On the morning of 29 October, a mid-size hosting provider watched three decoy PHP listeners light up within twelve minutes of the CVE-2026-109618 and CVE-2026-109619 brief landing on practitioner channels. Overnight, the same sensors had already logged SSH and Telnet probes that matched tooling patterns later tied to the 1.3 million TV-box proxy botnet dismantled by a multi-firm effort. Operators dumped every source IP into a shared blocklist and called the shift a win. By afternoon, a ransomware affiliate reused one of those prefixes for credential stuffing against a customer admin plane that still trusted the decoy feed as “noise already handled.”
The decoys worked. The intelligence pipeline did not. Mass exploit scouts and ransomware staging share ports, user-agents, and even payload wrappers during weeks when Microsoft ships large vulnerability batches, AI-assisted red-team tooling accelerates recon, and PHP advisories drive automated internet-wide scans. Treating every honeypot hit as the same class of threat intelligence collapses response into one slow playbook.
What broke in the decoy-to-ticket path
The provider ran low-interaction Cowrie and a custom PHP decoy that returned fake version banners and captured POST bodies. Sensors shipped JSON lines to a SIEM with fields for source IP, destination port, first bytes, and session duration. Analysts had one rule: any authenticated login attempt or RCE-like payload became a Priority 2 ticket and a candidate for upstream sharing.
That rule mixed three behaviors that need different clocks:
- Opportunistic PHP scanners that rotate /24s every few minutes and never return
- Proxy-botnet side channels that probe SSH/Telnet in short bursts from residential and set-top ASNs
- Human-paced staging that revisits the same decoy, fingerprints admin paths, and stages uploaders before pivoting to real tenants
When the PHP critical notices hit, ticket volume spiked and staging sessions aged past the shift. Upstream partners received a flat list of IPs with no dwell, revisit, or payload-stage tags. Edge quarantine treated a ransomware scout the same as a throwaway scanner that had already moved on.
Classify sessions before you promote indicators
Build a short taxonomy at ingest so every honeypot event carries a campaign class before humans open the ticket. Keep the classes few and operational:
- Spray scout — single touch, generic exploit string, ASN with high historical scanner density, no revisit inside four hours
- Proxy echo — protocols and banners aligned with known IoT or set-top abuse, short TCP lifetime, often overlapping peer graphs from recent botnet takedowns
- Stage probe — multi-step interaction, path enumeration beyond the first payload, credential attempts with realistic password lists, or a return visit from the same /32 or neighboring /29
Map each class to a different TI product. Spray scouts feed rate caps and temporary deny deltas. Proxy echoes feed ASN peer hunts and residential proxy detection. Stage probes feed same-hour hunts across auth logs, webshell detectors, and partner sharing packets with full session narrative.
Fields that make the taxonomy stick
Enrich every decoy session with:
- Interaction depth (banner only, command attempt, file write, second-stage download)
- Revisit count and inter-arrival time
- Payload hash and family hint when the sample matches known PHP RCE wrappers or webshell droppers
- Prefix tenure and hosting category (cloud VPS, residential, mobile, set-top/CGNAT)
- Protocol mix across your decoy farm for the same source in a 24-hour window
Store the classification decision on the event. Downstream blocklists and STIX packages should inherit that label so partners can choose spray hygiene versus staging response without re-analyzing raw PCAP.
A workable pipeline for the current threat week
Recent pressure is concurrent: PHP critical flaw waves, ransomware volume rising, AI tooling shortening recon loops, and leftover proxy capacity from large IoT botnet families. Design the pipeline for concurrency, not for a single campaign name.
Sensor placement that answers operator questions
Place low-interaction listeners on the same exposure classes attackers actually hit after advisories: PHP-FPM and common CMS entry paths on unused vhosts, SSH on non-production jump images, and one protocol decoy that mirrors remote-support ports your org really uses. Keep production out of the honeypot VLAN. Mirror only metadata and payloads into the TI bus; keep full session recordings in a short-retention vault for legal and abuse desks.
Same-hour joins, not overnight digests
Join decoy hits to edge WAF, IdP auth failures, and outbound proxy logs on source IP and /24 within the hour. When a stage probe coincides with a real tenant auth anomaly, escalate as intrusion support. When a spray scout floods PHP decoys and never appears on auth, keep the response in automated rate policy. The November threat intelligence cycle rewards teams that turn decoy signals into containment before the next advisory crowds the queue.
What to share upstream
Share stage probes with session story: timestamps, depth, hashes, and the classification rationale. Share spray scouts as aggregated prefix and ASN counts with time windows, not as one-IP-one-ticket dumps. After set-top proxy takedowns, include peer-membership hints when your decoys see the same families probing SSH or HTTP from new ASNs; family names often lag infrastructure by days.
Tradeoffs and caveats operators hit in practice
High interaction looks richer and burns more staff. Full shells and malware detonation produce excellent TTPs and attract lateral-movement tooling, including AI-assisted red-team style automation. They also create containment risk and false confidence that every interesting session yields an operator profile. Reserve high interaction for confirmed stage probes after low-interaction triage.
Aggressive auto-block from decoys poisons partner trust. Spray scouts sit on shared cloud and hosting prefixes. A permanent deny based on one honeypot touch can cut customer callbacks and monitoring agents. Prefer time-boxed caps keyed to campaign class, with manual promotion for stage probes.
Payload hashes alone age fast. PHP exploit kits mutate wrappers within hours of a widespread flaw notice. Keep behavioral features (depth, revisit, path set) as first-class TI even when the hash is novel.
Legal and privacy scope must stay explicit. Capture only what your policy allows on decoy networks. Strip customer PII if a misconfigured route ever mirrors production traffic. Document retention separately for malware samples versus connection metadata.
Remediation checklist for the next dual-wave week
- Add campaign-class labels at honeypot ingest with spray, proxy-echo, and stage-probe defaults.
- Require revisit or depth thresholds before a decoy IP enters human Priority 2 queues.
- Wire stage probes into SIEM hunts that already cover admin portals and PHP app pools; leave spray to edge rate policy.
- Package upstream shares with class, dwell, and hash so recipients apply the right control.
- Rehearse the playbook against a tabletop that starts with a PHP CVE drop and a simultaneous IoT-proxy probe surge, then inject one slow staging return visit.
- Review auto-deny TTLs weekly so ransomware staging that reused a “handled” scout prefix cannot ride stale assumptions.
Concrete example from a dual-signal day
At 09:14 UTC, decoy PHP-FPM accepted a POST matching public PoC structure for CVE-2026-109618 from a cloud VPS with tenure under three days. Session depth stopped at one payload. Class: spray scout. Action: /24 rate cap for six hours, no ticket.
At 09:41, the same decoy farm saw SSH password sprays from a residential ASN previously linked to set-top proxy peers. Class: proxy echo. Action: feed ASN into overnight peer scoring; block outbound to known C2 patterns from that peer set.
At 11:02, a /32 that had banner-grabbed at 08:55 returned, enumerated two fake admin paths, and attempted a second upload. Class: stage probe. Action: Priority 1 hunt across customer auth and WAF for that /29, temporary quarantine on matching JA3 and URI patterns, and a partner share with full narrative before the afternoon ransomware desk closed.
Three hits, three products. The decoy farm stayed quiet for operators because classification happened before triage, not after the ticket backlog formed.
Takeaways for TI and admin teams
Honeypots earn their keep when they sort attacker intent under pressure from PHP mass exploitation, IoT proxy residue, and rising ransomware staging. Promote indicators only after session class is attached. Match each class to a control with a clear TTL and owner. Keep high-interaction capacity behind that filter so AI-accelerated recon and human operators both produce usable intelligence instead of another undifferentiated IP list.
When the next advisory and the next botnet brief land on the same morning, your decoys will light up together. The teams that win the shift are the ones that already know which knocks deserve a hunt and which knocks only deserve a temporary cap.