“89% of phishing domains are now active for fewer than two days,” the source reports — and by the time a domain lands on a blocklist, the campaign has often already moved on. That single statistic frames a shift the authors say AI simply finished: indicator-based blocklists can no longer keep pace with phishing campaigns that are ephemeral, automated, and engineered to avoid analysis.
Disposable infrastructure and trusted hosting platforms
Phishing campaigns are disposable by design. According to the reporting, attackers routinely spin up and tear down infrastructure so quickly that only 6.5% of phishing domains survive past 15 days. Operators combine trusted hosting platforms — Cloudflare Workers, Railway, Vercel, Microsoft Dynamics, SharePoint, Adobe, Google Firebase, Google Sites, Jotform, Linode, Azure, Cloudflare, Atlassian, and others — with bot protection, referrer checks, browser fingerprinting and complex redirect chains to filter out researchers and automated scanners.
Push Security’s telemetry shows 95% of the in-browser attacks it detects use some form of bot protection, often layered so the crawler-facing page differs from the real victim-facing page. The result: a once-malicious page may no longer serve a payload by the time it can be analyzed and added to a blocklist.
AI, screenshot cloning and the collapse of static fingerprints
AI has compressed the cost and time of producing convincing phishing pages. Attackers can "vibe-code" entire phishing frontends from a screenshot in minutes — creating unique codebases that look like legitimate login pages but leave no static fingerprint for signature-based detection. Phishing delivery increasingly abuses legitimate services — the source lists AI chatbot sharing features, search ad placement, in-app messages and app-generated emails — which inherit platform reputation that blocklists would not flag.
Nobody's watching your logs at 2 AM.
Full SOC coverage without building one. Nubivance deploys and manages Rapid7 InsightIDR and MDR for organizations that need detection and response, not another dashboard.
Get coverageTooling erosion: PhaaS, kit fragmentation and rapid commoditization
The middle of David Bianco’s Pyramid of Pain — tool and kit fingerprints that once offered durable detection — is eroding. The phishing kit ecosystem fragments through forking, AI-assisted development and open-source-style sharing faster than defenders can track. Device code phishing moved from an early novelty in 2024 to broad criminal adoption by 2026, expanding from zero criminal kits to "25+ distinct kits (and counting)." Criminal PhaaS kits named in the source include EvilTokens (used against 340+ organizations in its first five weeks), Kali365 (which prompted an FBI advisory), ARToken, DEBULL, Forg365, and many others; established AiTM vendors like Tycoon 2FA have added device code phishing features as well.
These kits often include attacker-operated admin panels that control payload delivery and can "activate" malicious pages only when an admin performs a social-engineering step, further reducing the chance pages are seen more than a handful of times before rotation.
What survives: technique-level behavioral detection (AiTM, ClickFix, device code phishing)
Not all detection surfaces have vanished. The authors argue that genuinely new attack techniques still require human creativity, and that behavioral mechanics are harder for attackers to change than infrastructure or fingerprints. Examples called out in the reporting:
- AiTM (adversary-in-the-middle) phishing: push-style interception that proxies the victim session through attacker infrastructure; Push Security is tracking 75+ AiTM kit variants implementing the same interception mechanics.
- ClickFix: social engineering that injects malicious commands into the clipboard and instructs a user to paste and run them — the lure changes, the clipboard-manipulation behavior does not.
- Device code phishing: victims complete a legitimate device code auth grant; attacker behavior and destination polling pages remain constant even as kits and hosting rotate.
Detecting at that technique level requires two things many programs lack: visibility where the attack actually executes (inside browser sessions that network proxies and EDR may not see) and research velocity fast enough to extract behavioral signatures before techniques are industrialized into commodity kits.
Push Security’s agentic pipeline and the OAuth redirect case
The source describes how Push Security used an "agentic threat hunting pipeline" — AI agents operating as a force multiplier for human researchers across browser telemetry from "3+ million deployments" — to convert published research into a behavioral detection. After Microsoft published research on a novel OAuth error-handling redirect abuse, Push’s agents extracted the OAuth-redirect behavioral mechanics rather than relying on IOCs. Months later that detection fired on a different campaign with different lures, domains and a previously unseen phish kit; none of the original IOCs appeared in the attack chain.
Push says the same pipeline identified three additional novel browser-based techniques — InstallFix (malware delivery disguised as installation prompts), ConsentFix (OAuth consent phishing combined with ClickFix-style manipulation), and LLMShare (malware delivery via AI chatbot sharing features) — and in several cases blocked active campaigns before the techniques were publicly documented. The company reports the pipeline protected "60+ customers in the last three months," intercepting roughly 225 threat instances, and claims it has "tripled monthly detection output" by operationalizing behavioral detection at machine speed.
What this means for technologists, procurement leaders, and end users
- Technologists and security teams: The source argues teams need browser-session visibility and higher research velocity to detect technique-level behavior — because network proxies and EDR do not capture many in-browser interaction signals.
- Procurement leaders and enterprise buyers: Blocklists are described as structurally obsolete; the reporting suggests evaluating controls that operationalize behavioral, technique-level detections rather than relying primarily on indicator lists.
- End users and orgs exposed to platform features: Phishing increasingly leverages AI chatbot sharing, search ads, in-app messages and app-generated emails to inherit platform reputation, raising the risk that otherwise trusted channels will deliver convincing, short-lived campaigns.
AI hasn't merely accelerated phishing — the reporting contends it has changed the rules. Short-lived domains, automated cloning from screenshots, kit fragmentation, and the theft of platform reputation together make indicator-based blocklists a reactive exercise. What the source presents as durable is method-centric detection: visibility into browser sessions and a research pipeline fast enough to extract behavioral signatures before adversaries commoditize them. Whether defenders can scale that visibility and velocity across enterprises at large will decide if blocklists become a historical footnote or a continued, fragile layer of defense.
Source: How AI-powered phishing killed blocklists for good — BleepingComputer
