginx: reverse proxies and rate-limiting options
- Authentication
- CDN: edge networks and bot management