OVHcloud
2021 · France (Strasbourg) Type: Major fire in a data centre — origin in the power room (UPS), Eco-Room design fed the fire
Impact: SBG2 (2 MW, ~30,000 servers) completely destroyed; SBG1 severely damaged; 100 firefighters and 44 vehicles fought the fire for six hours. Hundreds of thousands of websites and services offline simultaneously. A class action initiated by over 140 customers for damages exceeding EUR 10 million; in 2023 a French ruling ordered OVH to pay more than EUR 400,000 for lost backup data, showing that standard exoneration clauses of cloud providers are strictly limited under the European regime.
Lesson: A pivot incident for the sector. NIS2 Art. 21(2)(c) (continuity / crisis management) + Section 13 of the Annex to Reg. 2024/2690 (physical and environmental security) require N+1 or 2N redundancy on power + cooling, clean-agent fire detection and suppression, and separation of UPS rooms from the data hall. For customers: geo-redundant off-site backup is no longer optional — it is part of the shared-responsibility model.
Public source ↗ Amazon Web Services (AWS US-EAST-1)
2021 · United States (Northern Virginia) Type: Automated scaling activity caused an internal denial-of-service on the AWS network
Impact: 7 December 2021, ~10:30 EST: network devices connecting the internal Amazon network with the AWS network became overloaded; foundational services (internal DNS, authorisation, EC2 control plane, monitoring) were affected; full recovery at 14:22 PST. Customers affected globally: Amazon Connect, Disney+, Netflix, Robinhood, Tinder, Coinbase, iRobot, McDonalds app and others.
Lesson: Hyperscalers are not immune. NIS2 Art. 21(2)(b) (incident handling) + Section 3 of the Annex to Reg. 2024/2690 (incident handling) require playbooks with automatic triggers for cascade isolation. For Romanian cloud customers: multi-region + multi-cloud architecture is no longer exotic — it is part of the 'proportionate measure' against single-vendor risk, especially where the CSP is designated CTPP under DORA.
Public source ↗ Microsoft Azure (global Microsoft WAN)
2023 · Global (EU impact included) Type: A change on a Microsoft WAN router triggered rapid BGP prefix re-announcement → global routing churn
Impact: 25 January 2023, 07:05-12:43 UTC: packet loss up to 100% on Azure routes for many locations; affected services: Azure, Microsoft 365 (Teams, Outlook, SharePoint), Power Platform. The bulk of the incident lasted ~90 minutes, residual effects the next day. Microsoft explained that a router command was propagated across the WAN and all routers recomputed adjacency / forwarding tables almost simultaneously.
Lesson: Configuration and change management on the cloud backbone are not back-office procedures — they are part of NIS2 Art. 21(2)(e) and Section 6 of the Annex to Reg. 2024/2690 (secure acquisition / development). For a CSP at scale, any production change requires staged rollout, automated rollback and prior simulation on an isolated environment.
Public source ↗ CrowdStrike — impact on the global cloud + on-prem Windows ecosystem
2024 · Global Type: Faulty update to Channel File 291 in the Falcon sensor (20 vs. 21 input fields mismatch, out-of-bounds memory read)
Impact: 19 July 2024 — about 8.5 million Windows systems locked (BSOD); the largest IT outage in history; affected sectors: aviation (Delta — 7,000 flights cancelled, ~USD 500 million in losses reported via 8-K to the SEC), hospitals, banks, retail, governments. CISA issued an advisory on 19.07.2024; global damages estimated at tens of billions of US dollars.
Lesson: The same security vendor present on millions of systems = a single point of failure. NIS2 Art. 21(2)(d) (supply chain) + Reg. (EU) 2024/2847 (CRA) require SBOM, vulnerability disclosure and pre-deployment testing on any kernel-mode update. Operational takeaway: kill-switch for third-party kernel drivers + staged rollout (canary, regional, global); Microsoft announced it will move EDR / antivirus out of the Windows kernel.
Public source ↗ Snowflake (UNC5537 campaign targeting customer accounts)
2024 · Global (EU customers included) Type: Snowflake customer accounts compromised via stolen credentials (infostealer malware) + lack of MFA on the customer side
Impact: May-June 2024 — hundreds of Snowflake instances compromised; affected customers include Live Nation / Ticketmaster (declared in 8-K to the SEC ~560 million customer records), Santander, Advance Auto Parts, AT&T (call records), Pure Storage, Neiman Marcus, LendingTree. The UNC5537 actor (tracked by Mandiant) used Snowflake credentials without MFA.
Lesson: The shared-responsibility model has real limits. NIS2 Art. 21(2)(j) + Section 10 of the Annex to Reg. 2024/2690 require mandatory MFA on any access to critical systems — the cloud operator cannot fully transfer that responsibility to the customer. Snowflake subsequently announced it would enforce MFA by default on new accounts. The lesson for Romanian cloud operators: MFA on-by-default is the minimum standard, not an optional feature.
Public source ↗