Published Wednesday, July 22, 2026 at 08:58 PM PT

<strong>0DAY RUBBISH PROJECT: AI-DRIVEN AUTOMATED ZERO-DAY DISCLOSURE AT SCALE</strong>

BLUF: Project “0day Rubbish” is actively publishing full technical analyses and reproducible exploits for AI-discovered zero-day vulnerabilities; first batch of 10 released July 22. Multi-LLM ensemble (Claude, OpenAI, DeepSeek, GLM) systematically identifies flaws. Threat actors now have weaponized disclosure model plus working proof-of-concept code. All connected infrastructure should assume 10 new unpatched vectors are in active reconnaissance/exploitation phases.

DETAILS:

  • Confirmed scope: First batch contains 10 documented 0-day vulnerabilities with reproducible, published exploits. Full technical analyses (not summaries) are publicly available.
  • Confirmed automation: Vulnerability discovery driven by multi-LLM ensemble combining Claude, OpenAI, DeepSeek, and GLM models; not manual research.
  • Confirmed intent: Project explicitly states continuous disclosure model — this is first batch; more to follow on an undisclosed cadence.
  • Confirmed distribution: Posted to public security mailing list (seclists) July 22, 2026. No gating or responsible-disclosure process mentioned.
  • Timing uncertainty: No visibility into whether these are zero-days in currently-deployed systems, legacy software, or research targets. Threat actors have likely begun reverse-engineering the exploits within hours.

IMPACT:

  • Immediate: Every organization running software in the affected CVE families faces unpatched attack surface with public working exploits.
  • Systemic: Establishes proof-of-concept that LLM ensembles can systematically mine vulnerability patterns faster than vendors can patch. This is a new class of at-scale disclosure risk.
  • Escalation vector: Prior incidents show threat actors rapidly weaponize published 0-day PoCs; expect exploit kits within 48–72 hours.

RECOMMENDED ACTIONS:

  • Now: Query your vulnerability management platform for any CVEs matching the 10 disclosed; assume they’re actively scanned.
  • Now: Brief security and engineering teams that public exploits exist; raise vulnerability triage priority to critical.
  • Within 24h: Contact your major software vendors (OS, cloud, application) asking ETA for patches; plan emergency change windows.
  • Ongoing: Monitor threat intelligence feeds (Shodan, CISA, vendor advisories) for active exploitation reports.

SOURCES:

  • Seclists post, Jul 22, 2026 (zz lin). Project: “0day Rubbish”
  • Multi-LLM ensemble: Claude, OpenAI, DeepSeek, GLM

Recent high-severity events at publish time:

Recent high-severity events