AI-curated intelligence for people who run networks.Daily coverage of AIOps, network automation, agentic operations, AI infrastructure, security and the vendors shaping them.
During routine FAA telecommunications maintenance on November 13, a newly installed Juniper router generated over 1,000 connection attempts every 4-5 seconds, cascading through the network and affecting 17 of 22 en-route air traffic control centers.
Why it matters The FAA is replacing 2,000 Cisco 600 routers with Juniper equipment while maintaining 24/7 operations, illustrating operational risks of brownfield router migrations and the need for staged canary deployments during equipment transitions.
Dreeze emerged from stealth on November 10, 2026, with $33M seed funding led by Norwest and Bessemer Venture Partners.
Why it matters Addresses a critical gap in cloud infrastructure resilience: traditional backup only restores data, not the configurations, secrets, and identity bindings needed to run applications—critical for SREs managing AI agent failures and cyberattacks.
AutoCon 6 (November 16–20, Tucson) brings network engineers for deep-dive technical talks on open-source tooling and infrastructure operations.
Why it matters AutoCon's practitioner-first model—avoiding vendor keynotes and prioritizing real operational lessons—is the primary venue for NetDevOps practitioners to exchange hands-on automation patterns across GitOps, source of truth, and agentic operations.
Network Automation Forum · Nov 15, 2026 · Industry news
AutoCon 6 (November 16–20, Tucson) brings network engineers for deep-dive technical talks on open-source tooling and infrastructure operations. Peer presentations examine federated Nautobot, replacing spreadsheet sources of truth, disposable labs, and resumable upgrades.
Why it matters AutoCon's practitioner-first model—avoiding vendor keynotes and prioritizing real operational lessons—is the primary venue for NetDevOps practitioners to exchange hands-on automation patterns across GitOps, source of truth, and agentic operations.
AutoCon 6 runs November 16–20 in Tucson with two days of practical network automation training (November 16–17), followed by conference sessions covering AI in network automation, intent and configuration, compliance, and automation ROI. The agenda reflects lessons from AutoCon 5 (June 2026 Munich), which featured production talks on GitOps delivery patterns, ops-led automation, fully automated backbone networks with 10,000+ devices, and LLM landscape navigation. Presentations cover federated Nautobot deployments across segmented networks, migration from spreadsheet-based sources of truth, disposable network labs, mock-device testing, resumable network upgrades, and agentic tools. For practitioners, this is the working-engineer venue grounding out production constraints: how spreadsheets become SoT liabilities, why upgrade automation must be resumable, what agentic tools operationally solve.
AI clusters require dramatically more optical interconnect per unit of compute than previous data center generations, pulling transceiver speeds from 800G toward 1.6T and driving early adoption of co-packaged optics. Moving from 800G to 1.6T preserves total fiber capacity and halves the module count; at an assumed 50% price premium, module spending for the same capacity falls 25%.
Why it matters Operators can reduce fiber run counts and total optics spend per data center while increasing port density, improving rack power budgets and reducing cooling load—the physics of AI density now favor 1.6T transceiver adoption.
AI clusters pull transceiver speeds from 800G toward 1.6T and drive early adoption of co-packaged optics because they require dramatically more optical interconnect per unit of compute than previous data center generations. In 2026, 800G is now the default for new AI cluster buildouts, while 1.6T is entering volume production enabled by 200G-per-lane lasers. Moving from 800G to 1.6T preserves total fiber capacity by halving the module count—fiber installed for 800G parallel optics can be reused—and at an assumed 50% price premium for 1.6T modules, module spending for the same 25.6 Tb/s capacity falls 25%. The 800G configuration uses 32 lanes at standard width, while the 1.6T configuration uses 16 lanes each twice as wide; traffic per lane doubles while total traffic stays at 25.6 Tb/s; transceiver modules across two endpoints fall from 64 to 32. This creates significant capital and power leverage for operators scaling hyperscale AI data centers: fewer transceivers, shorter fiber routes, reduced thermal dissipation per bit forwarded, and improved space utilization in OSFP cages.
Oracle Cloud Infrastructure introduced Oracle Collective Communications Library (OCCL) network plugin for distributed AI workloads on OCI-supported GPUs and RDMA-enabled infrastructure. Benchmarks on NVIDIA GB300 systems show OCCL increased average NCCL bus bandwidth by up to 10% for AllReduce operations across 16-128 nodes.
Why it matters OCCL addresses production constraints in large-scale AI training: congested network paths, failed links, and failed NICs interrupt collective communication and force training jobs to stall or restart—this plugin recovers automatically and reroutes around failures.
Large-scale AI training depends on collective communication among hundreds or thousands of GPUs; as clusters grow, a congested network path, failed link, or failed NIC can interrupt collective communication, lower GPU utilization, and force a training job to stall or restart. Oracle's OCCL network plugin is designed for distributed AI workloads and automatically reroutes traffic around supported link or NIC failures and restores connectivity when hardware recovers; it supports single-planar and multiplanar RDMA networks in a vendor-agnostic fashion. On NVIDIA GB300 systems with 16 to 128 nodes and four GPUs per node, OCCL increased average NCCL bus bandwidth by up to 10% for AllReduce operations. The design creates path diversity to isolate failures; networking intelligence for congestion control and link/NIC failure recovery sits in the collective-communication transport layer; AI frameworks continue to use NCCL or RCCL with the collective-communication library using its standard network-plugin interface for inter-node data transfers. This is a practical response to the constraint that large-scale clusters are failure-prone, with the plugin reducing both mean time to recovery and the performance tax of link flapping during training.
Tom's Hardware, Transportation.gov, Cellstream · Nov 13, 2026 · Industry news
During routine FAA telecommunications maintenance on November 13, a newly installed Juniper router generated over 1,000 connection attempts every 4-5 seconds, cascading through the network and affecting 17 of 22 en-route air traffic control centers. A specific router feature caused the overload; FAA staff disabled it globally across all Juniper routers. The ground stop lasted 4 hours 21 minutes and affected over 3,500 flights.
Why it matters The FAA is replacing 2,000 Cisco 600 routers with Juniper equipment while maintaining 24/7 operations, illustrating operational risks of brownfield router migrations and the need for staged canary deployments during equipment transitions.
On November 13, 2026 at 20:54 UTC (3:54 p.m. EST), the FAA issued a nationwide ground stop after a router at a remote facility began generating extraordinary volumes of connection attempts. Transportation Secretary Sean Duffy reported over 1,000 connection attempts every 4-5 seconds directed at other routers in the FAA telecommunications network. The traffic propagated through the decades-old copper-based infrastructure and caused instability affecting both air-ground communications and radar feeds. Eventually 17 of the FAA's 22 en-route air traffic control centers were impacted. Investigation revealed a specific feature on newly installed Juniper routers (built by HP Enterprise) was causing the system overload. FAA employees disabled that feature across the entire Juniper fleet. The FAA is continuously replacing 2,000 aging Cisco 600 routers with newer Juniper equipment while keeping the existing network operational—a complex migration that exposed the brittleness of legacy infrastructure when new gear behaves differently.
System Design Academy, Enow Software, Azure Blog UK · Nov 14, 2026 · Analysis
Technical analysis reveals Microsoft 365 outage was amplified by cascading routing failures. Initial health-check failures triggered traffic rebalancing, which itself generated a surge of outbound routing calls that further overloaded backend servers. Servers returned to service became unresponsive under renewed traffic, creating a self-amplifying deadlock requiring staged recovery.
Why it matters Large-scale cloud operators should recognize this pattern: remediation actions (traffic rebalancing, failover) can trigger secondary issues through control-plane overload and retry storms. Circuit breakers, traffic throttling, and gradual staged recovery are essential guardrails.
Analysis of the Microsoft 365 outage on November 12-13 shows a cascade pattern typical of large-scale infrastructure failures. The incident began as an Exchange Online connectivity problem under incident ID EX1550543. Early diagnosis attributed the issue to 'elevated load,' prompting traffic rebalancing across global regions. By late morning, engineers realized the load was not from genuine user traffic but from sudden collapse in backend server capacity, which dumped traffic onto surviving nodes. Automated health-check failures triggered the cascade by pulling servers from service. Remediation actions intended to relieve pressure instead amplified the problem: continuous traffic rebalancing caused a massive surge in outbound routing calls from Microsoft's routing services to backend servers. When those calls could not route properly, servers returned to service became unresponsive again, requiring Microsoft to apply aggressive traffic throttling, protocol adjustments, and gradual staged recovery. The outage lasted 30 hours, affecting 430+ million users across 70+ countries, illustrating how routing and authentication components interact under overload.
CISA added six actively exploited vulnerabilities to the KEV Catalog on November 10, including two Barracuda flaws (CVE-2026-116606, CVE-2026-111560) that allow root access when chained, plus Cisco BroadWorks XXE, Openwave, Microsoft Outlook, and Google Chromium vulnerabilities. Two of the Barracuda flaws hand an attacker root on Barracuda Email Security Gateway appliances.
Why it matters The two Barracuda flaws chain to enable unauthenticated root access: CVE-2026-116606 is a pre-authentication authorization bypass allowing crafted backup restoration, and CVE-2026-111560 is command injection that runs as root; together they bypass all authorization controls. Affects Email Security Gateway and five other Barracuda products; patch immediately with federal deadline November 13.
CISA announced six new Known Exploited Vulnerabilities (KEV) entries on November 10, 2026, based on evidence of active exploitation. CVE-2026-116606 is a pre-authentication authorization bypass allowing unauthenticated attackers to restore crafted configuration backups and change appliance settings, while CVE-2026-111560 is a command injection that normally requires admin access but runs as root; when chained, they permit root access without any login. The bypass carries a CVSS score of 9.3 and the command injection 8.6. CVE-2026-116606 affects only the Barracuda Email Security Gateway; CVE-2026-111560 also affects Email Security Gateway and five other Barracuda products: the Load Balancer Application Delivery Controller, Message Archiver, WAF Control Center, Web Application Firewall, and Web Security Gateway. Barracuda disclosed both flaws on November 2, 2026, and CISA added them to KEV on November 10 with a federal deadline of November 13. The Email Security Gateway is patched in version 9.4.0.030 and later.
A double-free and use-after-free vulnerability in the FRRouting BGP daemon (CVE-2026-120221, CVSS 8.2) allows unauthenticated remote attackers to cause a denial of service by crashing the bgpd process. Additionally, an out-of-bounds heap read in FRRouting bgpd (CVE-2026-129217) causes DoS when processing BGP NOTIFICATION messages with improperly validated data fields.
Why it matters These vulnerabilities affect FRRouting 8.3 through 10.5.1 and directly impact network routing infrastructure, particularly in ISPs, large enterprises, and service providers relying on open-source BGP implementations. Unauthenticated DoS attacks against BGP daemons can disrupt critical routing convergence and cause widespread outages; patches should be prioritized for edge and core routing deployments.
CVE-2026-120221 is a double-free and use-after-free vulnerability in the FRRouting BGP daemon that allows unauthenticated remote attackers to trigger a denial of service by crashing the bgpd process; it carries a CVSS score of 8.2. A separate vulnerability (CVE-2026-129217) involves an out-of-bounds heap read in FRRouting bgpd that causes DoS via BGP NOTIFICATION messages; in versions 8.3 through 10.5.1, the bgp_notify_decapsulate_hard_reset() function in bgpd/bgp_packet.c does not validate the length of the data field in received BGP NOTIFICATION Cease/Hard Reset messages per RFC 8538. FRRouting is widely deployed as the open-source BGP implementation in Linux-based routers and route servers used across ISP networks, hyperscaler edge infrastructure, and large enterprise core routing. Unauthenticated remote DoS against BGP daemons can destabilize routing convergence and trigger cascading outages, making these high-priority patches for network operators managing critical routing infrastructure.
Help Net Security / SecurityWeek · Nov 10, 2026 · Industry news
Dreeze emerged from stealth on November 10, 2026, with $33M seed funding led by Norwest and Bessemer Venture Partners. The platform automates recovery of entire cloud applications—not just data—by mapping dependencies, orchestrating infrastructure/configuration/identity restoration, and validating recovery before outages.
Why it matters Addresses a critical gap in cloud infrastructure resilience: traditional backup only restores data, not the configurations, secrets, and identity bindings needed to run applications—critical for SREs managing AI agent failures and cyberattacks.
Dreeze, founded in 2025 by Zerto and Kashya veterans (Ziv Kedem, Oded Kedem, Avi Raichel), emerged publicly with $33M seed to solve the 'recovery scramble'—the gap between data restoration and full application operability. Application Revival platform continuously scans cloud environments to map dependencies, back up all resources (infrastructure, configurations, identities, secrets, keys, service connections), and generate recovery orchestration plans stored in an isolated vault. Four core capabilities: Know (autonomous dependency mapping), Orchestrate (revival automation restoring data, infrastructure, configs, identities in dependency order), Validate (end-to-end testing without production impact), and Remediate. AWS support is generally available; Azure planned for early 2027. For infrastructure and SRE teams, this addresses a real operational blind spot: when AI agents or cyberattacks corrupt configurations, secrets, or IAM bindings, RTO extends from hours to days or weeks even with intact backups. Application Revival's dependency orchestration reduces recovery time from manual trial-and-error to automated, tested workflows.
Netceed appointed Eric Updyke as group CEO, effective November 10, 2026, succeeding Alper Turken. Updyke most recently served as CEO of Spirent Communications from 2019 to 2025, where he led the company's shift from hardware-centric vendor to software, automation and services provider.
Why it matters Updyke joins as Netceed pursues growth across broadband, AI infrastructure and energy markets following the company's 2026 financial restructuring—relevant for infrastructure teams managing carrier and data center vendor relationships.
Netceed is a global leader in delivering supply chain solutions across broadband, data center and energy infrastructure. Updyke brings more than 35 years of leadership experience in communications and technology infrastructure. Previously, Updyke held senior roles at Amdocs, including group president of Amdocs Services, a global organization with more than 10,000 employees. Turken's tenure included setting a strategy centered on infrastructure supply chain services, building an AI infrastructure business and overseeing the company's 2026 financial restructuring. The leadership transition places Updyke in charge of executing that strategy following the restructuring. For network and infrastructure practitioners, this signals continuity in Netceed's infrastructure supply chain strategy with leadership experienced in both legacy telecom transitions and modern software-driven architecture—relevant for teams managing fiber, data center, and broadband infrastructure procurement and logistics.
AT&T has deployed nearly 500 autonomous AI workflows across core operations, handling over 1 million transactions monthly and processing 45 billion AI tokens daily, with plans to triple production workflow count within a year. The Business Agent Foundry maps processes to eliminate steps, handoffs, and legacy dependencies rather than automating isolated tasks.
Why it matters Value comes from connecting agents to business rules, data, and legacy systems, not deploying independent assistants. AT&T requires security, privacy, and compliance review for customer-facing workflows and maintains human oversight until accuracy reaches mid-to-high 90%, providing a practical governance model for operators deploying agents at scale.
AT&T's Business Agent Foundry, established in early 2025, works with individual business units to map business processes using industry frameworks and eliminate unnecessary steps and organizational handoffs. The company maintains a relatively small set of agents executing business workflows using centrally managed business logic. Employees can create personal productivity agents independently, but workflows affecting customers must pass foundry review and undergo security, privacy, and compliance checks. Use cases include customer provisioning, network operations, sales prospecting, knowledge retrieval, and planning actions across legacy systems. Notably, AT&T halved the time required to provision dedicated internet connections. The approach requires one measurable high-volume workflow per team, mapping every handoff and exception, with explicit thresholds for when autonomy is allowed. Production AI investments return 'fivefold within a year,' according to AT&T's Luke Corbin.
Network Insight published a comprehensive technical analysis of Model Context Protocol architecture covering host–client–server topology, JSON-RPC communication, tools, resources, prompts, authentication, networking stages of remote MCP requests, and enterprise MCP security and observability architectures for AI agent deployments.
Why it matters Remote MCP deployments introduce standard enterprise networking requirements: DNS, HTTPS, TLS, proxies, firewalls, and load balancing all directly affect MCP availability. Network teams must now account for agent-to-server communication patterns in existing infrastructure, and MCP request failures can occur even when AI applications function correctly due to underlying network or TLS issues.
MCP (Model Context Protocol) provides standardized interfaces giving AI agents governed access to infrastructure tools, network devices, cloud platforms, observability stacks, and ITSM systems. The article details how MCP integrates with enterprise networking: clients must reach server endpoints through DNS resolution, routing, firewalls, proxies, and encrypted HTTP connections, while servers may need controlled access to internal systems. Streamable HTTP operates over standard HTTP networking, making connectivity, TLS negotiation, authentication, load balancing, and proxy behavior critical to MCP availability. Network Insight's analysis covers MCP connectivity dependencies on enterprise networks, enterprise security architecture for MCP, and production reference architectures. As of September 2026, 78 production-ready MCP servers span every major layer of the network and infrastructure stack, requiring network ops teams to extend observability and access control.
OpenAI, Anthropic, Microsoft and Mistral AI signed the Paris Call for AI Safety and Security on November 9, 2026, pledging to build mechanisms to stop, pause or shut down their models during training and internal use. OpenAI described Control Room, an internal service that stops research jobs on models in 35 seconds.
Why it matters The pledge has no binding force or deadline; the commitment is voluntary and covers only frontier models, signaling regulatory pressure without hard enforcement mechanisms.
The Paris Call for AI Safety and Security is a political declaration signed on Monday, November 9, 2026 by seventeen states, ten AI companies including OpenAI, Anthropic, Microsoft and Mistral AI, and ten civil society organizations. The developers of the most powerful models promise to build mechanisms to stop, pause or shut down the systems they control, including during training and internal use.
On November 9, OpenAI described Control Room, an internal service that stops research jobs on models in 35 seconds. However, the pledge itself is non-binding and sets no deadline for implementation. Between November 10 and 13 the list of signatures nearly tripled, highlighting how easy it is to put a name under a promise with no deadline. The commitment covers only frontier models and their risks, representing a narrower scope than comprehensive AI governance frameworks.
For practitioners, this signals growing regulatory pressure on frontier model developers but with limited immediate enforcement mechanisms. The Paris Call represents soft power rather than hard commitments—watch for subsequent EU or other jurisdictional requirements that may mandate binding shutdown capabilities with specified timelines.
Nvidia CEO Jensen Huang resigned on November 12, and three days later Anthropic filed an amended S-1 that could value the company at up to $1.4 trillion. Meta's acquisition of Conductor and Sequoia Capital's decision to close its London office signal capital at the top of the AI stack is being restructured, not just raised.
Why it matters Leadership crises at foundational infrastructure providers and record frontier-AI valuations alongside VC retrenchment reshape capital allocation patterns and governance risk assessment for enterprise AI deployment decisions.
Nvidia CEO Jensen Huang resigned on November 12 amid allegations involving export-control fallout and governance questions affecting financing arrangements for CoreWeave and Nebius. The timing exposes geopolitical and governance risks embedded in AI infrastructure supply chains. Simultaneously, Anthropic filed an amended S-1 that could value the company at up to $1.4 trillion, representing a historic capital event for a frontier-AI lab. Meta's acquisition of Conductor and Sequoia Capital's decision to close its London office and wind down two programs, combined with a fresh $10 billion fund raise, all reflect capital at the top of the AI stack being restructured. For infrastructure teams, this signals market consolidation around core players and potential shifts in support ecosystems for alternative compute platforms and specialized inference tools.
Red Hat OpenShift 5 integrates AI agents to automate Kubernetes cluster upgrades and troubleshooting at KubeCon NA 2026, reducing planning from weeks to a single click. Agents pull telemetry, logs, and Kubernetes metadata to investigate problems and suggest fixes with human operator approval remaining mandatory.
Today's 3 things that matter and every story with why it matters, in your inbox each morning. Free, and you can unsubscribe at any time. Prefer a reader? Follow the RSS feed.