Weekly Threat Intel: Agentic AI Abuse and September 2026 CVEs
Agentic-AI abuse, the September 2026 CVE wave and ransomware tradecraft — this week’s defensive checklist.
Agentic-AI abuse, the September 2026 CVE wave and ransomware tradecraft — this week’s defensive checklist.
On 17 January 2020, Microsoft published ADV200001, a rare out-of-band advisory for CVE-2020-0674, a remote code execution flaw in the scripting engine used by Internet Explorer 9 and 11 that the company confirmed was being exploited in limited targeted attacks. There was no patch yet, only mitigations and workarounds, and defenders spent nearly a month exposed until the 11 February 2020 cumulative update shipped the fix. This piece reconstructs the advisory, the memory-corruption mechanics in the script engine, why IE was still a live attack surface in 2020, and what the episode taught about mitigations-first disclosure.
In early October 2024, Cisco’s disclosure cadence stacked two unrelated but equally urgent problems: CVE-2024-20419, a CVSS 10.0 unauthenticated password-change flaw in Smart Software Manager On-Prem that let anyone with network access reset the admin API account, and CVE-2024-20399, a CLI command-injection bug in NX-OS already being exploited in the wild per the CISA KEV catalog. This account reconstructs both flaws’ mechanics, the patch timelines, and what this pairing says about authentication surface area in management tooling.
All through 2023, route leaks and suspected BGP hijacks kept redirecting chunks of internet traffic — events touching Rostelecom-linked infrastructure, financial services, and a persistent streak of cryptocurrency-targeting interception paths. None matched the famed mass redirections of prior years, but the pattern of brief, deniable, hard-to-attribute incidents kept routing security in the research headlines. This year-end review explains how BGP trust fails, walks the 2023 incident ledger with appropriately hedged attribution, and covers the defensive state of the art: RPKI signing crossing majority coverage, MANRS norms, and external route monitoring.
The July 2023 NetScaler ADC/Gateway pre-auth RCE hit government networks as a true zero-day — and post-patch webshell persistence taught defenders that appliance remediation means rebuild, not upgrade.
CVE-2023-34362 in MOVEit Transfer let Cl0p deploy the LEMURLOOT web shell at scale over Memorial Day weekend — pure data extortion, no encryption, hundreds of downstream victims.
Through 2024, digital skimming returned to threat reports’ front pages: Magecart-style attacks compromised hundreds of storefronts via compromised third-party JavaScript, supply-chain infections like polyfill.io’s June domain takeover injected malicious scripts into vast numbers of pages, and PCI DSS 4.0’s script-integrity requirements (6.4.3 and 11.6.2) approached their March 2025 enforcement deadline. This survey digests the modern skimming kill chain — injection, exfiltration, and evasion — the major 2024 campaigns, and the compliance clock turning client-side risk into boardroom math.
On October 23, 2024, Fortinet confirmed CVE-2024-47575 — a CVSS 9.8 missing-authentication flaw in the FortiManager FGFM protocol that China-nexus actor UNC5850 had exploited since summer to jump from exposed managers into fleets of managed FortiGates with a custom DeepMove implant. CISA KEV-listed it within days, forcing two-week patch deadlines across federal and enterprise fleets. This account reconstructs the protocol bug, the DeepMove persistence, the fleet-jump blast radius, and the management-plane hardening it made mandatory.
On April 12, 2024, WatchTowr disclosed CVE-2024-3400 — a CVSS 10.0 pre-auth command injection in PAN-OS GlobalProtect that state-sponsored actors had exploited since late March by chaining a cookie-controlled file write into Tcl execution as root. This account walks the two-flaw exploit chain, the scramble after the 10.2.9-h1 hotfix, CISA’s KEV clock, config-hidden persistence that survived reboots, the mitigation-versus-remediation confusion, and why 2024 made appliances patch with server-grade urgency.
January 2024 opened with the year’s first appliance crisis: two pre-authentication zero-days in Ivanti Connect Secure that nation-state actors had already exploited, followed by integrity-check failures and a reset wave across thousands of enterprise VPNs. This account covers CVE-2023-46805 and CVE-2024-21887, the mass exploitation between disclosure and patch, the customers whose breaches surfaced weeks later, and why edge appliances became the year’s most contested patch surface.
CVE-2023-4863, a heap overflow in libwebp’s Huffman decoding, let the BLASTPASS chain install Pegasus via a zero-click iMessage image — and forced every browser, OS, and Electron app on earth to patch the same line of C.
CVE-2023-38831 let a crafted WinRAR archive execute attacker code off a decoy file click — and researchers proved APT campaigns had been exploiting the 2004-era parser flaw for months before the August 2023 fix.