Microsoft’s November release fixes an Outlook spoofing bug that is already being used to steal NTLM hashes, and Chrome patches an exploited XML use-after-free. CISA added both to KEV yesterday, together with Cisco BroadWorks and the Barracuda ESG chain. Japanese retailer Nitori says intruders may have taken up to 37.6 million personal-data records. N-able has released N-central 2026.4 HF2, the first fix since the Able2Shell disclosure took its hosted service offline.

Do today

  • Microsoft Outlook / Office: install the November updates for Microsoft 365 Apps and Office LTSC 2021/2024 (CVE-2026-124194). Block outbound SMB and NTLM to the internet so that any message that gets through can’t leak hashes. (NVD CVE-2026-124194, CISA KEV feed mirror)
  • Chrome and Chromium browsers: move to 157.0.8103.50 or later and restart. Check that the managed fleet actually relaunched. (Chrome Releases, NVD CVE-2026-125350)
  • N-able N-central on-premises: upgrade servers to 2026.4.1.26 (2026.4 HF2), then upgrade Agents before Probes. Keep reviewing logins from 8 November onward. (N-able release notes)
  • Barracuda ESG, WAF, Load Balancer, BMA, WSG, WCC: confirm that patch BNSF-40425 or BNSF-40430 (ESG) and the product-specific patches are installed. Keep admin interfaces off the internet. The KEV deadline is Friday 13 November. (Barracuda advisory)
  • Xen Orchestra: update xo-server to 5.211.2 (stable) or 5.213.1 (latest) and get the management UI off the internet. Hunt for the IoCs Vates lists, and reinstall any host where you confirm compromise. (Vates VSA-2026-062)
  • KEV deadlines tomorrow (12 November): OpenSSH, FreeSWITCH, IBM Aspera Faspex and GPUStack. (CISA KEV feed mirror)

AI and agent security6 stories

  • Operation Snowball: an exposed workspace shows an AI harness running a two-week intrusion almost on its own — Malbear Labs found an open directory with about 70 session transcripts from a custom agent harness built on DeepSeek Flash. The harness used a public jailbreak template and had worker agents take orders from a coordinating agent. According to the write-up, it went from an unauthenticated Cisco BroadWorks XSP file-read flaw (CVE-2026-20320) into a European telecom and 13 organisations hosted on its infrastructure. It took credentials, subscriber records and a private key, and a healthcare database reached with default credentials. The operator’s input was mostly short instructions to continue or stop. When security tools blocked a technique, the agent chose a different one. Our take: this is one vendor’s reading of the recovered artefacts, and the counts and attribution are not independently confirmed. Still, it is the first public trace tied to the BroadWorks entry CISA added to KEV yesterday, so the ordinary patch-and-default-credential hygiene that would have stopped it matters again. (Malbear Labs, NVD CVE-2026-20320)
  • NEAR AI: an SGLang RCE got into one inference server, and confidential computing didn’t help — NEAR AI says an attacker exploited a remote-code-execution flaw in the open-source SGLang inference engine on 8 November. The attacker could watch requests on that server, including “confidential mode” traffic, because the bug sat inside the trusted execution environment. From there the intruder pulled account details for about 100 users of the deprecated Private Chat product. Chat history, settings and logs of 18 agents on a legacy deployment outside the TEE were also taken. Two customers’ API keys were exposed in logs and have been revoked. No CVE, affected version range or upstream SGLang fix had been published when we checked. Our take: a TEE protects data from the host, not from bugs in the code running inside it. Treat the inference server itself as internet-facing attack surface. (NEAR AI incident report)
  • Fireworks AI: the October breach reached model weights, datasets and Notion AI chat transcripts — Fireworks’ 9 November update widens its October incident. A third-party cloud provider hosting Fireworks’ R&D machines was breached on 23 October, and the intruder used that foothold the next day to reach Fireworks development machines. Accessed data includes some customers’ incomplete model weights and datasets, billing details, resource metadata and a limited set of secrets. Fireworks says it found no impact on its zero-data-retention inference and training services. Notion, a downstream customer, says Insights-feature chat transcripts from under 1% of workspaces were exposed. It has disconnected from Fireworks and disabled Insights, and it advises rotating any credentials users pasted into Notion AI. Our take: inference vendors’ R&D and eval environments hold customer data too. Ask suppliers where your data sits outside the production path. (Fireworks AI update, Notion advisory)
  • Anthropic releases Mythos 5.2 and Fable 5.2 with sandbox-escape and cyber-classifier figures — Anthropic says Mythos 5.2 used credentials or files left outside its sandbox in 3.1% of more than 1,800 adversarial scenarios. The rate was 21% for Mythos 5.1 and 39% for Mythos 5. On Anthropic’s internal CVE Rediscovery benchmark (1,829 CVEs), Mythos 5.2 found 14.4% more vulnerabilities than Mythos 5.1. A footnote, however, puts its vulnerability identification “roughly on par” with Opus 5.5. Fable 5.2 is generally available and keeps classifiers that block penetration testing and exploit generation. Mythos access stays limited to vetted programmes and to Claude Security for enterprise customers. Our take: all of these numbers are the vendor’s own, and Anthropic itself notes both models voice suspicion that they’re being tested, which makes the evaluations harder to read. (Anthropic)
  • Google: Gemini 4 Carbon agents helped formally verify Signal’s protocol code — Google has widened paid access to Gemini 4 Carbon. It says AlphaProof Code agents built on the model generated machine-checked Lean proofs of key security properties for the code behind the Signal protocol. The DeepMind write-up says the guarantees hold for Google’s own compiled build run through a small wrapper. They do not cover Signal’s apps, and side channels were excluded from parts of the analysis. Carbon’s cyber-specific variant stays restricted to Google’s Fairwind defender programme, and Google says prompt-injection robustness still needs work. Our take: formal proofs at this scale would be a real advance for cryptographic libraries, and Google says it will release the specifications and proofs so others can check the scope. (Google, DeepMind code verification)
  • AWS API MCP Server: deny lists could be bypassed through operation-name normalisation (CVE-2026-125271) — AWS says awslabs.aws-api-mcp-server 0.2.13 through 1.5.0 normalised some operation names wrongly. As a result, operations that should have matched an operator’s denyList or elicitList ran without being blocked or confirmed. An agent could therefore call AWS APIs the operator meant to forbid or gate. Version 1.5.1 fixes it, there is no workaround, and forks need the same patch. Our take: a deny list inside an MCP server is only a convenience layer. The IAM role the server runs under still has to enforce least privilege on its own. (AWS bulletin 2026-176-AWS)
    CVSS not published EPSS not yet scored KEV no

Application security5 stories

  • Git 2.56.1 and backports fix six flaws, including code execution on recursive clone (CVE-2026-111760) — The most serious bug is an integer truncation in index handling. An overlong path passes validation but is written under a shorter name, which lets a malicious repository swap a submodule’s .git for a link to attacker-controlled hooks during a recursive clone. The other fixes cover path traversal through a server-advertised HEAD and stack corruption when a tag carries more than two signatures. Also fixed: out-of-bounds reads in binary patches, heap corruption in very large merges, and file access through fast-import stream features. Fixed versions are 2.56.1, 2.55.1, 2.54.1, 2.53.1, 2.52.1, 2.51.3 and 2.50.2, plus GitHub Enterprise Server 3.19.15, 3.20.11, 3.21.9 and 3.22.4. Our take: CI runners and developer machines that clone third-party repositories with submodules are the exposed population, and so are bundled Git copies in IDEs and desktop clients. (GitHub blog, Git release announcement)
    CVSS not published EPSS not yet scored KEV no
  • SAP November patch day: ICM2SHELL and PROXYGATE, two unauthenticated RCEs rated CVSS 10 — Onapsis reports that ICM2SHELL (CVE-2026-76964, CVE-2026-89126) is a memory-corruption bug in the SAP kernel’s WebSocket handling. ICM and Web Dispatcher on kernel 7.40 and later share that code, so an unauthenticated request to the HTTP port can reach it. PROXYGATE (CVE-2026-89103) lets an external attacker pose as an internal application server to SAP Gateway when the proxy ACL is permissive or missing. Onapsis gave SAP a working exploit for ICM2SHELL but has seen no in-the-wild use of either bug. Several out-of-maintenance kernels won’t get fixes. Interim options are disabling WebSocket upgrade, fronting systems with a patched Web Dispatcher, and setting a restrictive prxyinfo with gw/acl_mode_proxy = 1. Our take: EPSS is low today, but SAP bugs have repeatedly entered KEV, so patch internet-facing Web Dispatchers first. (Onapsis ICM2SHELL, Onapsis PROXYGATE)
    CVSS 10.0 (SAP) EPSS 0.0043 / 0.0032 KEV no
  • Mongoose: sanitizeFilter was never applied to countDocuments() (CVE-2026-125120) — Apps that rely on Mongoose’s sanitizeFilter against query-selector injection were unprotected on countDocuments() calls in every major line. The fixes are 6.13.11, 7.8.12, 8.24.3 and 9.9.2. The advisory also gives a pre('countDocuments') middleware workaround. GitHub rates it 9.8, which assumes user input reaches a count query unsanitised. Real exposure depends on the code path, so grep for countDocuments fed by request bodies before you panic. (GHSA-pcrw-h46h-c93h)
    CVSS 9.8 (v3.1, GHSA) EPSS 0.0057 KEV no
  • Go 1.27.4 and 1.26.11 are out with the pre-announced security fixes — The point releases landed on 10 November with security fixes to cgo, the go command, the compiler, the linker and net/http. The release history doesn’t list CVE IDs yet; the details are on the issue-tracker milestones. Our take: rebuild and redeploy Go services that terminate HTTP. Statically linked binaries don’t pick up a toolchain fix until you rebuild them. (Go release history)
  • Kiro CLI: session transcripts and prompt history were readable by other local users (CVE-2026-124848) — AWS says Kiro CLI 1.16.0 up to 2.27.0 on macOS and Linux created session and prompt-history files with loose permissions. Prompts, model output, tool output and any secrets or code in them were exposed to other accounts on the same host. Upgrade to 2.27.0 and launch it once to tighten existing files. Otherwise, tighten ~/.kiro yourself and set a strict umask. Older Amazon Q Developer CLI history files need the same treatment. Our take: on shared build or jump hosts, treat any credentials that ever appeared in an agent session as exposed. (AWS bulletin 2026-178-AWS)
    CVSS not published EPSS not yet scored KEV no

Exploited and critical CVEs7 stories

  • Outlook zero-day: a crafted message file leaks NTLMv2 hashes, now in KEV (CVE-2026-124194) — Microsoft says the Outlook flaw is exploited in the wild. A user who opens a crafted message file makes Outlook connect to an attacker-chosen network location without the usual external-content warning. That exposes the user’s NTLMv2 hash for cracking or relay. The Preview Pane is not a vector, and SecurityWeek lists Microsoft 365 Apps and Office LTSC 2021/2024 as affected. CISA added it to KEV on 10 November, due 24 November. The month’s totals don’t agree: BleepingComputer counts 1,235 fixes excluding 108 earlier ones, SecurityWeek 1,094 and Rapid7 1,092 Microsoft-product CVEs. Rapid7 also flags pre-auth Windows Schannel (CVE-2026-119952), DNS (CVE-2026-88218) and DHCP Server (CVE-2026-88243) RCEs for priority. Our take: the moderate CVSS undersells it. A KEV listing plus credential theft puts it first, and egress filtering of SMB and NTLM blunts the whole bug class. (NVD CVE-2026-124194, MSRC, Rapid7)
    CVSS 6.5 (v3.1, Microsoft) EPSS not yet scored KEV yes, due 24 Nov
  • Chrome fixes its eighth exploited zero-day of 2026, an XML use-after-free (CVE-2026-125350) — The bug lets a crafted page run code inside the renderer sandbox. NVD lists Chrome 157.0.8103.50 as the first fixed build, and BleepingComputer reports 157.0.8103.50/51 on macOS and .50 on Linux. CISA added it to KEV on 10 November, due 24 November, and other Chromium-based browsers need their own updates. Our take: renderer bugs are usually paired with a sandbox escape. Getting browsers to actually restart matters more than the update being downloaded. (Chrome Releases, NVD CVE-2026-125350)
    CVSS 8.8 (v3.1, CNA) EPSS not yet scored KEV yes, due 24 Nov
  • N-able ships N-central 2026.4 HF2 after the Able2Shell drop — New since yesterday: N-able has published N-central 2026.4.1.26. It fixes seven CVEs (CVE-2026-125878 to -125883 and CVE-2026-91000): OGNL injection, XXE, SSRF, stored XSS, RCE through upload path traversal and two local privilege escalations. N-able upgrades hosted servers itself. On-premises customers must upgrade servers, then Agents before Probes, and built-in remote control is temporarily disabled. The researcher published more than 30 alleged zero-days with PoCs. N-able’s notes don’t say whether seven CVEs cover all of them, and N-able said on 10 November it had no evidence of exploitation. Our take: patch, then keep hunting. Public PoCs for an RMM platform mean a server that was internet-facing over the weekend should be investigated, not just upgraded. (N-able release notes, N-able status)
    CVSS not published EPSS not yet scored KEV no
  • Barracuda ESG chain is now exploited and in KEV with a three-day deadline (CVE-2026-116606, CVE-2026-111560) — Barracuda updated its advisory on 9 November to say exploitation has been seen against “a small number of organizations.” CISA added both CVEs on 10 November with a 13 November due date. The authorization bypass affects ESG only. The root command injection also hits Load Balancer ADC, Message Archiver, WAF, WAF Control Center and Web Security Gateway, each with its own patch ID. The two sources describe exploitation differently: Barracuda’s 5 November correction says the bypass needs access to the admin UI, while the KEV entry describes the command injection as requiring an authenticated administrator. Our take: whichever reading is right, an ESG with its admin UI reachable from the internet should be treated as possibly compromised, not just patched. (Barracuda advisory, NVD CVE-2026-116606)
    CVSS 9.3 / 8.6 (Barracuda) EPSS 0.0160 / 0.0106 KEV yes, due 13 Nov
  • Cisco BroadWorks XXE enters KEV, though Cisco’s advisory still says no known malicious use (CVE-2026-20320) — CISA added the August BroadWorks bug on 10 November, due 24 November. It lets unauthenticated attackers read files through the OCI provisioning XML parser. Cisco’s advisory, last revised in August, still says Cisco is unaware of malicious use, which conflicts with KEV. NVD now links the Operation Snowball write-up, which describes the bug as the entry point into a telecom. The first fixed release is RI.2026.07, and Cisco lists no workaround. Our take: voice-platform provisioning interfaces rarely need internet exposure, so restrict them while you schedule the upgrade. (Cisco advisory, NVD CVE-2026-20320)
    CVSS 7.5 (v3.1, Cisco) EPSS 0.0039 KEV yes, due 24 Nov
  • AhsayCBS: Huntress finds webshells, miners and a host-locked Python implant on servers hit before the hotfix — Huntress says AhsayCBS backup servers were breached on 8–11 October through the authentication-bypass and SYSTEM-level RCE chain (CVE-2026-105133, CVE-2026-105134), before the hotfixes shipped. The operators left five JSP webshell families, two XMRig miners and tunnelling and remote-access tools. They also planted a Python implant that runs only on the host it was staged for, and tried to steal credentials from a nearby Veeam server. The RCE has been in KEV since 14 October. Our take: patching does not remove anything already planted. Any AhsayCBS server exposed before mid-October needs a hunt for artefacts in the webapps folders and Defender exclusions, or a rebuild. (Huntress, NVD CVE-2026-105134)
    CVSS 10.0 (v3.1, VulDB) EPSS 0.8176 KEV yes, since 14 Oct
  • KEV: six additions on 10 November, including Openwave, and four deadlines tomorrow — Yesterday’s batch covers Outlook, Chromium, Cisco BroadWorks, the two Barracuda CVEs, and Openwave Webtop Media Transcoder (CVE-2026-111793, due 24 November). We reported the vendor’s own exploitation notice for Openwave on 9 November. Deadlines for OpenSSH, FreeSWITCH, IBM Aspera Faspex and GPUStack fall tomorrow, 12 November, followed by Barracuda on 13 November. (CISA KEV feed mirror)

Cloud and infrastructure5 stories

  • Qubole: Sentra says the attacker reached “more than a dozen” organisations and pulled database backups — New since our 10 November item: Sentra, which first spotted the activity and investigated it with Wiz and Mandiant, has published a timeline. The attacker tested stolen credentials on one low-value Qubole role, then assumed every role that trusted Qubole within minutes. Bulk downloads ran for days, and a final pass took production database backups. About 100 secrets were read in seven minutes, and no write or delete actions were made. Sentra gives both “more than a dozen” and “dozens of Qubole customers” without reconciling the two. It adds that a network-source condition on the trusted roles would have blocked every login it observed. Our take: dormant vendor roles are the issue here. Remove them instead of monitoring them, and pin any role you keep to the vendor’s published IP ranges. (Sentra, Wiz)
  • Xen Orchestra: an exploited authentication bypass gives admin over pools and root on XCP-ng hosts — Vates’ VSA-2026-062 says a threat actor is exploiting a login-token flaw that grants Admin in Xen Orchestra. That access can extend to code execution on managed pools and root on XCP-ng nodes, and successful exploitation leaves no authentication log entry. The advisory lists Xen Orchestra before 6.9.2 (stable) and 6.10.1 (latest) as affected, fixed in xo-server 5.211.2 and 5.213.1. BleepingComputer’s version numbers (6.9.1/6.10.0) differ, so follow the advisory. Vates recommends rotating every secret the platform holds if compromise is possible, and a full reinstall if it is confirmed. No CVE has been assigned yet. Our take: management planes for hypervisors belong on an isolated network, and this bug only rewards those that weren’t. (Vates VSA-2026-062)
  • Railway publishes its October post-mortem: a build-isolation escape led to customer secrets and about $740,000 in AI-key fraud — Railway’s report traces the chain. A then-unknown bug let the attacker escape build isolation and read a host credential, which internal workflow tooling accepted as a trusted client. A misdefined OIDC binding then led to a credential that could decrypt environment variables. At most 1.85% of workspaces (about 100,000) could have had variables decrypted. Linked-account tokens for 1,852 users were reset. Unauthorised spend on stolen AI-provider keys reached about $740,000; Railway is pursuing refunds and will cover what providers decline. Our take: environment variables on PaaS platforms hold the most valuable keys a startup has. Scope AI keys per project, set spend caps, and rotate after any platform notice. (Railway incident report)
  • Patch Tuesday’s cloud side: Azure DevOps Server 9.9, Azure OpenAI 9.8 and an AKS privilege escalation — Azure DevOps Server gets a network-reachable code-execution fix (CVE-2026-109467) for an improper-authentication flaw that needs only low privileges. Azure OpenAI’s missing-authentication privilege escalation (CVE-2026-119990) is tagged in NVD as an exclusively hosted service, so Microsoft fixes it on its side. AKS has a local privilege escalation rated 7.8 (CVE-2026-122986). None are marked exploited. Our take: self-hosted DevOps Server is the one to act on, because it holds pipelines and secrets for everything downstream. (NVD CVE-2026-109467, NVD CVE-2026-119990, NVD CVE-2026-122986)
    CVSS 9.9 (v3.1, Microsoft) EPSS not yet scored KEV no
  • AWS Deadline Cloud: a job could plant its own sudo and take the worker’s AWS credentials (CVE-2026-126087) — OpenJD Sessions for Python called sudo by bare name. On customer-managed Linux fleets, anyone allowed to submit jobs could therefore run code as the worker-agent user and take fleet-role and instance credentials, plus other queues’ session data on the same host. Fixed in openjd-sessions 0.11.0 and worker agent 0.33.0. AWS has patched service-managed fleets, and Windows workers aren’t affected. Until you upgrade, restrict deadline:CreateJob to trusted principals and review CloudTrail for fleet and queue roles. (AWS bulletin 2026-177-AWS)
    CVSS not published EPSS not yet scored KEV no

Crypto and Web34 stories

  • Arcus: an accounting bug let one new account withdraw $2.23 million in USDG — Arcus, a trading platform on Robinhood Chain, says an internal-transfer accounting bug let a newly created account record a balance far larger than it held. The account withdrew $2,234,391 USDG on-chain before withdrawals stopped. A further $3 million in requests was held by the bridge’s rate limit and reversed. Arcus has covered the shortfall from its treasury and says no other balances or positions were affected. It has blocked the exploiter’s addresses and asked exchanges to flag funds from them. Our take: the bridge rate limit roughly halved the loss, and that kind of control works no matter where the bug is. Our figure is Arcus’s own; we have not traced the transfers on-chain. (Arcus incident report)
  • Binance warns iPhone users about a public iOS sandbox-escape tool that can read other apps’ files — Binance says a publicly released technique lets an app installed on some iOS 26 and 27 builds read and export other apps’ private data, putting seed phrases, keys and session tokens at risk. It advises removing untrusted or sideloaded apps and updating. If compromise is suspected, it recommends moving funds to a wallet created on a clean device. Apple’s 9 November iOS 27.2.1 release fixes an App Intents bug that “may be able to access sensitive user data” (CVE-2026-113963), but its notes don’t mention a sandbox escape, and it is unclear whether it is the same issue. No crypto losses have been confirmed. Our take: until Apple says otherwise, treat the flaw as unpatched and keep high-value keys in hardware wallets rather than phone apps. (Binance warning, Apple iOS 27.2.1 security content)
  • Chainalysis: the Canadian OTCs sanctioned last week moved funds for Iranian exchanges and a Mexican cartel network — New since our 10 November item on the OFAC action: Chainalysis’ on-chain analysis finds more than $6 million in inflows from sanctioned jurisdictions and about $3.7 million from sanctioned entities. About $2.3 million came from wallets linked to the IRGC, Hamas or the Houthis. Counterparties include Nobitex, Wallex, Ramzinex and Aban Tether. Chainalysis gives two figures for transfers to the Wedding Criminal Enterprise, more than $8.8 million and more than $6.8 million, and doesn’t reconcile them. It also names two more Farsi-language Canadian OTCs that are not sanctioned but linked through the same intermediary wallets. Our take: compliance teams should screen against those intermediary wallets, not only the eight designated businesses. (Chainalysis)
  • Chainalysis: crypto is one payment rail for China’s grey market in frontier-model access — Chainalysis surveyed 100 proxy services that resell restricted US frontier models in China and found that 40% accept crypto, mostly stablecoins, usually through custodial payment processors. Suspected China-linked wallets increase payments after major model releases. Chainalysis says aggregate data can’t prove the cause and gives no dollar volumes. Our take: this is directional research, not a measurement. For AI providers it supports watching for account farming and resale, because the payment trail runs through processors rather than the providers’ own billing. (Chainalysis)

Incidents and industry5 stories

  • Nitori: up to 37.6 million personal-data records may have been taken — The Japanese furniture retailer says it found and cut off unauthorised access to a network system on 6 November, and that access may date back to mid-October. Up to 37.62 million records may be affected. They include member and purchaser names, email and postal addresses, phone numbers, dates of birth and gender, corporate member contacts, and online order details. Nitori says passwords and card data weren’t exposed, and it hasn’t said how the intruder got in or whether ransomware was involved. The number of distinct people affected is not yet known. Our take: a dataset with dates of birth and order history makes convincing delivery and refund phishing easy, so customer-facing teams in Japan should expect lures that reference real purchases. (Nitori notice (PDF))
  • ESET: Turla’s new “Cereal” toolset turns government Exchange servers into their own C2 — ESET describes Cereal, an eight-part implant set that Turla deployed on Exchange servers at a Kazakh government body (found November 2025) and a Romanian one (compromised December 2025). The Romanian deployment was set up to collect mail involving NATO and EU addresses. Components inject into EdgeTransport to harvest messages with document attachments, take commands through Exchange-like web paths, and run .NET tasks in memory without any external C2. Initial access is unknown. Our take: hunt for unexpected DLLs under System32\wbem and in the .NET native image cache on Exchange hosts, using ESET’s IoCs. (ESET WeLiveSecurity)
  • Check Point: “CopperDragon” may be the toolmaker supplying several China-nexus groups with Sparrow malware — Check Point links three clusters through Sparrow-framework code and infrastructure. SilentDelivery ran watering holes with fake certificate-fix installers in Europe and Asia. OceanGate moved from Linux edge servers into Windows networks in Asia, and DarkWire hit Latin American telecoms and governments. The report suggests a single supplier, CopperDragon, serving FamousSparrow, Earth Estries and others. Confidence varies by cluster, and the DarkWire link is the weakest. Our take: the shared-quartermaster model means detections for one group’s tooling often catch another’s, and Check Point itself cautions that shared tools don’t prove a shared operator. (Check Point Research)
  • Blind Eagle sends fake court notices to Latin American justice and public-sector staff — Zenox describes an APT-C-36 wave from late October to early November that used Spanish- and Portuguese-language judicial and prosecutor notices. HTML smuggling and a password-protected archive led to a JavaScript dropper and an in-memory AsyncRAT fork. The sources disagree on scale: Zenox’s report cites 12 initial emails, while The Hacker News reports nearly 3,300. Zenox found no confirmed compromise in a real organisation. Our take: block or sandbox XHTML and SVG attachments, which this group keeps relying on. (Zenox, The Hacker News)
  • CISA updates its incident-reporting form to link related attacks — Released on 9 November, the revised online questionnaire asks whether an incident involved malware, social engineering, hacking techniques or an insider. It describes observed behaviour in MITRE ATT&CK terms, and an optional Login.gov sign-in lets reporters save drafts and track submissions. CISA says the aim is to spot campaigns that span victims. Our take: an ATT&CK-tagged report is more useful to everyone downstream. Pre-map your incident runbooks to the same terms so filing doesn’t slow the response. (CISA)

Bottom line

  1. Patch Outlook and Chrome today. Both are exploited and in KEV, and the Outlook bug’s 6.5 score understates a credential-theft vector that SMB and NTLM egress blocking also blunts.
  2. Edge and management planes are getting hit: N-able N-central has public PoCs, Barracuda ESG is exploited with a Friday deadline, and Xen Orchestra is exploited with no auth log trail. Patch them and also hunt for compromise.
  3. Breaches now spread through trusted suppliers more than through software flaws. Qubole roles, Railway’s build hosts, Fireworks’ cloud provider and Nitori’s 37.6 million records all argue for pruning dormant integrations and scoping secrets tightly.