Two CISA KEV additions, an exploited Adobe Commerce upload flaw and the Brotli memory bug, lead a day that also brought a ransomware threat against 26,739 Onamae cloud servers in Japan, an uncoordinated Tailscale SSH disclosure and a heavy run of breach notifications.
🔴 CRITICAL (3)
Adobe Commerce and Magento upload flaw CVE-2026-48356 added to CISA KEV after exploitation Confirmed
CISA added CVE-2026-48356, an unrestricted file upload in the Adobe Commerce and Magento guest cart API rated 9.3, to its Known Exploited Vulnerabilities catalog on 27 October with a 10 November remediation deadline. Adobe shipped fixes in July under APSB26-73 and the flaw is reachable without authentication on stores that expose the REST cart endpoint, which makes unpatched merchants the exposed population. Confirm the running version, check whether the custom options upload path accepts guest sessions, and treat unexpected files in media or custom options directories as possible webshells.
Brotli code execution flaw CVE-2026-111497 confirmed exploited and added to KEV Confirmed
CISA added the Google Brotli out of bounds write tracked as CVE-2026-111497 to its Known Exploited Vulnerabilities catalog on 27 October with the same 10 November deadline, which confirms exploitation rather than a laboratory proof of concept. Brotli 1.0.3 through 1.2.0 is affected, with fixes in 1.0.10, 1.1.1 and 1.2.1. PHP rebuilt its Windows packages for 8.2 to 8.5 on 26 October because they bundle the library, and any service that decompresses attacker supplied Brotli content, including HTTP middleboxes and language runtimes, should be updated and reviewed for unusual child processes.
Sources CISA|The Hacker News
GMO tells Onamae Desktop Cloud customers to shut down after ransomware threat Under investigation
GMO Group urged customers of its Onamae.com Desktop Cloud VPS service to disconnect from the internet and power down machines after an intruder who had already planted malware on up to 26,739 customer servers threatened to launch a ransomware attack. The intrusion was confirmed on 20 October, the attackers have published material since, and GMO is issuing repeated notices in Japan. Customers should keep affected instances offline until clean rebuilds are offered, rotate every credential, key and token that lived on those servers, and check restored servers for persistence before returning them to service.
🟠 HIGH (10)
Tailscale scrambles to patch Tailworm flaws in Tailscale SSH Under investigation
Researcher Erica Windisch of Cyberdione Labs published the Tailworm report on 25 October with details of almost a dozen Tailscale SSH issues, including paths to root on connected hosts, enrolment credential interception and unauthorised device registration, before fixes were available. Testing covered Tailscale 1.104.2 and the self hosted headscale 0.29.5 control server. Tailscale is shipping patches in stages. Organisations that route SSH over Tailscale should upgrade as releases land, restrict the SSH access policy to named users and tags in the meantime, and review audit logs for unexpected node registrations.
Sources SecurityWeek|BleepingComputer
Keystone to destroy 18,835 hardware wallets after device verification key theft Confirmed
Keystone said an attacker who breached its online store obtained the factory key its hardware wallets use to prove they are genuine, which means counterfeit devices could pass the verification check buyers rely on. The company will destroy 18,835 unsold wallets, suspend shipments, and has notified 25,964 affected buyer records, some of which were kept beyond its 180 day deletion policy after an automated tool failed without alerting anyone. Recovery phrases and private keys were not involved. Anyone holding a Keystone should verify the device only through the vendor’s current process and treat unsolicited or second hand units as untrusted.
Sources The Hacker News|CoinDesk|Decrypt
CoinDCX confirms $16 million loss from retired wallets Confirmed
India’s CoinDCX confirmed that 15.99 million USDT was drained from wallets it had retired after last year’s $44 million hack, bringing cumulative losses to nearly $60 million, days after its chief executive publicly called the incident a false alarm. The exchange says it will absorb the loss rather than pass it to users. Dormant wallet infrastructure often keeps live keys and standing approvals, so exchanges and custodians should sweep and zero out retired addresses, and users should review withdrawal whitelists and connected approvals on any platform they use.
Sources Decrypt
HPE patches eight maximum severity flaws in networking products Confirmed
HPE released fixes for dozens of vulnerabilities across its networking portfolio, eight of them rated maximum severity and capable of remote code execution or authentication bypass. The affected products sit at the edge of enterprise networks, where management interfaces are frequently reachable. Network teams should schedule the update outside change freeze windows rather than wait for the next maintenance cycle, and restrict management access to a dedicated jump path in the interim.
Sources SecurityWeek
Microsoft Authenticator flaw let permissionless Android apps steal work tokens Confirmed
Microsoft fixed CVE-2026-80097 in Authenticator for Android version 6.2606.4246. In the researchers’ tests a malicious app holding no permissions could pose as a trusted Microsoft app such as Teams and collect sign in tokens for the victim’s work account. Stolen tokens defeat passwords and ordinary MFA prompts. Android fleet owners should force the update, then review Entra sign in logs for token replay from unexpected devices and revoke sessions where compromise is suspected.
Sources The Hacker News
Lazarus abuses South Korean security software in Operation FlashHole Confirmed
Kaspersky attributed a long running espionage campaign it calls Operation FlashHole to North Korea’s Lazarus Group, with at least three South Korean organisations hit across construction, software and wire communication equipment manufacturing. The group abused locally installed South Korean security software to deploy its malware and protected command and control with post quantum encryption, and the activity is assessed to date back to at least July 2025. Korean organisations should audit security agents for DLL side loading paths and hunt for the associated infrastructure rather than trust a binary because it carries a legitimate signature.
Sources The Hacker News
Warp breach exposes AI agent logs and telemetry for about 1% of users Confirmed
Warp disclosed that an attack on a third party vendor in its infrastructure exposed agent decision logs and telemetry data for some users of its agentic development environment, with the company estimating that roughly 1% of its 800,000 users were affected. Agent logs can hold source code, prompts, internal URLs and secrets that passed through the tool. Affected teams should rotate credentials that appeared in Warp sessions, review what the agent was allowed to reach, and treat the exposed logs as a reconnaissance advantage handed to an attacker.
Sources BleepingComputer
University of New Mexico Health ransomware attack diverts patients Under investigation
UNM Health Sciences Center disclosed a cyberattack that took systems offline, disrupted patient care and academic activity, and forced its main Albuquerque hospital to divert patients while more than 30 clinics closed or cancelled appointments. The incident has the shape of a ransomware event even where the university has not used the word. Healthcare operators should assume initial access through exposed remote services or identity providers, verify that offline backups are genuinely restorable, and harden identity now rather than during an outage.
Sources BleepingComputer|SecurityWeek
ShinyHunters keeps exploiting Oracle PeopleSoft SSRF to RCE flaw Confirmed
ZDI published technical detail on CVE-2026-35273, an SSRF to remote code execution chain in Oracle PeopleSoft, noting that Mandiant reported continued mass exploitation of the flaw by the ShinyHunters group on 25 September. ZDI reported three distinct vulnerabilities to Oracle in June that were consolidated under a single CVE, so patching one component may not close every path. Internet facing PeopleSoft installations remain the priority: restrict access to the application servers, apply the current critical patch update, and hunt for webshells and unusual outbound requests from middleware.
Sources ZDI
Australian Signals Directorate cuts Amadey botnet controller off from 180,000 machines Confirmed
The Australian Signals Directorate’s operators gained access to a criminal’s Amadey malware backend and disconnected the controller from roughly 180,000 infected computers worldwide, including hosts in Australia. The action severs command and control but does not clean the machines. Infected hosts still need remediation, so defenders should hunt for Amadey loaders and the payloads they delivered, and home users should run a full scan rather than assume the takedown fixed the device.
Sources The Hacker News
🟡 MEDIUM (11)
Microsoft ships out of band updates to distrust eMudhra certificates Confirmed
Microsoft released out of band Windows updates that remove trust from certificates tied to compromised eMudhra certificate authority keys, aimed at systems that do not take automatic trust list updates. Any certificate chaining to the affected roots should be treated as untrusted for code signing, TLS and authentication until it is reissued. Administrators should confirm the update is deployed and inventory internal services and signed binaries that depend on eMudhra issued certificates.
Sources BleepingComputer
Leaked vendor key exposed data of 238,000 Denver Health Plan members Confirmed
A leaked SquareML access key could have been used to reach a database holding personal and health information on 238,000 Denver Health Plan members, the plan disclosed. The pattern is an exposed third party integration key rather than a direct intrusion into the health plan. Vendor keys should be narrowly scoped, rotated on a schedule and monitored for use from unexpected networks, and covered entities should know which vendor holds which fields before an incident forces the question.
Sources SecurityWeek
DreamHost forces email and FTP password resets during unexplained incident Under investigation
DreamHost forced password resets for email and FTP/SFTP users after an incident that disrupted connections to its File Manager interface, without yet publishing a cause. Resets of that kind usually indicate credential or session exposure rather than a simple outage. Customers should reset and rotate any credential reused elsewhere, check mailbox rules and forwarding for tampering, and enable MFA where the provider supports it.
Sources SecurityWeek|BleepingComputer
Gheist campaign uses stolen GitHub tokens to leak private repository secrets Confirmed
An automated campaign tracked as Gheist is using stolen GitHub tokens to plant workflows that print a repository’s secrets into GitHub Actions logs, which the attackers download before deleting the runs. Wiz reports success in private repositories and GitHub Enterprise environments, with credentials harvested from thousands of developer repositories. GitHub’s safeguard that pauses workflows from new contributors helps but is not sufficient. Audit Actions runs for unexpected workflow commits, rotate any secret that appeared in logs, and review token scopes.
Sources DarkReading|The Hacker News
ENGIE discloses customer data breach via partner marketing platform Confirmed
French energy group ENGIE confirmed that attackers stole customer data after breaching a marketing platform used by its business partners. The exposure runs through a third party processor, which is where a large share of energy retail breaches now begin. Customers should expect targeted phishing that quotes real account details. Partners and processors should review which customer fields their marketing tools are permitted to hold.
Sources BleepingComputer
Card skimming group hijacks Convertr analytics script across e-commerce sites Confirmed
A credit card skimming group compromised a shared analytics script used by e-commerce platform Convertr to capture payment details during checkout. Because the script is shared, every store that loaded it inherited the malicious code until it was removed. Store owners should pin third party scripts with subresource integrity or a tag manager allowlist, and check checkout pages for foreign iframes, newly registered form handlers or requests to unfamiliar domains.
Sources BleepingComputer
Yoba malware chain hides behind signed ChatGPT and Claude binaries Confirmed
Check Point documented a two month campaign that uses fake GitHub projects posing as AI desktop applications, including invented builds such as GPT-5.6 Sol and GPT-6 Astra, to deliver a custom malware chain it calls Yoba. The installers lean on signed binaries from well known vendors so that the payload looks trustworthy to security tools. Developers should install AI tooling only from vendor domains, check publisher identity rather than the mere presence of a signature, and treat GitHub projects with no history or stars as hostile until proven otherwise.
Sources Check Point Research
Teams IT support scams deliver malware in about 15 minutes Confirmed
Researchers documented hundreds of cases where attackers use the Microsoft Teams external chat creation feature to impersonate IT support, often from throwaway tenants, then push users to install remote access tools and malware within roughly 15 minutes of first contact. The pattern overlaps with Black Basta linked activity. Defenders should restrict external Teams federation, brief staff that IT does not cold call from a new tenant, and alert on remote management tool installs by non administrators.
Sources DarkReading|The Hacker News
Aokigahara stealer targets developer keyrings and AI assistant sessions Confirmed
A stealer tracked as Aokigahara extends the usual infostealer playbook to AI tooling, harvesting local AI assistant sessions, developer keyrings and cloud credentials from a single compromised endpoint. That combination hands an attacker both business data and the ability to act inside developer infrastructure. Endpoint teams should hunt for its persistence, then rotate tokens found in developer home directories, browser profiles and AI client configuration files.
Sources DarkReading
CERT-UA warns of Dahua cameras and DVRs hijacked via P2P access Confirmed
Ukraine’s CERT-UA reported cases of Dahua IP cameras and DVRs compromised through P2P remote access and authentication weaknesses, with attackers viewing video streams, changing configuration and creating additional accounts. Exposed surveillance devices on flat networks remain an easy foothold. Owners should disable P2P and UPnP where possible, place cameras on an isolated VLAN, change default credentials and apply firmware updates.
Sources The Hacker News
ICS advisories: ICONICS authentication bypass, EPICS Base 9.8, Kiloview encoder flaws Confirmed
CISA published advisories for Mitsubishi Electric and ICONICS SCADA products with authentication bypass issues and a bundled libwebp flaw, EPICS Base with four issues rated up to 9.8, and Kiloview E3 NDI encoders at 9.9 where successful exploitation could fully compromise the device. Fixes exist for most affected products, with specific mitigations where they do not. OT owners should treat these as scheduled maintenance with network segmentation as the compensating control, since internet reachable engineering interfaces are the usual way in.
⚪ WATCH (4)
Instruction files in downloads folders used to steer AI coding agents Confirmed
Two reports describe the same weak point from different angles. DarkReading documented an in the wild technique that abuses how AI coding agents consume local Markdown instruction files, and Forever Security caught a malicious instruction file sitting in a developer’s Downloads folder trying to make an agent install a known bad npm package. The agent was stopped before any code ran and the file’s origin was never established. Teams should scope agent permissions to the task, require approval for installs and outbound calls, and treat instruction files outside the repository as untrusted input.
Sources DarkReading|The Hacker News
Project Zero: LLM bug finder reported more V8 flaws than Fuzzilli this year Confirmed
Project Zero compared Fuzzilli, its generative JavaScript engine fuzzer, with BigSleep, an LLM based bug finding system, across six months of V8 testing. BigSleep reported 45 vulnerabilities against 37 for Fuzzilli, with 25% overlap in the specific bugs found, and the write up is careful about what a raw count proves given the differing methodologies. The practical reading is that AI assisted discovery is now routine for browsers and other high value targets, which shortens the gap between patch and exploitation for defenders who lag.
Sources Project Zero
Solana bug bounty now charges 1 SOL per report after AI generated spam Confirmed
Solana’s bug bounty programme began requiring researchers to burn 1 SOL for every submission after a month long contest produced 809 reports, with AI believed to have helped generate or write nearly all of them. The change is a direct response to triage capacity being consumed by low quality submissions. Other programmes are likely to follow with stake to submit models, so researchers should budget for it and write reports that establish impact in the first paragraph.
Sources CoinDesk
OpenAI compaction feature used to extract hidden reasoning Confirmed
Security firm Xint showed that OpenAI’s compaction feature, which shortens a long conversation to save space, could be turned into a way to copy a model’s hidden reasoning back out in plain text rather than the encrypted item the service normally returns. It is the latest example of a summarisation or memory feature creating a read path around a protection that only covers the normal response path. Teams building on model APIs should assume memory and summarisation surfaces need the same threat modelling as prompt handling.
Sources The Hacker News
Methodology: compiled from vendor advisories, government feeds (CISA KEV, MSRC), security news sources and on-chain/security-firm alerts. Grouping: one incident, one entry, with every source cited. Unconfirmed reports are labeled. Crypto items are incident reporting, not investment advice. Corrections from prior digests are noted at the top when applicable.