Today’s digest is a reminder that “patched” and “exploited” can trade places more than once on the same bug, and that attackers are now reaching for the same AI coding assistants defenders use. A print-management server needed a second emergency fix in 48 hours, a ransomware crew planned its intrusions with a commercial coding agent, and an airport operator’s exposed front-end secret turned into a multi-million-record breach claim. Here’s what changed since yesterday’s digest.
Vulnerability watch: PaperCut ships a second emergency patch as CISA adds both chained RCE flaws to its KEV catalog
CISA added CVE-2026-81578 and CVE-2026-82078 to its Known Exploited Vulnerabilities catalog on August 31, confirming active exploitation of the PaperCut NG/MF flaw chain first disclosed last week. CVE-2026-81578 lets an unauthenticated attacker modify Application Server configuration; CVE-2026-82078 is an unsafe dynamic class-loading bug that turns that access into arbitrary Java bytecode execution — chained together, the pair delivers pre-authentication remote code execution with no credentials and no user interaction. PaperCut’s first emergency patch didn’t hold: within roughly 48 hours, working with watchTowr and Huntress researchers, the vendor shipped “Emergency Patch Release 2” on August 28 after attackers found a bypass for the original fix. The flaws affect all NG/MF versions prior to August 27, with fixes available for the 24, 25, and 26 branches on Windows, Linux, and macOS. BleepingComputer · Cybersecurity Dive
If you applied PaperCut’s first patch last week and stopped there, you’re still exposed — confirm you’re on Emergency Patch Release 2 specifically, not just “patched,” and treat any PaperCut Application Server that’s been internet-reachable in the last two weeks as a compromise-assessment candidate, not just a patch-compliance checkbox.
Threat actor watch: An Aurora ransomware affiliate used the Cursor AI coding agent to plan and run intrusions against 20+ organizations
CloudSEK and Gambit Security published findings from an exposed open directory belonging to a Russian-speaking Aurora (aka Aur0ra) ransomware affiliate, active against more than 20 organizations across nine countries between April and July 2026 — with CIS-region networks and domains explicitly excluded from targeting. Rather than hand-writing post-compromise scripts, the operator fed the Cursor agentic coding assistant an initial foothold, VPN credentials, or an existing SOCKS tunnel and had it plan and execute the next steps, with recovered chat logs showing the planning conversations conducted in Russian. The leaked directory also contained credential material, Kerberos tickets, shell history, custom NetExec modules, and Windows/Linux Aurora lockers, both written in Zig from a shared codebase. The Hacker News · CloudSEK
This isn’t AI writing novel malware — it’s AI collapsing the skill floor for hands-on-keyboard intrusion work, which is exactly the part of an attack that historically took the most operator time and left the most detectable noise. If your detections lean on “the attacker was slow and made mistakes,” this is a good week to review whether your EDR and identity alerting would still catch a faster, more consistent operator.
Breach watch: Extortion group FulcrumSec claims 86GB from Manchester Airports Group via API credentials exposed in client-side JavaScript
Manchester Airports Group (MAG) — operator of Manchester, London Stansted, and East Midlands airports — disclosed on August 27 that a third party had stolen customer data; MAG’s own count put the exposure at 8.7 million customers, with most limited to email addresses. Extortion group FulcrumSec has since claimed a considerably larger haul: roughly 86GB obtained by using airport-specific Iterable API credentials that were exposed directly in client-side JavaScript, covering car park, lounge, and Fast Track bookings plus airport WiFi registrations — email addresses, phone numbers, vehicle registrations, and postcodes, including close to 200,000 records tied to travel later this year. BleepingComputer validated at least one traveler’s record against the claimed sample but couldn’t independently confirm the full scope; MAG has declined to directly address FulcrumSec’s specific figures. BleepingComputer · Security Affairs
A third-party marketing API key sitting in plain sight in front-end JavaScript is one of the cheapest things to find in your own environment — if you haven’t grepped your production web and app bundles for embedded API keys and tokens this quarter, this is a good prompt to do it before someone else does.
Also noted
Microsoft opened a major incident (tracked internally as EX1464935 / MO1465074) after a wave of reports flagged authentication errors, mail-flow delays, and search failures across Exchange Online, with the disruption spilling into OneDrive for Business, SharePoint Online, Teams, Purview, and Defender XDR. Engineers traced a shared failure pattern to authentication and protocol connectivity and were testing a remediation against a portion of the affected infrastructure as of this writing. If your SOC leans on Defender XDR telemetry for detection coverage, treat any gaps during this window as a visibility blind spot to review once the incident closes, not just an inbox inconvenience. BleepingComputer
D-Link published advisory SAP10513 for the DIR-X1860Z router, fixing access-control and information-disclosure flaws in the device’s OpenWrt-based ubus JSON-RPC management interface (exposed on TCP port 23355 and the /ubus endpoint) that let an unauthenticated attacker on the local network change the admin password and recover Wi-Fi credentials. Firmware V1.0.7.260821.161908 fixes the issue — and D-Link specifically warns not to cross-flash DIR-X1860 firmware onto a DIR-X1860Z unit or vice versa, since the two are separate hardware revisions. Cybersecurity News
Final thought
Three stories, one shared lesson: the interfaces you trust by default — a print server’s config API, a router’s local management socket, a marketing vendor’s API key shipped to every browser that loads your site — keep being the fastest path in, whether the actor on the other end is a human operator or a human operator directing an AI agent to do the legwork. Patch PaperCut to Release 2 specifically, and audit what secrets your own front-end code is quietly handing out. If you want a second set of eyes on either, see our how-to guides or book a discovery call.