Oasis Security has analysed files recovered from an attacker-controlled, US-based server used against two of South Korea's largest churches, which it does not name. Researchers collected the files between 28 August and 1 September 2026; Cyber Security News reported the findings on 7 October. At the first church, a web shell on the enterprise resource planning (ERP) system let the attackers decrypt database settings, gain administrator access to Microsoft SQL Server and use xp_cmdshell to run commands and move to linked systems, opening member, accounting, access-control, library, chat and mail databases; they also got past a database-monitoring control, recovered a MariaDB root password and used hardcoded NAS credentials to reach backup storage. That haul held about 960,000 congregant records updated in the past two years, including names and resident registration numbers, about 330,000 donation records, 68,000 document records, more than 14,000 chats and 6,874 login accounts; 47.3 GB in 13,939 files was found staged in a compromised MinIO bucket. The second church was entered earlier with leaked credentials and insecure direct object reference (IDOR) flaws in its groupware and membership (SIMS) systems: from a normal member session the attackers could see other users' plaintext PINs and reset a manager-level account, then reached SAP and took about 89,000 congregant records, 383 employee records and approval documents. Oasis says third-party transfers mean the recovered volume is a lower bound and unique-person totals are unverified. Primary: Oasis Security; wire: Cyber Security News.
Oasis Security — Analysis of cyber intrusions targeting major religious organisations in South Korea
Security researcher Oliver Fish says he found a way out of an OpenAI sandbox that let him send requests to paid AI models with no API key or account, Cyber Security News reported on 7 October 2026. A screenshot he shared shows OpenAI awarding US$300 for a report titled "Unauthenticated Sandbox Escape Enables Access to Internal OpenAI Responses API". If accurate, the flaw crossed two boundaries at once, sandbox isolation and API authentication, letting a remote user skip the normal identity and billing checks. Technical details are private: there is no public proof of concept, endpoint, list of affected models, CVE, exposure window or fix note, and no public evidence that customer data was reached or that anyone else used it. Fish criticised the size of the payout on X; OpenAI's Bugcrowd programme pays from US$200 for low-severity findings to US$20,000 for exceptional ones, so the award may reflect a lower internal severity rating, but OpenAI has not explained it. This is a researcher claim, not a confirmed incident. Source: Cyber Security News.
Cyber Security News — OpenAI sandbox escape flaw allowed free access to paid AI models without an API key (7 Oct 2026)
The Office of the Victorian Information Commissioner (OVIC) has published its investigation into the student data breach the Department of Education reported on 7 January 2026, and found the department contravened Information Privacy Principles 4.1 (security) and 4.2 (destroying data no longer needed). The Australian Signals Directorate issued a critical alert at 10am on 27 October 2025; the department passed it to school technicians the same afternoon, but not every school patched, and the department had no way of checking that they did. Attackers got in through one school's internet-facing server on or before 6 November 2025. A vendor detected the activity on 2 December, forensic work confirmed data had left on 23 December, and media reporting began on 14 January. The stolen database held every government school student's first name and surname, school, year level, school email address and encrypted password, and the same details for hundreds of thousands of former students, which the department kept so email addresses would not be reissued. OVIC called that retention disproportionate. It also found the department's vulnerability scanning did not cover every school and that critical findings were not always fixed. OVIC saw no evidence of misuse soon after the attack but notes the copy could be used later, a particular worry for families escaping violence or under protection orders. OVIC made seven recommendations due by 31 December 2026, including a plan to align with the Essential Eight. The department has accepted them, promised an archive policy for old student records by December and an internal audit in 2027, and plans centrally provided vulnerability management for schools by the end of 2028. OVIC's report does not name the vulnerability. Primary: OVIC investigation report; wire: iTnews (7 Oct).
OVIC — Investigation into the data breach of student information held by the Department of Education (report, October 2026, PDF)
Solar 4RAYS, the threat research team of Russian security company Solar, published an investigation (reported by Cyber Security News on 7 October 2026) into a long intrusion at a medical organisation that it links to the group it calls Partisan Zmiy, based on overlapping infrastructure and techniques. The earliest evidence dates to early 2024 and the investigation began in December 2025, so the attackers had access for about two years; how they first got in is not known. They ran commands remotely, then moved through the network over legitimate remote desktop and Windows file sharing. VMware Tools had been installed long before but was not in use, so its folder became a trusted, unwatched place to drop malicious files named like real virtualisation components, with timestamps changed to match. Shortly before discovery the attackers swapped a genuine VMware library for an unsigned malicious one, keeping the original under another name. Windows services with plausible names started a previously undocumented loader, which ran an updated Vasilek backdoor and GOST tunnels on a schedule, including one tunnel open only from 10pm to 11pm each Saturday, likely as a backup way in. The organisation had two-way trust relationships with many subsidiary medical institutions, and the researchers read the case as espionage rather than disruption. Solar did not name the victim. Primary: Solar 4RAYS blog (Russian); wire: Cyber Security News.
Solar 4RAYS — Partisan Zmiy: атаковала медицинскую организацию (blog, Russian)
Have I Been Pwned (HIBP) added the Angel One breach on 7 October 2026, so people can now check whether their email address is in it. HIBP says that in July 2024 the Indian stock brokerage confirmed that data leaked online related to a breach in April 2023, and that the leaked data included 7.9 million user records with 6.8 million unique email addresses, along with names, physical addresses, phone numbers, bank account numbers, Permanent Account Numbers (PANs, India's tax ID) and portfolio holdings; HIBP also lists dates of birth. In its July 2024 statement, reported by The Hindu BusinessLine, Angel One said no new data breach had occurred, that the issue related to an April 2023 incident that was promptly reported to the relevant authorities, and that "this incident has no impact on client securities, funds, or credentials". It also said the authenticity of the leaked data had not been validated. UPDATE 22 Oct 2026: added Angel One's own statement and attributed the breach details to Have I Been Pwned (source: The Hindu BusinessLine, 10 Jul 2024).
Have I Been Pwned — Angel One breach (added 7 Oct 2026; breach April 2023)
The Harbor project published GitHub security advisory GHSA-2phj-cp9f-qq6w on 6 October 2026 (no CVE assigned): a project administrator can point an HTTP or Slack webhook at localhost, private ranges, link-local, or the cloud metadata endpoint 169.254.169.254. When an event fires, Harbor jobservice POSTs from inside the deployment; if the target returns an error status, the response body is written to the webhook execution log readable by project members — so a ProjectAdmin can reach internal services and skim error responses (including metadata-style leaks). CVSS 3.1 8.5 (AV:N/AC:L/PR:L/UI:N/S:C/C:H/I:L/A:N). Affected: Harbor with webhooks from 1.7.0 onward when untrusted users can become project admin (default Project Creation: Everyone makes every registered user a project admin of projects they create) and jobservice can reach internals. Fixed in 2.13.6, 2.14.5, 2.15.3 and 2.16.0+, which reject private/loopback/link-local webhook targets unless HARBOR_ALLOW_PRIVATE_NETWORK_ACCESS / network.allow_private_network_access is deliberately re-enabled. Until then: set Project Creation to Admin Only, restrict jobservice egress, block 169.254.169.254, prefer IMDSv2. Wire: The Hacker News ThreatsDay (16 Oct) via OX Security. Primary: Harbor GHSA.
Harbor GHSA-2phj-cp9f-qq6w — SSRF in webhook delivery (6 Oct 2026)
CISA published ICS advisory ICSA-26-279-02 on 6 October 2026 for the SMTP client application shipped with lwIP, the lightweight open-source TCP/IP stack (hosted on Savannah) that is compiled into a huge range of microcontroller and embedded products. CISA lists lwIP SMTP client 2.2.1 as affected and says the client does not check the size of its inputs, so a buffer overflow can crash the device and may allow remote code execution; it lists energy and water and wastewater as the critical infrastructure sectors and says deployments are worldwide. CVE-2026-15340 was published to the CVE list on 9 October with a CVSS 3.1 score of 9.8 (CVSS 4.0 9.3). The fix, credited to xchglabs and committed to lwIP's git repository by maintainer Simon Goldschmidt on 12 May 2026 as 'smtp: fix server-driven AUTH line overflow into tx_buf' (bug #68313, commit 614420f), changes only src/apps/smtp/smtp.c, which means the overflow is triggered by the mail server's side of an AUTH exchange and only matters for firmware that builds and uses the SMTP client module. CISA's decision data marks exploitation as proof-of-concept, with no reports of attacks. Because lwIP is embedded inside vendors' firmware, most owners cannot patch it themselves and need a firmware update from the device maker. Primary: CISA advisory; vendor: lwIP fix commit.
wolfSSL released wolfSSH 1.6.0 on 6 October 2026 with five security fixes. CVE-2026-16516 (Critical per wolfSSL): the client did not check that the ECDSA curve in a server's host key matched the negotiated algorithm, so a man-in-the-middle could swap in a key on another curve and pass signature verification with its own private key; it also needs the application to use a lax public-key check callback. Affects all versions through 1.5.0. CVE-2026-83540 (High): wolfSSHd on Windows shared one authentication context and its logon token across concurrent connections, so a user with a valid account could end up logged in as another, more privileged user (1.4.15 to 1.5.0; non-Windows builds unaffected). Medium: CVE-2026-84897 lets an unauthenticated client make a server run client-side DH group-exchange handling and costly primality checks (about half a second of CPU per 1 KB packet); CVE-2026-81535 admits forwarded-tcpip channels without the forwarding policy callback in --enable-fwd builds; CVE-2026-83742 is an SFTP path bug in wolfSSH_RealPath() that writes one NUL byte past a stack buffer and can crash the process. 1.6.0 also turns on strict key exchange (Terrapin protection) by default, requires RSA user keys of at least 2048 bits and limits failed logins to six. wolfSSH is mostly found embedded in devices and firmware, so fixes arrive through vendor updates. No exploitation is reported. Primary: wolfSSH GitHub release notes; wire: Cyber Security News.
wolfSSL — wolfSSH v1.6.0-stable release notes (6 Oct 2026)
Gitea released 28.0.0 on 30 September 2026 (the project has dropped the old '1.' prefix) with 20 security fixes, then 28.1.0 on 6 October with nine more. The most serious, CVE-2026-103059 (CVSS 9.1 as reported by Cyber Security News), affects instances using Gitea's built-in SSH server: public keys were looked up with a case-insensitive SQL LIKE, so a forged case-variant RSA key whose private key the attacker controls could be matched to another user's account. Keys are now matched by fingerprint. Other 28.0.0 fixes close several ways for repository migrations, mirrors and push mirrors to reach internal hosts despite the outbound allowlist (DNS rebinding, multiple DNS answers and approved-domain IP skips, including CVE-2026-70357, CVE-2026-101027, CVE-2026-101029 and CVE-2026-89430); Gitea now sends Git network traffic through an internal proxy that enforces the rules at connect time. Gitea Actions fixes stop cancelled-then-rerun fork pull request runs from reaching self-hosted runners without approval (CVE-2026-104632, CVE-2026-104626) and stop a maintainer's event from running an untrusted contributor's workflow (CVE-2026-94205). There are also stored XSS fixes (container registry blobs, CVE-2026-103667) and a duplicate Git tree entry check that hid files from review (CVE-2026-95106). 28.1.0 adds fixes for push mirror API permission checks (CVE-2026-97208, CVE-2026-86684), repository media stored XSS (CVE-2026-96594) and an OAuth2 refresh grant that accepted access tokens (CVE-2026-101023). No exploitation is reported. Upgrading changes egress settings (set EGRESS_MODE = strict for deny-by-default) and needs Git 2.25 or later, so read the release notes first. Primary: Gitea blog; wire: Cyber Security News.
Gitea blog — Gitea 28.1.0 is released (6 Oct 2026)
Elastic posted 14 advisories (ESA-2026-185 and ESA-2026-187 to 199) on 6 October 2026. The highest-rated, ESA-2026-187 / CVE-2026-102406 (CVSS 3.1 8.8), is in Kibana's Fleet package installation: a user who has been given Fleet privileges to install custom (uploaded) integration packages, but no direct Elasticsearch admin rights, can claim a data stream identifier already used by another user or team in the same Kibana deployment. Fleet then applies the attacker's index and ingest-pipeline settings, so newly ingested data flows through infrastructure the attacker controls and stops reaching its intended destination. Elastic warns the interception can continue after the malicious package is removed, so affected infrastructure must be checked and repaired separately. It affects Kibana 8.14.0 to 8.19.21, 9.0.0 to 9.4.6 and 9.5.0 to 9.5.3, self-managed and Elastic Cloud Hosted, where delegated users can upload custom packages. Other notable fixes: ESA-2026-199 / CVE-2026-103009 (7.1), a cross-cluster search authorisation bypass that only affects clusters acting as a fulfilling cluster under Remote Cluster Security 2.0; ESA-2026-197 / CVE-2026-103007 (7.2), privilege escalation through the delegated manage_roles privilege when it uses wildcard or regex index patterns; and ESA-2026-193 / CVE-2026-102412 (6.5), where limited Fleet users could read Fleet Server host TLS private keys (Kibana 9.3 to 9.5). The rest are mostly authenticated denial-of-service bugs in Elasticsearch, plus an Elastic Endpoint crash loop on Windows hosts with Chinese, Japanese or Korean locales that can switch off Elastic Defend protection. Elastic does not report exploitation. Primary: Elastic security announcements.
ABC News reported on 6 October 2026 that the Australian Competition and Consumer Commission received 138 scam reports involving LinkedIn in July and August 2026, with reported losses of up to $254,449. The pattern is recruitment fraud rather than a platform breach: fake job ads, cloned profiles of real recruiters and employees (one jobseeker was approached in the name of a real Aurecon staff member, and Aurecon confirmed it had not contacted him), and offers that end in an upfront fee. A common version, described by Contract Services Australia managing director Mark Hopcroft, is a 'recruiter' who says your CV fails the employer's keyword screening and then offers their own HR specialist to fix it, in his case for about A$8,600 (US$6,000); he only caught it because the role was not on the company's own careers site. Users say approaches rise sharply after switching a profile to 'open to work'. LinkedIn's Job Search Safety Pulse survey found about 16 per cent of Australian respondents say they have fallen for a job scam (25 per cent of Gen Z), 61 per cent have seen an ad they suspected was fake, and nearly half say cost-of-living pressure makes them take more risks. ANZ warned last month that realistic job scams have become a common impersonation scam. Practical checks: confirm any role on the employer's own website, contact the company through a number you look up yourself, and never pay for CV writing, training or equipment to get a job. Report to Scamwatch and, if money was sent, to your bank straight away.
ABC News — Jobseekers exploited by recruitment scammers on LinkedIn (6 Oct 2026)
SOCRadar's Threat Research Unit published research on 6 October 2026 into CyberXero, a Russian-speaking, financially motivated initial access broker that pairs common attack tools with an AI orchestration layer on its own servers. Everything surfaced through one mistake: an open directory served the actor's live working folder, more than 90,000 files including AI session logs, scripts with plaintext tokens and stolen data, and the actor stayed active while researchers watched. On one workstation it configured up to 51 Claude Code agents for discovery, password testing, exploitation and data theft; a second setup connected PentAGI, an open-source AI penetration-testing framework, to a Cobalt Strike team server through an AI provider's API, with model choice and token budgets set per task. Session logs show the operator telling the model the targets were its own systems under authorised testing and, when refused, opening a fresh session with the same cover story; some refusals held, including requests to install a backdoor, disable a firewall, move across a network and plant a webshell. The broad campaign used an internal wp2shell package that abuses the WordPress REST API batch endpoint to inject SQL, create a rogue administrator and drop a WSO-family webshell, and it also hit Magento and used Support Board CVE-2026-4815 within 30 days of disclosure; one automated run scanned 4,708 targets, found 429 reachable WordPress admin panels and placed 32 shells in 61 seconds. A separate hand-picked pipeline mapped seven Ukrainian energy and utility organisations, including the national transmission operator; four Ukrainian organisations had confirmed data theft, a Kharkiv district-heating provider lost 564,073 subscriber records, and data on more than 628,000 Ukrainians was found. More than 40 organisations were affected worldwide, with activity also seen in Poland, China and Pakistan; SOCRadar found no direct proof of an access sale. Infrastructure was first seen in July 2026 across eight nodes in European hosting and Tencent Cloud. Primary: SOCRadar; wire: Cyber Security News (7 Oct).
SOCRadar — CyberXero: an AI-augmented initial access broker targeting Ukrainian critical infrastructure (6 Oct 2026)
