Blog

    In-house analysis and situation reports from the NEOSEC Intel newsroom. Deeper than Twitter, shorter than a whitepaper.

    • NEOSEC

      APT28 at a NATO naval shipbuilder: two months inside, mostly using built-in Windows tools

      Symantec has documented an attack by the Russian group APT28 on a naval shipbuilder in a NATO member state. After initial access via a Word macro, the attackers relied mainly on tools already present on Windows clients: Edge, curl, the OpenSSH client and PowerShell. These steps can be detected with widely deployed endpoint telemetry.

      APT28Threat IntelligenceRüstung
    • NEOSEC

      php2shell (CVE-2026-109618, CVE-2026-109619): background, patches and detection

      On 21 October 2026, PHP released security updates for two vulnerabilities rated CVSS 9.2. Practically all PHP versions are affected; fixes exist only for the 8.2 to 8.5 branches. Technical details were public before the release. Under Apache with mod_php and under LiteSpeed, updating the package is not enough; the PHP processes must be fully restarted.

      PHPZero-DayRCE
    • Benjamin Espagné

      Europol has counted the gaps in digital forensics. You have most of them too.

      The European Commission and Europol examined where law enforcement fails with digital evidence. The list reads like an audit of mid-sized corporate IT – and it says what needs doing before the incident.

      DFIRDigitale ForensikEuropol
    • NEOSEC

      Microsoft Patchday September 2026: 974 Schwachstellen, zwei aktiv ausgenutzt

      Microsoft schließt im September einen Rekordzyklus. Je nach Quelle sind es 966, 970, 972, 974 oder über 1.100 Schwachstellen. Keine dieser Zahlen ist falsch, und keine beantwortet die Frage, die im Betrieb zählt.

      MicrosoftPatchdaySchwachstellenmanagement
    • NEOSEC

      StyleSmuggler (CVE-2026-75650): Patchen allein reinigt keinen Magento-Shop

      Adobe hat am 7. September einen Notfall-Hotfix für eine Magento-Lücke mit CVSS 10.0 veröffentlicht. Angegriffen wurde bereits seit dem 4. September. Wer jetzt nur patcht und den Vorgang abhakt, übersieht den entscheidenden Teil.

      MagentoAdobe CommerceZero-Day
    • NEOSEC

      SAP Patchday September 2026: CVSS 10.0 im Kernel, kein Workaround

      SAP schließt am 8. September 17 Schwachstellen, vier davon kritisch. Die schwerste erreicht CVSS 10.0, ist vorauthentifiziert ausnutzbar und lässt sich nicht abschalten. Der Fix verlangt einen Kernel-Austausch — und damit ein Wartungsfenster, das Wochen entfernt liegt.

      SAPPatchdayNIS2
    • J. Benjamin Espagné

      CVSS 10.0, Priority Score 0.00 — and Still This Week's Most Urgent Patch

      N-able is patching out of band: CVE-2026-86218 allows code execution on the N-central server with no login at all. CVSS v4 10.0 — alongside an EPSS of 0.4 percent and a priority score of 0.00. Why both figures are correct, why the flaw still needs closing today, and what the case reveals about the limits of any external prioritisation.

      schwachstellenmanagementrmmn-able
    • J. Benjamin Espagné

      The First 24 Hours After an AI Agent Incident

      An agent holding live credentials does something nobody authorised. What happens in the next 24 hours decides both the damage and your ability to notify — and conventional playbooks miss, because they reach for host isolation instead of revoking an identity. The sequence hour by hour, with the second clock that runs in parallel in Europe: 24 hours to the NIS2 early warning, and the logging duty under Article 12 of the AI Act.

      ai-securityincident-responseagenten
    • NEOSEC Intel Redaktion

      OT Remote Access: How the VPN Became the Attack Surface

      BeyondTrust puts more than half of 2024's ransomware incidents on a remote service as the entry point — a VPN appliance, an RDP server. In OT environments those are the same tunnels and jump hosts that keep production running, and they open the first door for attackers.

      otnis2kritis
    • NEOSEC Intel Editorial

      Mirage2FA: The Victim Completes MFA Correctly — and the Attacker Takes the Session

      ANY.RUN has documented Mirage2FA, a phishing kit rented out as a service that does not break the Microsoft 365 sign-in but relays it: the victim completes MFA correctly and the attacker keeps the session cookie. Why a password reset afterwards achieves nothing, what one mail-gateway rule removes — and why 1 September settles the decision regardless.

      phishingaitmmicrosoft-365
    • NEOSEC Redaktion

      Ten Days, Eight Hours — and Your Logs Are Gone

      The SIRIUS Report 2025 from Europol and Eurojust describes how investigators obtain digital evidence. For companies it holds more than the title suggests: it is about your own log retention, about how well a criminal complaint holds up — and about who has been on the receiving end of binding orders since 18 August.

      e-evidencenis2dfir
    • NEOSEC Threat Intel

      Dysphoria: 296,000 IoT Devices Abused for DDoS and Residential Proxying

      Shadowserver reports 296,000 compromised IoT devices, every entry rated CRITICAL. Dysphoria has grown from a DDoS botnet into relay infrastructure, and one variant has dropped the attack function altogether. Why 155 UPnP port mappings are the best detection signal, and what network owners should check now.

      botnetiotthreat-intel
    • NEOSEC Redaktion

      Cybersecurity Is Survival

      Howard Solomon's CSO Online piece says plainly what the industry thinks and rarely states: an organisation that still budgets security mainly as prevention is buying theatre. His most uncomfortable point lands squarely in the European scoping debate — companies work actively to avoid being seen as critical, because the label brings obligations and liability.

      nis2resilienzciso
    • NEOSEC Redaktion

      Google Ads as Malware Delivery: The MacSync Stealer Case

      A Google ad impersonating Anthropic's Claude Code sends macOS users to a Google Sites page carrying an obfuscated terminal command. In six phases, browser passwords, cloud tokens and — where present — the seed phrase of a Ledger wallet end up with the attacker. What the case means for developers and their employers, and the one question to settle first after a suspected infection.

      malvertisingmacosstealer
    • NEOSEC Redaktion

      Did Iran Take Its Cue from China? The Barati Arrest and the Mabna Institute

      An Iranian hacker arrested in Montenegro, an eight-year-old US case, more than 31 terabytes of stolen research data. Kim Zetter asks whether Tehran began building an economic-espionage capability around 2013 — the same year Mandiant exposed China as APT1. What the case means for European universities and research institutions.

      iranaptwirtschaftsspionage
    • NEOSEC Intel-Redaktion

      22,000 breaches – what they reveal about real incident preparedness

      The Verizon DBIR 2026 analyses more than 22,000 confirmed security incidents across 145 countries. Three findings change the terms for mid-market organisations: exploitation of vulnerabilities is now the leading initial-access route for the first time, nearly half of all incidents run through a third party, and AI-assisted attackers are closing the patch gap faster than before.

      incident-responseransomwaresupply-chain
    • J. Benjamin Espagné

      Trust Issues: How MCP Servers Can Hijack Your AI

      Every MCP server is a trust boundary. Since April the specification has fixed authorization — which does nothing for tool poisoning or prompt injection. New since then: Microsoft's demonstration that poisoned tool descriptions are a general-purpose exfiltration technique, and the finding that nine of eleven MCP registries waved a malicious submission through unreviewed.

      ai-securitymcpprompt-injection
    • J. Benjamin Espagné

      The Augmented Analyst — and Why ISC2 No Longer Puts a Number on the Workforce Gap

      ISC2 dropped its workforce gap estimate in 2025 — not because the gap narrowed, but because respondents consider skills more pressing than headcount. That changes the case for AI in the SOC. Plus the attack path few people mention: alert text is largely attacker-controlled, and the triage layer reads everything by definition.

      ai-securitysocalert-fatigue
    • J. Benjamin Espagné

      48,185 CVEs: Why Prioritisation Is No Longer a Question of Diligence

      48,185 CVEs were published in 2025, and nearly 10,000 in July 2026 alone. More important than the volume is what changed in April: the NVD now enriches only 15 to 20 percent of intake, and Mandiant puts the mean time to exploit at minus seven days. Why CVSS no longer carries prioritisation — and why the KEV catalogue does not measure what many read into it.

      schwachstellenmanagementcveepss
    • J. Benjamin Espagné

      BSI C5:2026 — and How to Read a Cloud Attestation Properly

      C5:2026 adds container management, confidential computing, post-quantum cryptography and substantially expanded supply chain criteria. More important than the additions is what to do with an attestation: type 2 rather than type 1, read the exceptions, check the scope — and work through the obligations the report assigns to the customer. Plus a correction: the reference to § 8a BSIG no longer holds after the German NIS2 transposition.

      nis2compliancecloud
    • J. Benjamin Espagné

      XIEM® or Splunk: The Question Is Not the Tool, but Who Operates It

      A comparison in our own interest, openly labelled as such. With a correction on licensing — Splunk now also bills by workload, not only by volume — and with the section missing from the original: when XIEM® is the wrong choice. The decisive question is not which SIEM is better, but who operates it the day after deployment.

      siemxiemsplunk
    • J. Benjamin Espagné

      Detection Coverage: 21 Percent — and 13 Percent of Rules Never Fire

      In production environments, detection rules exist for 21 percent of ATT&CK techniques — while the telemetry already ingested would support over 90 percent. And 13 percent of existing rules are broken and never fire. The second figure is the more dangerous one: in any heatmap it looks exactly like coverage.

      detection-engineeringmitre-attacksiem
    • J. Benjamin Espagné

      Nineteen Days Without a Model: What the Mythos Suspension Teaches About AI Dependency

      In April the subject was a model that finds vulnerabilities autonomously. In June, Mythos 5 and Fable 5 were disabled worldwide under US export controls — nineteen days, across every platform at once, non-US users first. The real lesson is therefore not about the speed of attacks but about the fact that an AI model is a supplier.

      ai-securitylieferkettenis2
    • J. Benjamin Espagné

      "A Board Matter" Means Article 20: What NIS2 Requires of Management Personally

      Since NIS2, "a board matter" is no longer a question of attitude but a set of duties with a citation. The decisive provision is not the much-quoted Article 21 but Article 20: approval, oversight — and a training obligation that applies to management itself, not only to staff. Plus six questions where the time taken to answer is already the answer.

      nis2governancehaftung