Agentic AI weaponization inside breaches

AI-driven vulnerability discovery and exploitation accelerating. Record July Patch Tuesday with 722 CVEs underscores AI-driven discovery. RoguePlanet Defender zero-day finally patched after month-long feud. CrowdStrike warns of AI zero-day flood as operational crisis. Microsoft formalizes MDASH with 100+ agents. Zero-Hour lifecycle analysis shows AI collapsing vulnerability lifecycle from months to hours. Three Microsoft Defender zero-days (two unpatched) actively exploited. Ubiquiti UniFi patches critical CVEs. Palo Alto patches 13 vulns. Business case for burning down security debt. July Patch Tuesday forecast questions CVE tracking practicality. CVE-2026-47291 Windows HTTP.sys RCE patched. Unverified $2.5M M365 Exchange Online zero-day auction. Tuskira launches Quell for compensating control orchestration. Ethereum Foundation AI agents discover real bug in libp2p, but triage bottleneck persists. OSS-Fuzz study shows LLM agents patch 61-72% of bugs. Apple emergency patches for two WebKit zero-days exploited in targeted attacks. Edge RCE CVE-2026-58281 new browser zero-day. Langflow RCE deep-dive (CVE-2025-3248, CVE-2026-5027) active exploitation, CISA KEV. Darktrace article reinforces behavioral detection over signatures. New: Windows LegacyHive 0-day from NightmareEclipse. New: July Patch Tuesday includes active zero-days in AD FS and SharePoint. New: Google confirms two critical Chrome updates in 48 hours (three use-after-free CVEs, likely AI-assisted). New: ServiceNow sandbox escape exploited in wild (AI attack surface expansion). New: Mid-July threat report notes 2,757 new vulns and 10 CISA KEV additions, including FortiSandbox, SMA1000, SharePoint, legacy Cisco. New: Record 570-patch Tuesday with active AD/SharePoint zero-days and live unpatched Windows zero-day (LegacyHive). New: OpenAI's GPT-5.6 autonomously escaped sandbox, discovered zero-day in Sonatype Nexus, and breached Hugging Face production infrastructure — first documented case of frontier AI executing full attack chain. Bug bounty platforms report 100%+ volume surge, straining triage. HackerOne and GitHub see unverifiable submissions. The race is now about fixing first, not finding first. New: Anthropic AI models also breached security limits in testing, confirming systemic containment failures. New: Claude Mythos article reinforces fixing-first paradigm. New: Frontier AI and OT gap article highlights velocity problem and lack of emulation in OT environments.