Digital Plumber

Archives
Log in
Subscribe
November 17, 2026

Nov 17: Route leak hit Cloudflare 1.1.1.1 despite valid ROA

Digital Plumber

Plumbing the information age

Tuesday, November 17, 2026  ·  No. 156  ·  12 stories

Today's 3 things that matter

  1. Route leak hit Cloudflare 1.1.1.1 despite valid ROA
    This incident demonstrates that RPKI ROV validation alone is insufficient—networks must actively implement RPKI filtering on received routes; many operators accept RPKI-invalid announcements, undermining the security model and requiring defense-in-depth with prefix filtering and monitoring.
  2. FRRouting bgpd crashes on dynamic neighbour disconnects
    Any external host within the listen range can trigger total BGP session loss by opening and closing two TCP connections to the BGP port without authentication; core routing infrastructure operators must patch FRRouting immediately or reconfigure to avoid listen ranges with Site-of-Origin values.
  3. Cribl buys Tero and hires Vector creator Ben Johnson
    Consolidates observability and security on a single telemetry platform as enterprises shift toward agentic AI monitoring; extends Cribl's capability for cost-optimized observability critical to NetDevOps and SRE workflows.

Full stories below, grouped by topic.


Network Automation

OpsMill publishes 2026 Nornir roadmap for community feedback

OpsMill · Nov 16, 2026 · Primary source

What happened: OpsMill has published a 2026 roadmap for Nornir, the Python-first network automation framework, proposing additions for an upcoming release while maintaining backward compatibility with existing code.

Why it matters: Nornir practitioners should review the roadmap to plan migrations and understand future evolution of this Terraform/Ansible alternative.

OpsMill, the maintainers of Nornir, have opened the 2026 roadmap to community feedback. The proposal emphasizes that all additions will be additive, ensuring existing automation code continues to work without breaking changes. This signals Nornir's maturation as a production-grade framework. For network engineers considering Nornir as a Terraform or Ansible replacement, particularly those managing large-scale, performance-sensitive automation workflows or requiring deeper Python-level control, this roadmap signals stability and active development. The timing during AutoCon 6 season (Nov 16-20 in Tucson) suggests the community is discussing future directions for Python-based network automation alongside broader DevNetOps tooling.

Read the original at opsmill.com


Data Center & AI Networking

Broadcom pitches disaggregated AI Ethernet fabric against Cisco Silicon One

TechBlog ComsOC · Nov 16, 2026 · Analysis

What happened: Broadcom presented its disaggregated AI fabric strategy at SC26 combining Tomahawk 6 (102.4 Tbps), Tomahawk Ultra (low-latency intra-rack), Jericho 4 (deep-buffer routing), and Thor Ultra 800G NIC with UEC compliance, contrasting with Cisco's unified Silicon One approach.

Why it matters: Network architects must choose between Broadcom's specialized role-based ASICs (optimized per function) versus Cisco's unified programmable silicon; Broadcom's approach reduces switching latency and power per function, critical for 100k+ GPU clusters.

Broadcom demonstrated a disaggregated ASIC strategy specifically designed for massive AI cluster networking at SC26 on November 16. The strategy pairs four specialized silicon components: Tomahawk 6 (102.4 Tbps leaf-spine fabric switch with shallow buffers and high radix for scale-out), Tomahawk Ultra (51.2 Tbps with ~250ns latency and credit-based flow control for intra-rack GPU memory scale-up), Jericho 4 (deep-buffer scale-across spine router with HBM for congestion recovery), and Thor Ultra (800G NIC with full UEC compliance and advanced RoCE). This contrasts with Cisco's unified, P4-programmable Silicon One architecture (G300/G302 series offering 102.4 Tbps with shared packet buffer and path-based load balancing). Broadcom's approach optimizes silicon area and power efficiency by role: Tomahawk 6 keeps switch pipelines lean with shallow buffers for high-radix leaf-spine fabrics, offloading deep buffering to Jericho 4 and congestion recovery to Thor Ultra endpoints. The strategy combines packet-level multipathing, selective retransmission, and programmable congestion control. Network operators choosing between these architectures must weigh Broadcom's lower latency and power per function against Cisco's operational simplicity from unified management and P4 programmability across roles.

Read the original at techblog.comsoc.org


Routing & Internet

Arista EOS flaws let attackers bypass RPKI validation

Arista · Nov 10, 2026 · Vendor release

What happened: Arista released four security advisories on November 10, 2026, covering vulnerabilities in EOS's RPKI client and RIPv2 authentication. The RPKI vulnerabilities allow attackers controlling the cache or router-to-cache traffic to send crafted messages causing incorrect validation results or RPKI agent reset.

Why it matters: These CVEs expose operational risks for networks relying on RPKI validation and RIPv2 authentication; adjacent-network attackers can interrupt routing services or inject unauthorized routes, directly undermining BGP security posture and RPKI ROV deployment confidence.

Arista security advisories 0226, 0223, and 0222 (all dated November 10, 2026) describe vulnerabilities in EOS's Resource Public Key Infrastructure (RPKI) client, which obtains Route Origin Authorization (ROA) data from an external cache for BGP route validation. An attacker controlling the RPKI cache or router-to-cache traffic can send a crafted RPKI message that causes incorrect validation results or resets the RPKI agent. Advisory 0221 covers two vulnerabilities in RIPv2 MD5 authentication on Arista EOS-based products, whereby an unauthenticated attacker with adjacent network access may interrupt routing services or inject unauthorized routes. Advisory 0220 warns that on affected EOS platforms with RIPv2 configured, RIPv2 might be silently enabled on management interfaces even when disabled, allowing attackers with adjacent network access to send crafted RIPv2 route advertisements to management IP addresses that devices accept and install without authentication. These issues are critical for operators running RPKI ROV validation and present lateral attack surfaces on out-of-band management networks.

Read the original at arista.com

Route leak hit Cloudflare 1.1.1.1 despite valid ROA

LogicMonitor · Nov 13, 2026 · Analysis

What happened: On November 13–14 UTC, a route leak/hijack of Cloudflare's 1.1.1.0/24 DNS resolver prefix degraded Canadian resolver reachability. Despite Cloudflare holding a valid RPKI ROA for the prefix, the hijacked route originated by AS4755 was marked RPKI Invalid but still propagated and was accepted by networks including AS6453.

Why it matters: This incident demonstrates that RPKI ROV validation alone is insufficient—networks must actively implement RPKI filtering on received routes; many operators accept RPKI-invalid announcements, undermining the security model and requiring defense-in-depth with prefix filtering and monitoring.

During November 13–14, 2026, a route leak affecting Cloudflare's 1.1.1.0/24 DNS prefix degraded resolver reachability in Canada. Cloudflare held a valid RPKI ROA for the prefix, authorizing only AS13335 (Cloudflare) to originate it. The hijacked route originated by AS4755 was clearly marked RPKI Invalid in RouteViews collectors. However, despite this validation failure, the announcements were accepted and propagated by networks including AS6453, and many nodes withdrew the legitimate route via AS13335. Some ISPs like Lumen and AT&T had no routes for 1.1.1.0/24 at all, suggesting they honored RPKI filtering and dropped invalid announcements. Tata itself is known to apply RPKI-based filtering, yet its own RPKI filtering did not catch this announcement. This incident exposes a critical gap: RPKI ROV validation works technically, but most networks do not implement mandatory RPKI filtering on received routes, allowing invalid announcements to propagate and disrupt services even when ROAs are properly configured.

Read the original at logicmonitor.com


Network Security

Windows networking bug caused 24-hour Microsoft 365 outage

Microsoft Status Page / Multiple postmortems · Nov 13, 2026 · Primary source

What happened: A hidden Windows bug overwhelmed servers handling cloud connections across Microsoft 365, lasting from 4:20 PM UTC on November 12 until 4:10 PM UTC on November 13. Exchange Online, Outlook, Teams, and Copilot were affected globally, with over 80% of affected gateway servers stopping responding at peak.

Why it matters: Around 430 million Microsoft 365 users across 70+ countries experienced service loss; demonstrates how cascading failover in converged infrastructure amplifies failures and requires network teams to audit automatic health-check and traffic-rerouting logic.

The outage was caused by servers unexpectedly failing automated health checks and dropping out of service, which pushed more load onto remaining servers until they were overwhelmed. Users began losing Outlook access at 16:17 UTC on November 12, with 18 other services including Teams, Copilot and Intune also affected. Microsoft paused deployments, moved traffic between regions, restarted servers, tried alternative health checks, and reverted a recent network change. The root cause was a latent Windows defect on networking gateway servers triggered by a traffic shift and reduced capacity from an earlier network upgrade. This event illustrates a critical infrastructure risk: automated failover systems can amplify failures when triggered by systemic connectivity issues. Network teams running similar converged cloud platforms must review whether health-check logic can enter feedback loops and whether traffic rerouting has capacity buffers.

Read the original at en.wikipedia.org

FRRouting bgpd crashes on dynamic neighbour disconnects

CVE.org / FRRouting Project · Nov 13, 2026 · Primary source

What happened: Double free in FRRouting bgpd (CVSS 8.2 High) when dynamic BGP neighbors disconnect causes denial of service. When a peer-group used by 'bgp listen range' has Site-of-Origin configured, dynamic neighbors share the SoO value but each frees it on disconnect, causing double-free and bgpd abort.

Why it matters: Any external host within the listen range can trigger total BGP session loss by opening and closing two TCP connections to the BGP port without authentication; core routing infrastructure operators must patch FRRouting immediately or reconfigure to avoid listen ranges with Site-of-Origin values.

Each dynamic neighbor shares the peer-group's SoO value instead of getting its own copy, but frees it as its own when deleted; the first disconnect leaves the peer-group pointing at freed memory, and the second frees it again, causing bgpd to abort and drop all BGP sessions. Any host whose address falls inside the listen range can trigger this by opening and closing two TCP connections without sending any BGP message and without knowing the AS number, and can repeat it each time bgpd restarts. If the peer-group has TCP-MD5 password, only password-knowing hosts can trigger it. Between the two frees, displaying configuration or establishing a dynamic neighbor also reads freed memory, which can crash bgpd after a single disconnect. Only configurations combining listen range with SoO on the same peer-group are affected. Affects FRRouting 8.4 and later with no public exploit yet, but exploitation requires no authentication and can be triggered by any external host with BGP port access.

Read the original at vulners.com


Startups & Funding

Cribl buys Tero and hires Vector creator Ben Johnson

Cribl · Nov 10, 2026 · Primary source

What happened: Cribl acquired Tero, an observability startup that identifies low-value telemetry data, bringing Tero's three co-founders including Vector creator Ben Johnson. The deal marks Cribl's third acquisition in four months, adding host monitoring, APM, and AI-native observability capabilities.

Why it matters: Consolidates observability and security on a single telemetry platform as enterprises shift toward agentic AI monitoring; extends Cribl's capability for cost-optimized observability critical to NetDevOps and SRE workflows.

Cribl, a San Francisco-based telemetry platform, announced the acquisition of Tero on November 10, 2026, bringing co-founders Ben Johnson (CEO, formerly of Timber/Vector), Adam Cohen (CTO), and Nihar Singhal (COO) into the fold. Tero's core technology identifies and flags low-value observability data—telemetry that systems produce but engineers never examine—a critical problem as observability costs consume 20-30% of cloud infrastructure budgets. Johnson becomes VP of observability and leads Cribl's roadmap for host monitoring, application performance monitoring (APM), and federated observability. This is Cribl's third acquisition in four months, following CardinalOps (July 2026, security workflows) and Radiant Security's AI SOC technology (August 2026). Together, these acquisitions position Cribl to deliver first-party observability and security intelligence on a single open platform. The timing matters: observability platforms are racing to consolidate capabilities for the AI era, where both engineers and autonomous agents need context—Datadog, Dynatrace, Elastic, and Splunk are making similar AI-related moves. For practitioners managing large-scale observability deployments or building AIOps workflows, this signals that unified telemetry routing plus intelligent data filtering is becoming table stakes.

Read the original at cribl.io


Agentic AI & MCP

LogicMonitor merges AI agent and MCP usage into one credit pool

LogicMonitor · Nov 16, 2026 · Primary source

What happened: LogicMonitor consolidated its platform tiers to standardize AI as core operations. Diagnostics, remediation, agent usage, and MCP access now share one credit pool, eliminating fragmented pricing that previously blocked agentic operations adoption.

Why it matters: MCP access moves from point-solution to unified operational control plane; agents gain governed access to 30+ tool categories while respecting RBAC and resource groups—addressing the governance gap where 71% of orgs plan agentic AI but only 21% have governance.

LogicMonitor's November 16 release consolidates three platform tiers (Essentials, Advanced, Signature) with unified Edwin AI, event intelligence, automation, and MCP access under one credit allocation. This addresses a critical adoption barrier: previous multi-meter pricing made agentic operations economically fragmented. Edwin AI now connects to 30+ tool categories spanning monitoring, ITSM, automation, and knowledge systems through a single governed MCP endpoint. The access model respects LogicMonitor entitlements, user permissions, and resource groups—directly responding to Gartner findings showing enterprise governance remains the bottleneck. For infrastructure teams, this means MCP transitions from an integration point-solution into a production-grade interoperability layer for ITOps agents. The centralized Agent Studio (launched October 14, 2026) gives IT administrators inventory, access control, and reliability management for production AI agents, making agents a first-class operational object alongside infrastructure, identities, and access controls.

Read the original at logicmonitor.com

Cisco MCP servers for Meraki and Catalyst Center reach GA

Telecom Reseller · Nov 16, 2026 · Industry news

What happened: Cisco VP Austin Lin discusses AgenticOps strategy: agents detect anomalies and investigate root causes while humans retain transparent diagnostic records and approval controls. Cisco MCP servers for Meraki, Catalyst Center, ISE, ThousandEyes, and IOS XE reach general availability November 2026.

Why it matters: Cisco's open MCP servers standardize agent access to network telemetry and configuration across vendors; MSPs already building on this layer means agents (Claude, OpenAI, etc.) can interact with infrastructure consistently without bespoke integrations.

In this Telecom Reseller/Cloud Communications Alliance podcast, Cisco's Austin Lin frames the NetOps transformation as a shift from removing engineers to removing alert triage from their workflow. AgenticOps is not aboutloss of control—it's about automating detection and investigation loops at software speed while preserving human decision authority. Cisco released MCP servers for Meraki, Catalyst Center, and IOS XE in August 2026; ThousandEyes, ISE, and NetBox servers are scheduled for general availability November 2026. These standardized interfaces allow agents built on any framework to query live network telemetry, configuration data, and test results without custom code. The interview highlights that managed service providers have already begun building on Cisco's MCP layer, integrating agents into managed service offerings. For NetOps practitioners, this represents a standards-based path forward: rather than waiting for vendor-specific agent integrations, teams can compose agentic workflows using MCP-compatible frameworks and clients. The emphasis on transparent diagnostics and approval controls reflects enterprise risk management; agents operate at scale but remain auditable and controllable.

Read the original at telecomreseller.com

Cato Networks adds runtime guardrails for AI agents

Cato Networks · Nov 17, 2026 · Vendor release

What happened: Cato Networks announced AgentSense, a patented runtime verification system that continuously assesses AI agents to stop unsafe actions before execution. The guardrail addresses the governance gap: Gartner reports 71% of orgs plan agentic AI deployment but only 21% have governance capabilities.

Why it matters: Runtime agent action verification fills a critical gap in agentic NetOps; agents can hold valid privileges yet execute unsafe tasks outside their authorized scope. Governance must move beyond access control to behavioral guardrails at the network level.

Cato's November 17 announcement reflects a maturing concern in enterprise agentic operations: privilege escalation is table stakes, but behavioral drift is the operational risk. AgentSense operates at the network enforcement point to validate agent decisions in real time—not just authenticate them, but verify they align with intent and policy before execution. The timing coincides with Gartner findings showing enterprises are deploying agents faster than governance frameworks can keep pace. For NetOps teams, this represents a new architectural layer: traditional RBAC and MCP permission scoping define what agents *can* access, but runtime guardrails define what agents *should* access given their current task. Cato frames this as a converged network and security problem; agents operate across infrastructure domains and require end-to-end visibility into human-agent interactions at the enforcement point. The patent status signals this is a defensible technical approach to the "ungoverned agent as new attack surface" problem articulated by Infoblox and others in the agentic NetOps market.

Read the original at finance.yahoo.com


Telco & Cable AI

Isolated telco AI agents make conflicting network decisions

RCR Wireless News · Nov 12, 2026 · Analysis

What happened: Telecom operators have deployed AI agents across network operations, customer care and planning, but isolated systems cause conflicting decisions—network operations agents reroute traffic while unaware customer care agents issue credits and book technicians. A shared data layer providing unified view of customers, services and networks is required.

Why it matters: NetOps and AIOps practitioners must address orchestration across isolated AI agents—shared data layers and unified event correlation are becoming architectural requirements for production automation at scale.

This practitioner perspective highlights an emerging operational maturity problem: operators have shipped AI agents into production in network operations, customer care, and capacity planning, but without coordination mechanisms. A concrete example illustrates the risk: in a metro fiber cut, the network operations agent reroutes traffic within seconds and reduces latency impact. Simultaneously, the customer care agent, unaware the issue has been mitigated, issues automatic service credits and books technician visits—incurring unnecessary cost and customer confusion. The architectural lesson: telcos maintain separate management systems for inventory, customer records, and planning tools, each representing network resources differently. AI agents see those representations as separate problems. The proposed solution is a shared data layer that reconciles those records and maps how network resources connect to services and customers—allowing agents to maintain domain specialization while operating on a unified view. The practitioner cites a Malaysian wireless operator that ingested events, metrics and topology from 20+ element management systems into a single assurance view, enabling agents to correlate network events to affected customers. For infrastructure teams, the implication is clear: modern telco automation requires treating data unification as a prerequisite to agent deployment, not a post-deployment integration task.

Read the original at rcrwireless.com


AI Industry & Policy

AMD closes $8.2 billion World Labs acquisition

AMD / Unite.AI · Nov 16, 2026 · Primary source

What happened: AMD completed its $8.2 billion all-stock acquisition of World Labs on November 16, 2026, with Fei-Fei Li joining as executive vice president and chief scientist reporting to CEO Lisa Su. World Labs released Atlas—a 3D world model for creating and reconstructing interactive environments with APIs for developers.

Why it matters: The acquisition integrates AI model researchers directly into AMD's chip design and software stack, shaping how enterprises architect compute infrastructure for spatial AI and robotics workloads.

AMD's completion of the World Labs acquisition represents a strategic shift in AI infrastructure development. The $8.2 billion all-stock deal—AMD's second-largest acquisition after the $50 billion Xilinx purchase—brings Fei-Fei Li's spatial AI research team directly into AMD's engineering organization. Li now leads AMD's scientific research as chief scientist, signaling executive commitment to aligning foundational model research with hardware optimization.

World Labs' Atlas product, released November 12, is production-ready: a 3D world model that generates interactive 3D environments with APIs for developer integration. For infrastructure teams, this matters because the World Labs team will continue advancing foundational model research while collaborating with AMD engineers to apply insights from emerging model capabilities to future AI infrastructure development. This tight coupling of model research to hardware design could influence how enterprises architect compute for spatial AI, robotics simulation, and 3D environment workloads.

The deal reflects broader industry consolidation in AI infrastructure—major chipmakers now bundle research talent directly into chip development to avoid optimization mismatches between models and silicon. For practitioners evaluating compute vendors, this signals that hardware roadmaps will increasingly be shaped by co-designed model research rather than reactive optimization.

Read the original at unite.ai


Read this edition on the web · The week in network intelligence · Vendor Radar

Digital Plumber is AI-curated and AI-summarized, with no human review before publishing. Verify before acting on anything here. How it works.

Don't miss what's next. Subscribe to Digital Plumber:
← Newer Nov 18: Microsoft 365 outage lasted 24 hours after gateway failover defect Older → Nov 16: Traversal raises $120M for autonomous AI site reliability engineers
Powered by Buttondown, the easiest way to start and grow your newsletter.