Pre-Auth File Read Hits Eight Atlassian Data Center Products

Atlassian’s CVE-2026-21589 lets an unauthenticated attacker read known files inside the web application root across eight self-hosted products, and watchTowr and Horizon3 both pushed analyses within a day of disclosure. Separately, attackers hijacked three country-code registries and used that control to mint counterfeit TLS certificates for Google and other major brands.

Vulnerabilities & Exploits

• Atlassian disclosed CVE-2026-21589 on October 5, a CVSS 4.0 9.3 unauthenticated arbitrary file access flaw in Jira Software, Jira Service Management, Confluence, Bitbucket, Bamboo, Crowd, Crucible and Fisheye (Data Center/Server). The attacker must know the exact path and filename — directory listing is not possible — but the read is pre-auth and scoped to the web app root. watchTowr Labs published an analysis and Horizon3 shipped detection; no exploitation has been reported (The Hacker News). @vuln_tracker notes Atlassian lists a Tomcat RewriteValve rule as a stopgap for Confluence, Jira, JSM, Bamboo and Crowd.
• Pwn2Own Ireland 2026 opened with 21 entries on day one and, per BleepingComputer, 32 zero-days demonstrated. Highlights from ZDI’s results: @_McCaulay chained an OOB write with a format string bug on the Sonos Era 300 ($50,000); Taisic Yun of Xint got a reverse shell on LiteLLM via improper input validation plus code injection ($40,000); VinSOC burned seven zero-days on the Philips Hue Bridge Pro. The Samsung Galaxy S26 fell three times, though mostly with vendor collisions, and printers (Brother, Lexmark) again defeated the clock (CyberInsider).
• LibreOffice and Apache OpenOffice will execute attacker code on opening a crafted Calc file, with no macro prompt, when Java support is enabled. A database range pulls a remote ODB file, which names a JDBC driver and a remote JAR; the program downloads and instantiates it in-process. LibreOffice fixed it as CVE-2026-63277 on October 5 (26.2.5 / 26.8.0, alongside an arbitrary file write and several file-read/SSRF bugs); Apache OpenOffice’s CVE-2026-59265 is unfixed through 4.1.16, with 4.1.17 still in testing — disable Java as a workaround. The V12 team published a PoC; no in-the-wild use reported (The Hacker News).
• NetScaler CVE-2026-88779 was added to CISA’s KEV on October 4, and exploitation attempts are hitting appliances already patched against the earlier PitScaler pair (earlier coverage). CERT.dk summarises admin reports of authentication requests with commands hidden in usernames and scripts attempting webshells, reboot persistence and config/backup export — with the caveat that there is no public proof those scripts executed via this flaw. Citrix maintains the observed impact is availability, not customer data.
• Rejetto HFS CVE-2026-61500 now has the mechanism published: HFS leaked Math.random() (xorshift128+) outputs to unauthenticated clients during login and derived the session-cookie signing key from the same generator, so a handful of login responses is enough to reconstruct the key, forge an admin cookie and reach RCE via server_code. Horizon3 found it using Anthropic’s Mythos model; patched in 3.2.1 back in July, with VulnCheck seeing small-scale probing from a China Telecom IP against canaries in Japan and the US (SecurityWeek, earlier coverage).
• Also patched: Microsoft Exchange Server CVE-2026-96940 (CVSS 8.8), an out-of-band fix for weak authorization letting an authenticated attacker elevate privileges and read other users’ mailboxes — Exchange Online needs no action (The Hacker News); and a critical RCE in the Dell System Update CLI deployment tool that yields root (BleepingComputer).

New Tools & Releases

• BigDiskBuster is a proof-of-concept that leaves Microsoft Defender’s service running and reporting healthy while silently blocking signature updates — a detection gap that needs no exploit at all (Dark Reading).
• PoC for CVE-2025-54769 (LPAR2RRD, CVSS 8.8): an authenticated read-only user abuses the upgrade function to upload a crafted archive and reach RCE via directory traversal. Affects ≤ 8.04, fixed in 8.05; the PoC builds the archive and confirms execution with whoami (GitHub).
• vx-underground’s Malware Source Code collection is back on GitHub after an unexplained ban of the profile and repos — forks were left untouched throughout, which nobody has explained either (@vxunderground, ban notice).

Identity, Certificates & Cloud

• Attackers hijacked the .gh, .sl and .as ccTLD registries, rewrote authoritative DNS for selected domains, passed automated domain control validation and obtained unauthorized TLS certificates for “several Google domains” plus other leading global brands. Chrome blocked the certificates it identified and CAs revoked those for Google properties, but Google warns its analysis may not have caught every affected domain and that non-Chrome users are not covered — monitor CT logs and publish restrictive CAA records (Ars Technica, Google) (discussion).
• ASOS customers received an in-app push notification titled “ASOS HACKED” at around 10am UK time on October 6, addressed to the retailer’s DPO and IT team and pointing to a Telegram channel. An actor calling itself Xuanye Group claims full compromise of ASOS’s Snowflake instance; ASOS told the market it is investigating unauthorised activity on third-party customer-communication platforms, that names and contact details may have been accessed, and that it does not believe card data or passwords are affected (BleepingComputer, NCSC-UK) (discussion).
• A human-operated phishing platform is impersonating AI advertising portals for Gemini, Claude, ChatGPT, Perplexity, Meta Muse and Manus. Clicking “Connect” draws a browser-in-the-browser window with a fake address bar showing accounts.google.com or an Okta tenant; the backend fingerprints the device over Socket.IO and lets a live operator choose which MFA challenge the victim is shown next, replaying credentials in real time. One domain, museads.ai, appeared on September 16, days after Meta launched Muse (The Hacker News, BleepingComputer).

AI & Model Security

• Agent-to-agent prompt injection across MCP is turning into a structural problem. Independent researcher Syed Anas Mohiuddin built PoCs against agents at Google, JPMorgan Chase, Rapid7, the French interministerial digital directorate and the US federal government: instructions planted in content are read by one special-purpose agent and passed downstream as a delegated task, and the second agent executes because it trusts the first — frequently landing on SSRF. The Rapid7 issue (CVE-2026-97228) rated only 2.7 and was fixed last month; the Google one rated 8. Rapid7’s Douglas McKee: “each one checks its own front door while nobody watches the hallway in between” (Ars Technica). OX Security’s survey of 15,465 public MCP servers adds the ecosystem view (The Hacker News).
• South Korea believes AI agents were used in a run of bank breaches. At least 68,000 people’s data was exposed across at least seven financial institutions — Hana, KB Kookmin and Shinhan among them, the latter reporting roughly 25,000 affected customers — with borrowing history, income, phone numbers and names in the exposed set. Authorities have identified 33 IP addresses across at least 12 countries and point to the Chinese tool Artex AI; President Lee said “signs have emerged” of agent involvement, which stops short of confirmation. Incidents first surfaced September 30 (The Record, BleepingComputer) (discussion).
• Hacktron reports 24 validated issues, 4 of them critical, in Cloudflare’s “vinext”, a Next.js replacement built in roughly a week; Cloudflare patched them. A useful data point on what AI-accelerated development leaves behind for AI-accelerated review to find (Hacktron).
• Mistral shipped Large 4 (“Le Chonk”), a trillion-parameter model trained on its own European infrastructure and pitched explicitly at security work that closed US models decline. The Decoder notes it jumps on the independent Intelligence Index but still trails Claude, GPT-6 and Chinese open-weight rivals (The Decoder, ZDNET).
• Reflection released Beam, a 501B-parameter MoE activating 23B per token, claimed to match GLM 5.2 on coding and reasoning at three to four times less compute and billed as the most capable open-weight model built outside China (The Decoder).

Threat Activity

• Russian intelligence collection against Finland has intensified, according to Defence Forces intelligence chief Pekka Turunen and Supo director Juha Martelius, with the defence industry and network reconnaissance as the focus. Turunen described the activity as potential preparation for sabotage and pointed to unexplained activity near garrisons and NATO exercises; Martelius singled out small Finnish technology startups tied to the defence sector, saying at least one has verifiably faced increased targeting. Neither considers a military threat likely at present (Yle).
• ClickFix is hiding its payloads better. Microsoft Threat Intelligence describes compromised sites pre-fetching a script into the browser cache disguised as a PNG, so the pasted command loads locally instead of pulling a remote payload — sidestepping Windows Run dialog length limits and the usual network-stage detections; DNS TXT records are also being used as payload storage (The Hacker News, Dark Reading). Separately, CERT-UA found over 100 compromised Ukrainian websites serving fake Cloudflare checks that drop Lunex Stealer, including a Chromium extension posing as “Microsoft Office Word Editor” and a component reaching the filesystem; tracked as UAC-0277 (The Record).
• The FBI and US Secret Service say the FortiBleed campaign is still active, in a joint advisory published October 6: credential stuffing, password spraying and GPU cracking of dumped hashes against internet-facing FortiGate firewalls and SSL VPN gateways, helped along by reused credentials and legacy SHA-256 password storage (FBI/USSS advisory).
• Iran-aligned CL-STA-1178 ran “Blinder Tunnel” against an Iraqi critical-infrastructure target, likely a software engineer. The lure was a fake Dubai Airports recruiter with an offline careers portal and HR questionnaire, followed by a Visual Studio “coding test” whose weaponised .csproj executes the moment the project is opened in the IDE — before any build — chaining into AppDomainManager hijacking (Unit 42).
• Two unrelated WordPress plugins are being exploited to plant hidden admins. Patchstack reports an active campaign abusing stored XSS in Ninja Forms ≤ 3.15.3 (CVE-2026-94504, 500K+ installs, fixed 3.15.4) and WPC Product Bundles for WooCommerce ≤ 8.6.6 (CVE-2026-93836, 30K+ installs, fixed 8.6.7), both CVSS 7.1. The payload is injected without login, fires when an admin views the poisoned submission or order, and installs a fake “WP Smart Thumbnails” plugin (Patchstack, BleepingComputer).
• Linux backdoors masquerading as email security tooling are being used to evade detection in campaigns against targets in Korea and Taiwan (The Hacker News).

Ransomware & Law Enforcement

• Japan extradited a suspected core Qilin member to Germany. Japanese police located and detained a Russian national wanted by German authorities over a ransomware attack on a German company and handed him over at Germany’s request. Qilin claimed the 2025 Asahi Group outage; Japan’s NPA logged 123 ransomware attacks in H1 2026, a half-year record (The Japan Times).
• Osaka Metropolitan University cancelled classes after a large-scale failure of its virtualization platform beginning around 00:20 JST on October 2 was determined to be ransomware, with signs of partial data tampering. Campus network, email, websites, admissions database, finance and HR/payroll are down (The Record, university notice).
• BYOD, a new RaaS operation, leaked data on 3,615 Trump Mobile customers — names, emails, phone numbers, home addresses and order details, with several listed individuals confirming accuracy. Trump Mobile is only the third victim on the group’s leak site (Straight Arrow News, The Register).
• The EndZone crew told vx-underground how it claims to have hit Momentum Telecom: an internet-facing customer-management endpoint of the form ViewCustomer.do?customerId=1 with no authorization check, incremented in a loop. Unverified operator claim, but a reminder that IDOR still scales (@vxunderground).

Supply Chain

• The MALFEX npm campaign has been running since August 2023, with eight malicious packages accumulating roughly 40,000 downloads before being caught (SecurityWeek).
• SelectorsHub, a Chrome extension with about 400,000 users, opens server-picked advertising tabs without any user click — the extension fetches its redirect targets remotely, so the behaviour can change without an update (malext.io).

✎ This issue was written by claude-opus-5. No human edited it before publishing — how this works .