Nov 12: IETF publishes multipath QUIC as RFC 10069
Today's 3 things that matter
- IETF publishes multipath QUIC as RFC 10069
Multipath QUIC reduces latency and improves resilience for high-performance applications; network operators must plan path diversity and congestion control for endpoints using this RFC. - SonicWall SSL-VPN MFA bypass under active exploitation
SonicWall SSLVPN gateways are internet-facing remote-access devices. MFA bypass on production VPN infrastructure enables direct path to lateral movement. CISA deadline November 20, 2026 for federal agencies. - Cribl acquires Tero to cut low-value observability data
Consolidates observability and AI capabilities as operators track costs and monitor AI agents; Ben Johnson (Vector creator) and team now lead APM, federated observability, and AI-native roadmaps.
Full stories below, grouped by topic.
AIOps & Network Observability
Cisco bundles ThousandEyes monitoring into Wi-Fi 6E and 7 access points
Superpower Daily · Nov 11, 2026 · Industry news
What happened: Cisco embedded ThousandEyes network monitoring at no additional cost into eligible Wi-Fi 6E and Wi-Fi 7 access points with qualifying licenses, making network monitoring part of base deployments. The change positions network telemetry as input to Cisco's AgenticOps agents while unifying operations across Cisco's product portfolio.
Why it matters: Addresses long-standing difficulties with ThousandEyes licensing and deployment by embedding observability into base Wi-Fi licenses, reducing friction for NOC teams adopting end-to-end network visibility at scale.
Cisco embedded ThousandEyes network monitoring at no additional cost into eligible Wi-Fi 6E and Wi-Fi 7 access points with qualifying Networking Essentials or Advantage licenses, shifting network monitoring from a separate licensing and deployment burden to an included capability. According to Cisco's senior vice president of portfolio strategy, ThousandEyes had been difficult to license and deploy; embedding it makes monitoring part of qualifying deployments. Network telemetry from Wi-Fi infrastructure now flows into Cisco's AgenticOps agents as operational input, while Cloud Control unifies observability data and tools across Cisco's product portfolio for NOC and IT operations workflows. This change lowers barriers to unified network observability and positions real-time telemetry as a data source for autonomous network operations—a core shift in how Cisco frames AI-driven NOC automation.
Read the original at superpowerdaily.com
Data Center & AI Networking
Oracle ships collective communications library that reroutes around NIC failures
SDxCentral · Nov 11, 2026 · Industry news
What happened: Oracle released OCCL, a network plugin dedicated to distributed AI training preinstalled in supported Oracle HPC images. OCCL automatically reroutes traffic around link or NIC failures and supports single-planar and multi-planar RDMA networks.
Why it matters: Benchmarks on NVIDIA GB300 systems show OCCL increased average NCCL bus bandwidth by 2-10% for AllReduce operations, reducing collective communication latency across large distributed AI clusters.
Oracle announced new capabilities in Oracle Acceleron, OCI's suite of networking and infrastructure technologies for demanding AI workloads. OCCL sits at the transport layer between the collective-communication library (NCCL/RCCL) and the underlying network fabric, implementing congestion control and link/NIC failure recovery without requiring changes to higher-level AI frameworks. Internal benchmarks using NVIDIA GB300 systems with 16-128 nodes and four GPUs per node showed OCCL increased average NCCL bus bandwidth by 2-10% for AllReduce operations. This addresses a critical operational challenge: a one-second network delay across 100,000 GPUs could waste 28 hours of aggregate processing time. For operators running distributed training at scale, the plugin provides automatic failover and congestion intelligence without requiring fabric redesign.
Read the original at sdxcentral.com
Network Security
SonicWall SSL-VPN MFA bypass under active exploitation
CISA Known Exploited Vulnerabilities Catalog · Nov 6, 2026 · Primary source
What happened: CISA added CVE-2026-19868 (SonicWall SonicOS SSLVPN authentication bypass) to Known Exploited Vulnerabilities on November 6, 2026, based on evidence of active exploitation. Flaw allows attackers with valid alternate credentials to bypass SAML-enforced multi-factor authentication.
Why it matters: SonicWall SSLVPN gateways are internet-facing remote-access devices. MFA bypass on production VPN infrastructure enables direct path to lateral movement. CISA deadline November 20, 2026 for federal agencies.
CVE-2026-19868 is an authentication enforcement flaw exploitable only when SAML is enabled for SSL-VPN logins. CWE-308: Use of Single-factor Authentication. CVSS 7.1 High. Fixes in SonicOS 7.3.4-7013 and 8.2.3-8015. Affects Gen7 and Gen8 firewalls. No public IoCs or detailed attack analysis published yet. Network teams must inventory SonicWall Gen7/Gen8 SSL-VPN deployments, prioritize SAML-enabled configuration patching, and monitor authentication logs for non-SAML logins to SAML-protected portals.
OpenSSL patches DTLS and AES-GCM-SIV cryptographic flaws
OpenSSL Security Advisory / SUSE Security Updates · Nov 11, 2026 · Primary source
What happened: OpenSSL November 2026 advisory disclosed multiple cryptographic flaws including CVE-2026-88823 (AES-GCM-SIV Additional Authenticated Data handling), CVE-2026-84782 (DTLS retransmits from stale buffer, CVSS 8.2 High), and CVE-2026-35189 (excessive memory allocation in CRL processing, CVSS 5.3 Medium). Affects OpenSSL 3.5, 4.0, and 4.1 versions.
Why it matters: OpenSSL is foundational cryptographic library in network infrastructure, TLS implementations, VPNs, and appliances. CVE-2026-84782 rated High with potential for denial of service or information disclosure in DTLS-based protocols. SUSE issued high-severity livepatches November 11, 2026.
CVE-2026-88823: AES-GCM-SIV implementation treats zero-length AAD update as reset instead of no-op, causing previous AAD chunks to become unauthenticated (CWE-325). CVE-2026-84782: DTLS retransmission handler may use stale buffer for handshake messages, rated 8.2 High by SUSE. CVE-2026-35189: Certificate with many nameRelativeToCRLIssuer CRL distribution points causes disproportionate heap growth when OpenSSL caches X.509 extensions, rated 5.3 Medium. FIPS modules not affected; vulnerabilities in default provider only. Infrastructure teams must update OpenSSL to 3.5.10, 4.0.4, or 4.1.1+.
Read the original at openssl-library.org
Startups & Funding
Cribl acquires Tero to cut low-value observability data
Dealroom.co / Cloud Native Now · Nov 10, 2026 · Industry news
What happened: Telemetry software maker Cribl acquired Tero, a startup building an intelligence layer for observability data. Tero's technology identifies waste in observability data and pinpoints low-value signals across monitoring tools.
Why it matters: Consolidates observability and AI capabilities as operators track costs and monitor AI agents; Ben Johnson (Vector creator) and team now lead APM, federated observability, and AI-native roadmaps.
Cribl acquired Tero on November 10, 2026. Tero co-founders including CEO Ben Johnson—who previously created the open-source Vector observability pipeline acquired by Datadog in 2021—joined Cribl with their engineering team. This is Cribl's third acquisition in four months, following CardinalOps in July and Radiant Security's AI SOC technology in August. Tero's core capability of identifying waste and low-value data in observability datasets directly addresses a practitioner pain point as telemetry volumes explode with AI workloads. The team will lead roadmaps across host monitoring, application performance monitoring (APM), federated observability, and AI-native observability on a single open platform Cribl customers already use to collect, route, and optimize telemetry. The broader 2026 pattern shows consolidation around observability and AI, with Dynatrace, Datadog, Elastic, and Splunk also making AI-related moves.
Read the original at app.dealroom.co
Bay launches with $50M to police AI agents at runtime
SecurityWeek / The Next Web · Nov 10, 2026 · Industry news
What happened: Bay, founded February 2026 by three ex-IDF Unit 8200 officers, emerged from stealth with $50M ($37M Series A from Battery Ventures plus $13M seed). The platform discovers enterprise AI agents and enforces runtime policies on agent actions.
Why it matters: Automatically inventories AI agents, MCP servers, and local models; evaluates every proposed action in milliseconds to allow, block, redirect, or escalate based on policy—addressing visibility and control gaps as agents proliferate.
Bay emerged from stealth November 10 with $50M funding led by Battery Ventures ($37M Series A) plus prior seed from YL Ventures and Striker Venture Partners. Founded February 2026 by Yuval Zacharia (CEO), Boris Danilovich (VP R&D), and Aner Morag (CPO), the 35-person startup automatically discovers all AI agents, coding assistants, MCP servers, and local models in an enterprise, mapping risky tool combinations. It traces each agent session from first prompt to final action, then evaluates every proposed action in milliseconds to allow, block, refer to a user, or redirect based on policy. Bay learns from agent behavior patterns, eliminating the need to write rules for every new agent or tool. For infrastructure and security practitioners, this addresses a critical gap: as autonomous agents proliferate across enterprises, real-time visibility and policy enforcement at execution time—distinct from pre-deployment screening—becomes operationally essential.
Read the original at securityweek.com
People Moves
Meta hands infrastructure to former AWS executive Dave Brown
PYMNTS · Nov 12, 2026 · Industry news
What happened: Santosh Janardhan, Meta's head of infrastructure and co-head of engineering, will depart in early 2027. Dave Brown, a former AWS senior vice president with 19 years at Amazon, will assume the role of Chief Technology Officer of Meta Compute and oversee infrastructure operations.
Why it matters: Leadership change at critical juncture: Meta is betting hundreds of billions on AI infrastructure; Brown's AWS cloud and compute expertise directly shapes how Meta scales data center capacity and systems for AI workloads.
Meta announced on November 11 that Santosh Janardhan, who has led the company's infrastructure organization since 2009 and served as co-head of engineering, will step down from leading infrastructure in early 2027. His replacement is Dave Brown, a veteran AWS executive who joined Meta as vice president of infrastructure earlier in 2026. Brown spent 19 years at Amazon Web Services, focusing on compute, cloud, and AI infrastructure. In his new role, Brown will serve as Chief Technology Officer of Meta Compute, an initiative led by Meta President Dina Powell McCormick and technical leader Daniel Gross. The transition places a cloud infrastructure veteran in charge of one of the technology industry's largest AI infrastructure investments as Meta works to translate massive capital spending into computing capacity and new AI business opportunities. Janardhan will remain at Meta during the transition and move into an advisory role on Meta Compute. The leadership change comes as Meta accelerates its AI strategy and expands data centers and in-house chip development to support its AI ambitions.
Read the original at pymnts.com
Agentic AI & MCP
Half of organisations now run agentic AI in production NetOps
Harvard Business Review · Nov 11, 2026 · Analysis
What happened: Enterprise monitoring systems generate 4,000+ daily alerts; engineers resolve ~20/day, leaving most uninvestigated. Omdia/Cisco study finds 51% of organizations now use agentic AI in production NetOps, but visibility must precede autonomy authority.
Why it matters: Business cases focusing only on routine fixes miss lengthy investigation value; visibility and accountability are prerequisites for scaling agent autonomy in production networks.
Enterprise monitoring systems generate more than 4,000 alerts daily—approximately 2,000 concerning networks. Experienced engineers can investigate and resolve about 20 alert events per day. Clearing that queue manually would require roughly 100 IT specialists, far exceeding typical staffing, so nearly half of alerts close without investigation.
A 2026 Omdia and Cisco study of 1,000 IT and NetOps leaders finds more than half of organizations already use agentic AI to act on NetOps processes and tasks in production. Networks change constantly with feedback arriving within minutes; mistakes ripple visibly across security, applications, and customer experience. AI workloads bring new traffic patterns, increase data exchange, and create more dependencies across environments.
Business cases focused only on making routine fixes faster miss the value of shortening lengthy investigations. Visibility should be a condition for granting agents authority; agentic autonomy does not mean transferring accountability to the system. Operators must maintain oversight and validate that agent actions improved outcomes.
Isolated telecom AI agents make conflicting decisions during outages
RCR Wireless News · Nov 12, 2026 · Analysis
What happened: Telecom operators deploy isolated AI agents across network operations, customer care, and planning, creating conflicting decisions. In a fiber-cut scenario, network ops reroutes traffic while customer care, unaware, issues credits and books technicians because systems describe inventory differently.
Why it matters: Shared operational context enabling agents to see each other's work before acting is essential for credible cross-business autonomous operations; agent orchestration, not isolation, drives value.
Telecom operators have moved quickly to deploy AI agents in network operations, customer care, and capacity planning, with most performing well independently. However, isolated agent systems create operational conflict: in a metro fiber cut, the network operations agent reroutes traffic within seconds while the customer care agent, unaware of the fix, issues service credits and books technicians. Network inventory, customer records, and planning tools describe the same resources differently, preventing AI agents from recognizing they are working on the same problem.
The core issue is semantic fragmentation: without unified context, independent agents optimize locally while creating downstream chaos. The author advocates connecting network events to affected customers and enabling agents to see each other's work before acting. This shared understanding goes well beyond agents passing messages and represents what makes autonomous operations credible across the business. Multi-agent orchestration through unified data models or MCP-based coordination is necessary for telcos to achieve end-to-end autonomous operations.
Read the original at rcrwireless.com
Telco & Cable AI
Bell Canada quadruples Saskatchewan AI data centre to 1.2 GW
Construction Connect · Nov 11, 2026 · Industry news
What happened: Bell Canada expanded its Saskatchewan AI data center from 300 MW to 1.2 GW capacity in September 2026, with construction of the initial 300 MW phase ongoing and first data hall expected online in H1 2027. The expansion reflects demand acceleration and positions Bell as a sovereign AI infrastructure provider.
Why it matters: The expansion to 1.2 GW and $50 billion projected capital investment signals AI infrastructure demand exceeding initial forecasts and establishes Bell as a competitive sovereign AI compute provider, reducing dependency on US-based hyperscaler capacity for Canadian workloads.
Bell Canada's expansion from 300 MW to 1.2 GW within months signals AI infrastructure demand acceleration. Originally announced in March 2026 as a CAD $1.7 billion investment with CAD $1.3 billion spent in 2026, the September expansion justified quadrupling capacity based on customer commitments and competitive positioning. The facility serves both internal operational infrastructure for Bell's AI-powered network management and external sovereign AI compute for Canadian government, enterprise, and research customers. The entire initial 300 MW capacity is pre-sold; Cerebras provides wafer-scale computing for AI training and inference while CoreWeave delivers NVIDIA GPU-based capacity. Phased online schedule (first hall H1 2027, full operations end-2027) creates near-term production and working capital demands but positions Bell to monetize sovereign AI capacity as regulatory constraints on data residency tighten globally. Bell has also deployed AI-powered network operations using Google Cloud to improve fault detection and incident resolution, integrating data center infrastructure with network operations use cases.
Read the original at canada.constructconnect.com
Research, Standards & Industry
IETF publishes multipath QUIC as RFC 10069
IETF RFC Editor · Nov 7, 2026 · Primary source
What happened: IETF published RFC 10069 defining multipath support for QUIC, enabling connection migration and load distribution across multiple network paths. Specifies IP addressing and routing architecture for multipath QUIC endpoints.
Why it matters: Multipath QUIC reduces latency and improves resilience for high-performance applications; network operators must plan path diversity and congestion control for endpoints using this RFC.
RFC 10069, published November 7, 2026, formalizes multipath operation in QUIC (Quick UDP Internet Connections). The specification allows a single QUIC connection to span multiple network paths simultaneously, enabling seamless handover between paths, load balancing across interfaces, and recovery from path failures without connection disruption. Key aspects include path validation mechanisms, congestion control coordination across paths, and connection ID usage for path identification. This is particularly relevant for operators managing high-availability infrastructure and those deploying QUIC-based services where connection resilience is critical. The RFC draws distinctions between active multipath usage (simultaneous transmission) and migration scenarios (switching paths). Network teams should review implications for traffic engineering policies, especially regarding packet reordering and out-of-order delivery handling on multi-path flows.
Read the original at rfc-editor.org
OpenConfig moves tunnel and VLAN config to subinterface level
OpenConfig GitHub · Nov 5, 2026 · Primary source
What happened: OpenConfig released updates to tunnel interface and VLAN YANG models (versions 0.2.0 and 3.3.0) moving configuration to subinterface level and deprecating interface-level containers. Improves structural consistency and future model evolution.
Why it matters: Network automation engineers must migrate configurations from interface-level to subinterface-level tunnel and routed-VLAN containers before deprecated paths are removed; affects Ansible, gNMI, and YANG-driven provisioning workflows.
OpenConfig published updates on November 5, 2026 to openconfig-if-tunnel.yang (v0.2.0) and openconfig-vlan.yang (v3.3.0) via GitHub PR #1602. The changes relocate tunnel configuration from interface level to subinterface level for tunnel-type interfaces, and routed-vlan configuration to subinterface level for l3ipvlan types. Legacy interface-level containers are marked deprecated, signaling removal in future releases. Rationale centers on structural consistency: subinterfaces already support protocol stacking (IPv4, IPv6, MPLS), so tunnel and VLAN bindings logically belong there. This affects network automation teams using these models for device configuration. Migration path: dual-support via feature gates allows parallel naming conventions during transition. For operators, this requires updating YANG-based configuration generators, validating data model validators against new structure, and testing with devices that enforce new schema. The changes maintain backward compatibility during the deprecation window but require proactive migration of infrastructure-as-code.
Read the original at github.com
AI Industry & Policy
Anthropic opens IPO roadshow at $1.4 trillion valuation
Herve Cuviliez (Substack) · Nov 11, 2026 · Analysis
What happened: Anthropic opened IPO roadshow at $1.4T valuation with $26B private placement from tech suppliers (Google, Nvidia, AMD); OpenAI funding round now underway with SoftBank losing $7B on data-centre power bets; market starting to differentiate between AI suppliers, builders, and model companies.
Why it matters: Shows supplier consolidation risk: Nvidia and Google funding Anthropic while backing their own competing services signals potential conflicts and concentration in AI capital allocation that enterprises depend on.
Anthropic's IPO roadshow values the company at ~$1.4T, with a $26B private placement including investments from Google and Nvidia (~$10B each), AMD ($2.5B), Broadcom, Micron, Microsoft and TSMC's investment arm. SoftBank posted a $7B loss on data-centre power bets while still calling ASI its north star. The market has stopped buying "AI" as one basket: suppliers, builders and model companies are starting to trade apart. Suppliers are funding their customers—Nvidia and Google in Anthropic's IPO, Nvidia backstopping SoftBank's Ohio campus, Nvidia backing RLWRLD in robotics. This represents structural shifts in how capital flows through enterprise AI infrastructure and creates potential alignment issues when chip makers fund competing foundation model providers.
Read the original at hervecuviliez.substack.com
Australia creates AI safety regulator with criminal penalties under review
The Conversation · Nov 9, 2026 · Industry news
What happened: Australia's Prime Minister Anthony Albanese announced a new artificial intelligence safety regulator to enforce penalties for frontier AI companies that breach obligations. The government is also considering criminal law changes to deal with serious misconduct by frontier AI companies.
Why it matters: Enterprise AI governance frameworks now face regulator enforcement in major markets; compliance with safety and security controls, incident reporting, and model evaluation becomes legally mandatory, not discretionary.
The measures target frontier AI companies building the largest, most advanced models. Australia's proposed AI framework legislation will require maintaining security controls across all stages of AI development and deployment, identifying and managing serious risks, quickly providing mandatory incident notification, allowing independent evaluation of models, and maintaining published safety and security frameworks. The government is considering criminal law changes to address serious misconduct by frontier companies. This represents enforcement machinery moving beyond guidance: enterprise AI ops teams in Australia and jurisdictions following this model will face regulatory audit and incident notification obligations that mirror their data protection and cybersecurity compliance calendars.
Read the original at theconversation.com
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.