CISA added four exploited vulnerabilities to its Known Exploited Vulnerabilities catalog in a single day, led by a CVSS 10.0 Arista VeloCloud Orchestrator flaw with fixes still missing for two release trains, Check Point disclosed a pre-authentication zero-day in its Security Management Server, and WordPress rushed out 7.1.2 for an unauthenticated file inclusion bug that attackers began probing within hours.
🔴 CRITICAL (2)
Check Point fixes a pre-authentication zero-day in Security Management Server exploited since July Confirmed
Check Point released emergency hotfixes on 22 September for CVE-2026-93616, a path traversal flaw in the Security Management Server web service that lets an unauthenticated attacker upload and run scripts on the server that controls firewall policies for the gateways it manages. The company says a handful of customers were attacked, with the first exploitation seen on 23 July, and the flaw carries a CVSS score of 9.8. Check Point also warned that attackers have been trying since 12 September to exploit CVE-2026-85102, a certificate validation flaw it fixed on 9 September that affects Spark firewalls for small businesses as well as management servers; both flaws are now in CISA’s Known Exploited Vulnerabilities catalog. Administrators should check their release and Jumbo Hotfix take against support article sk1000171, install the fix, and work through Check Point’s hunting guidance, because patching does not by itself show whether a server was already accessed.
Sources Check Point|Check Point support sk1000171|The Hacker News|BleepingComputer
Arista VeloCloud Orchestrator flaw CVE-2026-93952 is exploited, with no fix yet for two release trains Confirmed
Arista disclosed on 22 September that attackers are exploiting CVE-2026-93952, a maximum-severity flaw rated CVSS 10.0 in on-premises VeloCloud Orchestrator, the server that manages the Edge devices in a VeloCloud SD-WAN deployment. Only orchestrators configured to authenticate Edges with certificates are exposed, and an attacker also needs network access to the VCO web interface plus the public part of an Edge authentication certificate. Fixed releases exist for the 5.2 and 6.4 trains, while 6.1 and 7.0 have no fix yet; CISA added the flaw to its Known Exploited Vulnerabilities catalog on 22 September with a federal remediation deadline of 25 September. Until an upgrade is possible, restrict the VCO web interface to trusted administrative networks, monitor for unexpected outbound traffic, backdoor daemons and webshells, and review recent administrative accounts.
Sources CISA|Arista advisory 0183|The Hacker News
🟠 HIGH (6)
GreyNoise links a Chinese-speaking actor to Zyxel and WordPress exploits at 49 organisations in 29 countries Confirmed
GreyNoise researchers report that a Chinese-speaking actor exploited known flaws in Zyxel GS1900 switches and the WordPress wp2shell chain, CVE-2026-63030 and CVE-2026-60137, reaching 996 devices and stealing more than 18,500 records from backend databases while breaching at least 49 organisations in 29 countries. One intrusion at a Western government organisation saw the attacker run extensive reconnaissance against Microsoft Defender and AMSI, then use database credentials found on the host to password spray an internal SQL server holding accounts, plaintext passwords and government and law enforcement PII. Activity from the same source addresses has been recorded since early June and also targeted PAN-OS GlobalProtect, FlowiseAI, Nuclio, Proxmox and Ubiquiti. Apply the Zyxel and WordPress fixes, rotate credentials that were stored in affected databases, and check edge appliances and web servers for webshells.
Sources BleepingComputer|SecurityWeek|CISA
Microsoft seizes EvilTokens infrastructure after device code phishing compromised 12,000 accounts Confirmed
Microsoft’s Digital Crimes Unit, acting with authorisation from a US district court, disrupted the EvilTokens phishing-as-a-service platform, seizing 50 websites and disabling more than 150 domains with help from Cloudflare, Coinbase, OpenAI, Health-ISAC and Shadowserver. Microsoft says the platform compromised more than 12,000 Microsoft accounts across over 10,000 organisations, using device code phishing, AI-assisted lures and automated infrastructure to steal tokens rather than passwords. Because device code flows hand an attacker a legitimate token once a user approves a code, domain blocking is not sufficient on its own: review sign-in logs for device code grants, revoke suspicious tokens and refresh tokens, and restrict or monitor the device code flow wherever it is not operationally needed.
Sources Microsoft Security Blog|The Hacker News|BleepingComputer|Dark Reading
ShinyHunters claims an FBI intrusion through an Oracle PeopleSoft zero-day Unconfirmed report
The ShinyHunters extortion group says it used a previously unknown vulnerability in Oracle PeopleSoft to reach FBI systems on the night of 21 September, then moved into FBI-managed AWS GovCloud infrastructure and took between 2TB and 3TB of data covering current and former employees, job applicants and internal records. The group also shared a screenshot appearing to show the FBI Jobs site defaced, and says it is now using the same flaw against other organisations including Fortune 500 companies. The FBI has confirmed it is investigating unauthorised activity affecting FBijobs.gov but has not confirmed a breach or data theft, and neither Oracle nor the agency has verified the zero-day; BleepingComputer has not independently verified the claims. Treat the report as unconfirmed, and if your organisation runs PeopleSoft, review the external exposure of its internet-facing components and apply Oracle guidance as it appears.
Sources BleepingComputer|Nextgov/FCW
WordPress 7.1.2 closes unauthenticated file inclusion flaw CVE-2026-87902 as probing starts Confirmed
WordPress released 7.1.2 on 22 September to fix CVE-2026-87902, rated critical with a CVSS score of 9.2, after finding that one of the template file names core builds from the web address was not passed through the project’s own directory traversal check. An attacker needs no account and no action from a signed-in user to make an affected site load a local PHP file from outside its theme folders, and on servers that already hold a useful PHP file that can extend to running attacker-chosen code. Every version from 4.7 through 7.1.1 is affected, including 7.1.1 from the 17 September security release, so a site updated last week still needs this one, and Patchstack reported that attackers began probing sites hours after the patch landed. Update to 7.1.2, or to 7.0.6, 6.9.9, 6.8.10, 6.7.9 or 6.6.9 on older branches.
Sources WordPress.org|The Hacker News|Patchstack
BigDiskBuster zero-day blocks Microsoft Defender updates and has no patch Confirmed
A former Microsoft security researcher, Abdelhamid Naceri, published a proof-of-concept tool called BigDiskBuster on GitHub on 19 September that stops Microsoft Defender installed on Windows from installing platform and signature updates by filling all available disk space. The tool has no CVE, no vendor advisory and no patch, and SecurityWeek reports that the same person is the exploit publisher Nightmare Eclipse, also known as Chaotic Eclipse, whose earlier Defender exploits were reused in real attacks. An endpoint left on stale signatures loses detection for newer threats, so monitor free disk space and Defender update health, and treat unexplained disk exhaustion on a workstation as a security event rather than a storage problem.
F5 BIG-IP APM heap overflow CVE-2026-94127 added to KEV with exploitation reported Confirmed
CISA added CVE-2026-94127 to its Known Exploited Vulnerabilities catalog on 22 September, a heap-based buffer overflow in the F5 BIG-IP access policy manager that sits in the Traffic Management Microkernel, the process handling data plane traffic. It is reachable when a virtual server is configured with both an APM access policy and an OAuth profile, where crafted traffic can overflow a heap buffer during OAuth processing, and the flaw carries a CVSS score of 9.8. F5 has published fixed versions and guidance, and federal agencies have until 25 September to remediate under BOD 26-04. Patch the affected virtual servers, then check them for signs of compromise, because exploitation is already confirmed.
Sources CISA|F5|CISA KEV Catalog
🟡 MEDIUM (13)
Sweden fines Miljödata SEK 1.8 million over a 2025 breach affecting 2.2 million people Confirmed
Sweden’s data protection authority, IMY, has fined the IT systems provider Miljödata SEK 1.8 million, about $183,000, for inadequate security measures that contributed to a breach in August 2025 affecting 2.2 million people. The decision concerns failures in the controls protecting personal data rather than the intrusion itself, and the penalty is small relative to the number of people affected. Suppliers that hold personal data on behalf of public bodies should review access control, logging and breach detection against their contractual and regulatory obligations, in Sweden and in their own jurisdiction.
Sources BleepingComputer
ClosedQuorum is the first reported malware implant to run its own AI-driven command and control Confirmed
Cisco Talos analysed a Windows malware family it calls CLOSEDQUORUM, found through its CAIRN research project, which calls large language models including Google Gemini, DeepSeek, Qwen and Mistral to decide autonomously what to do during post-compromise stages of an intrusion, making it the first reported fully autonomous AI command and control implant. Talos frames the significance as effort displacement: parts of the attack chain that previously needed an operator can now run without one. Expect longer periods of activity with no human pacing, and lean on behavioural detection of tooling plus monitoring for outbound model API traffic from servers that have no reason to make it.
Sources Cisco Talos|BleepingComputer
Malicious npm package posed as a Twilio bug-bounty probe Confirmed
Researchers disclosed a malicious npm package named tw-pkgprobe-7731, uploaded in mid-August 2026 by an account called twdepprobe7731, which presented itself as a security tool for developers integrating Twilio while attempting to harvest credentials and other sensitive data from the machines where it was installed. Packages that arrive disguised as security or bug-bounty tooling exploit the trust developers place in tooling aimed at developers. Audit recently added dependencies, look for the package and its publishing account in lock files and build logs, and rotate anything that was reachable from an affected workstation.
Sources The Hacker News
Shai-Hulud activity costs CrowdSec 170 private GitHub repositories Confirmed
Threat actors stole 170 private repositories from the security firm CrowdSec using an OAuth token taken from a former employee’s computer, in activity tied to the Shai-Hulud campaign that spread through the TanStack npm supply chain attack. The theft shows how far a single stolen token reaches inside a software organisation, where private code and internal tooling can be worth taking on their own. Inventory long-lived OAuth tokens and personal access tokens, revoke those belonging to former staff, and put expiry and review cycles on the rest.
Sources Dark Reading
Bifrost AI gateway flaw CVE-2026-90898 allows unauthenticated command execution Confirmed
A critical flaw in Bifrost, an open-source gateway that routes requests to more than 20 LLM providers, lets an unauthenticated attacker run arbitrary commands on the gateway server with a single HTTP request when management authentication is disabled, which is the default configuration. The flaw carries a CVSS score of 9.8 and is reached through an unauthenticated POST to the management API that registers a stdio MCP client, causing Bifrost to start the supplied command as the gateway process user before any handshake. Because the gateway stores API keys for every connected provider, that access exposes those credentials. Move to transports v2.1.0, enable governance authentication with strong credentials, keep the management listener off untrusted networks, and rotate virtual keys and provider API keys on any instance that ran exposed.
Sources The Hacker News
D-Link warns of an unpatched maximum-severity flaw in DIR-822A routers with public exploit code Confirmed
D-Link has warned that legacy DIR-822A dual-band routers carry a maximum-severity stack-based buffer overflow, CVE-2026-86296, in the DHCP server component, which an attacker on the same local network can trigger without authentication or user interaction by sending crafted DHCP packets. The researcher who reported it has published proof-of-concept code and there is no patch; D-Link is also investigating CVE-2026-86510, an out-of-bounds write in the L2TP control message parser in the same model, which also has public exploit code. These devices regularly end up in botnets, so keep them off the internet, disable remote management, and limit administrative access to trusted systems and users.
Sources BleepingComputer
CISA publishes nine ICS advisories covering Siemens products, lwIP and OpenPLC Confirmed
CISA released nine industrial control system advisories on 22 September covering Siemens Industrial Edge Management, Desigo CC, SIPLUS and SIMATIC products, WTV676 and WTV776, SIMOVE Fleetmanager and SIPLANT, and Siveillance Control, along with the lwIP TCP/IP stack and its MQTT client application and OpenPLC Runtime v3. Several of the issues are remotely exploitable in components that are difficult to patch once they are in production, which is why they are worth reading before the next maintenance window rather than during it. Asset owners should map affected versions, apply vendor mitigations where updates cannot be scheduled, and confirm the affected interfaces are not reachable from untrusted networks.
Sources CISA: Siemens Industrial Edge Management|CISA: lwIP|CISA: OpenPLC Runtime v3
Linux kernel flaw CVE-2026-89775 lets ARM64 KVM guests read and write host memory Confirmed
A flaw in the Linux kernel’s KVM virtualisation code for ARM64 processors can leave a freed region of host memory exposed to a guest virtual machine on hosts with nested virtualisation enabled. The bug, CVE-2026-89775, allows a guest to read and write host kernel memory, and the researcher who found it says it can be used to escape the guest and run code on the host. Hosts running untrusted or multi-tenant ARM64 guests should track the upstream fix and treat guest escape as a plausible part of their threat model rather than a theoretical one.
Sources The Hacker News
Malicious npm package indexed-btree mimicked a popular utility and reached millions of downloads Confirmed
Researchers at Checkmarx and elsewhere reported that a malicious npm package named indexed-btree, which imitates the legitimate sorted-btree B-tree utility, hid its loader inside application code rather than in install lifecycle scripts, a change in technique that appears designed to sidestep recent registry and CI controls. The package accumulated millions of downloads before it was removed. Check lock files and build logs for the package, remove it where it appears, and rotate any secrets that were present in environments where it was installed or executed.
Sources SecurityWeek|The Hacker News
SharePoint flaw CVE-2026-65660 is an authenticated RCE, not the spoofing bug Microsoft listed Confirmed
A researcher at Viettel Cyber Security published technical details showing that CVE-2026-65660, which Microsoft initially classified as a spoofing flaw with a CVSS score of 6.5, in fact allows authenticated remote code execution on SharePoint Server 2016, 2019 and Subscription Edition. Patches are available, so the practical step is to confirm the relevant SharePoint updates are installed and to treat site collection administrators as privileged accounts whose sessions deserve protection. Prioritisation decisions taken when this was a 6.5 spoofing bug should be revisited now that it is an RCE.
Sources The Hacker News
Japan dismantles its first North Korean laptop farm as allies publish a joint report Confirmed
Japan has broken up its first known North Korean laptop farm, where remote IT workers obtained employment under false identities in order to send their earnings abroad, as the United States, Japan, Germany and Australia published a joint report detailing the WaterPlum campaign. The scheme places workers inside companies with legitimate access to code, systems and sometimes funds, which makes it a supply chain and insider risk rather than purely a financial crime. Employers hiring remote staff should verify identity and location independently, control what newly onboarded remote contractors can reach, and monitor for the indicators described in the joint report.
Sources SecurityWeek
SideCopy expands from government targets to Indian academic institutions Confirmed
Trellix researchers report that the SideCopy actor is using spear-phishing lures against academic institutions in India, broadening a campaign that had previously focused on government entities. The lures abuse mshta.exe to run malicious scripts and get past standard controls, and the final payload is the ReverseRAT backdoor. Universities and research bodies should treat unsolicited documents and links that lead to script execution as hostile, restrict or monitor mshta.exe in user contexts, and watch for outbound connections from workstations to unfamiliar infrastructure.
Sources The Hacker News
WordPress Comment2Shell flaw CVE-2026-93485 turned anonymous comments into server-side code execution Confirmed
Patchstack has detailed Comment2Shell, CVE-2026-93485, fixed in WordPress 7.1.1 on 17 September, which let an anonymous visitor leave a comment containing a crafted line break inside the attribute of an allowed HTML tag; when core reformatted the comment for display, the tag was split and the attacker’s text ended up where the browser treated it as an event handler that ran automatically as the page loaded. The script ran with the access level of whoever opened the page, and if a signed-in administrator opened it, the attacker could use that session to upload a plugin containing a web shell. The chain depended on a block theme, the default since Twenty Twenty-Five, and there is no evidence of exploitation. Confirm your sites are on 7.1.1 or later.
Sources The Hacker News
⚪ WATCH (6)
Rogue external MFA providers can capture passwords during normal sign-ins Confirmed
Researchers have demonstrated an attack in which an adversary who already holds privileged access registers a rogue external multi-factor authentication provider, then collects users’ passwords during otherwise legitimate sign-in attempts. It is a persistence and credential harvesting technique rather than an entry point, and it shows how a compromised administrator can subvert the controls meant to contain them. Restrict who can register and consent to authentication providers, review the providers already configured in your tenants, and alert on changes to authentication method policy.
Sources BleepingComputer
Dark Reading: more than 80,000 relay servers mask Chinese access to frontier AI models Confirmed
Dark Reading reports that more than 80,000 AI relay servers help users in China hide their identities while accessing cutting edge large language models, probably in order to clone them. Reselling access is a business model as well as an evasion technique, and it sits outside what most provider terms allow. Teams running model APIs should look for relay patterns in usage data, such as shared accounts accessed from dispersed locations, and enforce the account and rate controls their agreements assume.
Sources Dark Reading
White hats move 52 bitcoin from the Coldcard exploit into a recovery trust Confirmed
White-hat actors have consolidated 52 bitcoin linked to the Coldcard exploit into a fresh address carrying an OP_RETURN message pointing to a Crypto Recovery Trust, according to Galaxy Research, which says the funds represent about 2.8 percent of the total taken. The move gives affected holders a possible route to recovery, but the bulk of the haul remains under attacker control. Hardware wallet users should treat this as a reminder to verify key generation and entropy sources for themselves rather than trusting a device’s defaults.
US prosecutors examine whether Binance knowingly allowed Iran-linked transactions Under investigation
Federal prosecutors in Manhattan and the Department of Justice are reported to be investigating whether Binance knowingly allowed users to bypass US sanctions on Iran, with investigators asking whether the compliance changes the exchange agreed to in 2023 held. Binance says it has a zero tolerance policy for sanctions violations. This is an enforcement matter rather than a technical incident, but it lands on a compliance programme that much of the sector treats as a benchmark, and the outcome is likely to reset expectations for screening and transaction monitoring.
Malware can silence Meta's Muse assistant and redirect dictated prompts Confirmed
Researcher Patrick Wardle has shown in a proof of concept that malware already running on a Mac can change a hidden setting in Meta’s Muse assistant so that when the user dictates a prompt with the microphone, the words go to the attacker instead of to Meta, giving the attacker whatever access the user granted the assistant. The issue sits in a third-party assistant’s configuration rather than in macOS itself, and it needs an existing infection as a starting point. It is a worked example of why assistant and agent permissions should be scoped per application, and of why what those integrations can reach deserves a review.
Sources The Hacker News
US proposes an AI incident alert system in talks with China Confirmed
Treasury Secretary Scott Bessent said the United States has proposed an AI incident alert system in talks with China, which would give governments an early warning channel for significant AI failures or attacks. The proposal sits alongside domestic resistance to slowing AI development. Whatever comes of the talks, organisations deploying AI in regulated sectors should expect incident reporting expectations to extend to model behaviour and agent failures, not just conventional breaches.
Sources SecurityWeek
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.