AI-curated intelligence for people who run networks.Daily coverage of AIOps, network automation, agentic operations, AI infrastructure, security and the vendors shaping them.
Traversal Inc. raised $120 million in fresh funding, bringing its valuation to $1 billion, as its AI agents automate site reliability engineering tasks.
Why it matters The autonomous SRE platform directly addresses operational burden SRE and NetDevOps teams face in production-incident management and infrastructure problem remediation.
Cisco released security updates for CVE-2026-76418, a high-severity zero-day affecting the Remote Access SSL VPN service in Cisco Secure Firewall ASA and FTD software; Cisco disclosed the vulnerability on October 28 after its PSIRT became aware of active exploitation during October, and CISA added it to KEV on October 29.
Why it matters CVE-2026-76418 is an actively exploited, unauthenticated DoS flaw (CVSS 8.6) caused by logic errors when secondary authentication is configured on tunnel groups; network teams must patch immediately or implement mitigation to prevent attackers from reloading firewalls and taking VPN infrastructure offline.
Verizon and Ericsson announced nationwide rollout of the Ericsson Intelligent Controller supporting AI-native rApps deployment.
Why it matters Enables RAN automation at scale with open, programmable interfaces; reduces vendor lock-in while delivering quantified performance gains for closed-loop optimization.
Datadog named a Leader in The Forrester Wave™: AIOps Platforms, Q4 2026, with highest scores in Innovation, Roadmap, and ModelOps/AI Governance. Platform delivers autonomous operations via Bits AI detecting, investigating, and remediating within guardrails.
Why it matters Native AI observability unifies application monitoring, agent observability, and digital experience in one correlation flow—when integrated with network tools, eliminates tool-switching during incident investigation and reduces MTTR.
Datadog and Dynatrace emerged as the two leaders in a 10-vendor Forrester evaluation. Datadog's unified platform spans observability, security, and AI workloads with Bits AI providing cross-silo event correlation and autonomous remediation. Native AI observability ingests application traces, agent logs, and user experience metrics in one flow, enabling teams to trace a single customer request from service activity through agent reasoning to business impact. For network operations, platform consolidation matters: organizations can now correlate network telemetry with application performance and security events without manual context assembly, accelerating root cause analysis and reducing mean time to resolution.
ThousandEyes deprecated legacy OpenTelemetry attribute (thousandeyes.source.agent.connection.type) effective October 29, 2026, standardizing on network.connection.type. Splunk Synthetic Monitoring integration reached GA, enabling test creation and application-network correlation within Splunk dashboards.
Why it matters OpenTelemetry semantic alignment reduces instrumentation drift; Splunk integration closes the NOC pain point of correlating network path metrics with application performance traces without manual context-switching during incident response.
ThousandEyes' October release reflects two operational priorities: semantic standardization and cross-domain APM correlation. Legacy attribute deprecation forces teams to update dashboards and alert rules to use vendor-neutral semantics, accelerating adoption of OpenTelemetry conventions across infrastructure. Splunk integration GA enables network and application teams to view ThousandEyes synthetic tests and path analysis alongside Splunk application performance metrics in a single dashboard, with drill-through capability to detailed network analysis. For NOC teams, this means operators can correlate network latency, loss, and path changes with application errors and slowdowns in one observability interface, reducing the time spent switching between network and APM tools during incident investigation.
Arista CEO Jayshree Ullal forecasted the company's AI fabrics business doubling to approximately $7 billion in 2027, with starting point at about $3.5 billion in 2026. Arista raised its full-year 2026 guidance to $13.3 billion, representing 48% growth, signaling market dominance in Ethernet-based AI fabric architectures.
Why it matters Demonstrates Arista's emerging leadership in purpose-built AI networking fabrics; the company's AI-focused strategy and deep relationships with hyperscalers position it as the primary Ethernet alternative to NVIDIA Spectrum-X and proprietary InfiniBand, with tangible revenue metrics confirming market share capture.
Arista reported third-quarter results and provided first look at 2027 guidance after reporting record AI-driven results. CFO Chantelle Breithaupt set the 2026 AI fabrics starting point at about $3.5 billion, with CEO Ullal expecting doubling to approximately $7 billion in 2027. Arista raised its 2026 guidance for the fourth time, to $13.3 billion, or 48% growth. For the fourth quarter, Arista guided to revenue of $3.9 billion-$4.0 billion, a 48%-49% non-GAAP operating margin and EPS of $1.26-$1.30. Both Arista and Cisco reported record AI-driven results in August, yet their stocks sold off hard, as one company offers margin-rich focus while the other controls its own silicon, with scale-across design wins potentially defining data center networking for the next decade. Arista's non-AI, non-campus revenue implies only about 11% growth in 2027, while total growth decelerates from 48% to 38%-40%.
Internet Society Pulse · Nov 3, 2026 · Primary source
Study measuring latency impact on networks when individual IXP paths are withdrawn. Among 30 largest IXPs, 6.8–86.4% of members experience measurable performance degradation, with 64.2% of affected ASes losing >10 ms latency and some exceeding 50 ms increases.
Why it matters Shows IXP importance varies by path quality, not size alone; helps operators size redundancy requirements and understand their real exposure to IXP failures.
Internet Society Pulse published detailed analysis of anycast network resilience when specific IXP paths are withdrawn via BGP community manipulation. The study found that while most anycast networks retain connectivity when losing a single IXP path, latency impact is substantial and highly variable across exchanges. Among the 30 largest IXPs globally, the fraction of member networks experiencing measurable latency degradation ranged from just 6.8% to 86.4%, indicating that IXP criticality depends more on the quality and diversity of alternative paths available to members than on the size of the exchange itself. Median latency increases across affected networks were severe: 64.2% experienced >10 ms loss, 29% >20 ms, and 0.9% >50 ms (with one outlier reaching 71.2 ms). The impact is geographically uneven, with operators in lower-GDP regions experiencing disproportionate vulnerability. This challenges the assumption that larger IXPs provide inherently better resilience, suggesting instead that network topology and path diversity are the critical factors for operational planning.
Cisco released security updates for CVE-2026-76418, a high-severity zero-day affecting the Remote Access SSL VPN service in Cisco Secure Firewall ASA and FTD software; Cisco disclosed the vulnerability on October 28 after its PSIRT became aware of active exploitation during October, and CISA added it to KEV on October 29.
Why it matters CVE-2026-76418 is an actively exploited, unauthenticated DoS flaw (CVSS 8.6) caused by logic errors when secondary authentication is configured on tunnel groups; network teams must patch immediately or implement mitigation to prevent attackers from reloading firewalls and taking VPN infrastructure offline.
The vulnerability exists in the Remote Access SSL VPN functionality due to logic errors when secondary authentication is configured on a tunnel group. An attacker could exploit this by sending a crafted request when connecting to the SSL remote access VPN; a successful exploit causes the affected device to reload, resulting in a DoS condition. Cisco says its security team became aware of active exploitation in October 2026 but has not named an attacker, identified targets, or said how frequently the flaw has been used. This marks the second Cisco firewall VPN zero-day since August; Cisco's earlier SSL VPN denial-of-service advisory was published on August 11, 2026 and revised on September 16. Cisco offers a mitigation (disabling the tunnel group list or secondary authentication) that reduces VPN security; upgrading is the recommended fix.
CISA Known Exploited Vulnerabilities Catalog · Oct 28, 2026 · Primary source
CISA added two new vulnerabilities to its Known Exploited Vulnerabilities Catalog on October 28, 2026, based on evidence of active exploitation: CVE-2026-41709 (Broadcom VMware ESX Insufficient Logging) and CVE-2026-46300 (Linux Kernel Missing Initialization of a Variable).
Why it matters Both CVEs are in CISA KEV as actively exploited; infrastructure teams running VMware hypervisors and Linux systems must prioritize patching to prevent adversary activity from going undetected and to prevent memory corruption exploitation across production environments.
CISA's Known Exploited Vulnerabilities catalog tracks flaws confirmed to be exploited in real-world attacks. BOD 26-04: Prioritizing Security Updates Based on Risk establishes vulnerability management requirements for Federal Civilian Executive Branch agencies. Both vulnerabilities now carry federal remediation deadlines. VMware ESX insufficient logging could allow attackers to cover tracks on hypervisor platforms controlling multiple guest VMs, reducing observability and incident response capability during active compromise. The Linux kernel memory initialization bug poses a more direct operational risk across bare-metal and containerized Linux deployments, potentially enabling privilege escalation or denial of service. Federal agencies and critical infrastructure operators must treat these as immediate priorities given active exploitation.
Traversal Inc. raised $120 million in fresh funding, bringing its valuation to $1 billion, as its AI agents automate site reliability engineering tasks. Monthly investigations grew over 100-fold during 2026 from 513 to nearly 73,000, with usage extending into alert triage, release risk, CMDB validation, and security.
Why it matters The autonomous SRE platform directly addresses operational burden SRE and NetDevOps teams face in production-incident management and infrastructure problem remediation.
Traversal, a NYC-based AI SRE platform provider, raised $120M at $1B valuation in a round co-led by Kleiner Perkins, Redpoint Ventures, and Khosla Ventures with participation from Sequoia Capital, Battery Ventures, and Amex Ventures. Total funding now stands at $173 million following the $48 million raised when emerging from stealth in June 2025. The company demonstrates significant operational traction: monthly investigations grew over 100-fold during 2026, with ninefold year-to-date growth in annual recurring revenue, more than 150,000 customer investigations in the most recent quarter and support for over 4,000 engineers. Agarwal's background in causal machine learning underpins the platform's approach, with the company striving to build a superintelligent AI SRE to help enterprises keep pace with rapid velocity of AI-enabled developer teams. Funds will extend AI across the software development lifecycle by making production systems more resilient, and support hiring across research, engineering, and go-to-market.
CrowdStrike disclosed plans to acquire Fiddler AI for $400M, a Palo Alto startup whose platform tests AI agents for risks before launch, spotting issues like malicious code and training data leaks. The company also acquired Dreadnode for $95M, an automated security testing platform for AI models and MCP servers.
Why it matters The acquisitions extend CrowdStrike's Falcon platform across AI deployment stages with pre-deployment testing and runtime monitoring capabilities—critical for infrastructure teams managing AI agent risk in production.
CrowdStrike will pay approximately $400 million to acquire Fiddler AI, founded in 2018 and headquartered in Palo Alto. Fiddler AI develops software for monitoring and explaining agent behavior. Dreadnode takes a complementary approach by finding vulnerabilities in the MCP servers that AI agents use to carry out their work, mapping out AI agents in a company's environment and ensuring they meet cybersecurity standards. Dreadnode offers an AI Kill Switch enabling administrators to manually shut down risky agents and prevent relaunching. CrowdStrike plans to connect Dreadnode's testing results with Falcon Guardian, its runtime AI security offering, so findings can inform protection policies. Dreadnode's automated AI security testing is expected to become generally available between now and February. The transactions are primarily cash-funded with some restricted stock included.
Barret Zoph, former OpenAI VP of Research, Post-Training, joined Google DeepMind as VP of Research. He brings nine years of AI research leadership spanning Google Brain and OpenAI roles.
Why it matters VP-level AI research appointments at frontier labs signal competitive talent consolidation in large-scale model development and infrastructure planning.
Barret Zoph joined Google DeepMind as VP of Research, effective November 2026. Zoph previously served as VP of Research, Post-Training at OpenAI from September 2022 to October 2024, where he led post-training research for frontier models. Before that, he worked as a research scientist at Google Brain from June 2016 to August 2022, contributing to AutoML and neural architecture search. His appointment to lead DeepMind's research function reflects continued executive consolidation among frontier AI labs competing for top talent in model scaling, training efficiency, and post-training methodologies. The move comes as both OpenAI and Google DeepMind expand infrastructure-heavy AI research programs requiring deep technical leadership.
Anthropic announced expansion of Claude Marketplace on November 4, 2026 to include 25+ partner agents targeting security operations, developer infrastructure, and productivity workflows.
Why it matters Pre-built, vetted agents reduce build-versus-buy friction for operations teams deploying autonomous agents; ecosystem maturity signals shift from experimental to production-ready infrastructure.
The marketplace expansion curates partner agents for security operations (threat detection, incident response), developer infrastructure (CI/CD, observability), and operations domains. For NetOps and SRE teams evaluating agentic workflows, pre-built agents eliminate custom implementation work and provide known governance models, audit trails, and vendor support. The timing follows Anthropic's November 12 policy update restricting agent use for surveillance and biometric profiling, suggesting marketplace serves as gating layer for operational agents. Teams no longer need to build agents on raw Claude API; they can deploy vetted partner agents with policy compliance baked in. This accelerates enterprise adoption by reducing complexity, liability, and time-to-value for autonomous operations.
OpenAI's November 2 documentation update enables custom MCP server connections for ChatGPT Business and Enterprise/Edu plans, allowing workspace admins to provision infrastructure MCP servers directly into ChatGPT without code.
Why it matters MCP support across Claude and ChatGPT means operations teams can standardize infrastructure agent tooling (NetBox, Meraki, Splunk, etc.) and choose LLM clients based on regulatory, cost, or performance constraints rather than MCP availability.
OpenAI's documentation now describes full MCP support for ChatGPT Business and Enterprise on the web, completing MCP adoption across major LLM platforms. Workspace admins can add custom MCP server URLs via configuration; agents automatically discover tools and resources. This matches Anthropic's Claude offering and completes the platform-agnostic infrastructure layer for AI agents: NetBox MCP, Meraki MCP, ThousandEyes MCP, Splunk MCP, and others now work with multiple client implementations. For operations teams, this removes lock-in risk and enables heterogeneous deployments (some agents on Claude, others on ChatGPT) based on workload requirements. The shift signals MCP maturation from experimental to production infrastructure—formal governance via AAIF/Linux Foundation, 10,000+ servers in production, July 2026 stateless specification finalized, and now dual-platform support from Anthropic and OpenAI.
Verizon and Ericsson announced nationwide rollout of the Ericsson Intelligent Controller supporting AI-native rApps deployment. Field trials showed up to 8.9% improvement in downlink throughput and 8% in uplink throughput.
Why it matters Enables RAN automation at scale with open, programmable interfaces; reduces vendor lock-in while delivering quantified performance gains for closed-loop optimization.
Verizon and Ericsson announced nationwide deployment of the Ericsson Intelligent Controller (EIC), an open platform enabling AI-native rApps across RAN infrastructure. Field results from autonomous optimization rApps including Advanced Antenna System Cell shaper demonstrated 8.9% downlink and 8% uplink throughput improvements. The deployment showcases an open architecture allowing Verizon to develop self-designed rApps, positioning the operator toward Level 4 cognitive automation. Verizon plans national rollout this year. This matters for NetDevOps because it establishes a commercially viable pattern for closed-loop RAN automation at scale using open interfaces, reducing vendor lock-in while delivering quantified performance gains that justify operational complexity. The architecture also supports Verizon's stated goal of achieving Level 4 network autonomy through distributed, intent-driven management with active closed-loop execution.
Lumen announced Emerald AI, an intelligent network capability that dynamically reroutes AI workloads to optimize for power, cost, and operational factors, plus quantum-resistant security capabilities via Fortinet-based SASE.
Why it matters Addresses the critical constraint: by 2030, power limits will force 75% of I&O leaders to mandate energy-aware AI placement, making network-level workload routing directly actionable for operations teams.
Lumen Technologies unveiled three initiatives positioning its fiber network for the AI infrastructure era. Emerald AI dynamically reroutes AI workloads based on power consumption, cost, and operational metrics—directly addressing the data center power budget constraint. Lumen also announced quantum-resistant security via Fortinet SASE. The initiatives reflect strategic repositioning: Lumen is moving beyond transport-only services to position fiber, edge sites, and automation as an intelligent control layer for AI and quantum services. Emerald AI connects energy intelligence with programmable network routing, enabling operators to make dynamic workload placement decisions tied to real-time power and cost signals. This matters for NetOps because it directly addresses the operational constraint dominating 2027–2030 planning: energy limits, not compute capacity, will drive AI infrastructure decisions. The integration of network-level visibility with workload placement gives practitioners a new control point for cost optimization.
An interview with Keycloak maintainers on integrating native OpenTelemetry support. Key insight: backpressure in push-based telemetry signals remains a critical blocker for production readiness in legacy and high-stakes systems.
Why it matters Engineering teams adding observability to authentication-critical or legacy systems should understand backpressure constraints in OpenTelemetry push receivers and design telemetry pipelines as controlled boundaries, not passive data routers.
Keycloak, a CNCF authentication project, pursued 'batteries-included' native OpenTelemetry instrumentation on the Quarkus Java framework to improve production observability. Their experience revealed that architectural trade-offs matter more than feature coverage. The central concern: lack of built-in backpressure in push-based signals (metrics, logs, traces sent to an external Collector). If the observability stack itself becomes overloaded or slow, it risks degrading the protected authentication service—an unacceptable outcome in high-stakes systems. Maintainers recommend treating the OpenTelemetry Collector as a controlled boundary: constrain ingestion points, classify data by source trust (unauthenticated UDP multicast vs. authenticated mTLS), restrict network access, and distinguish between data freshness needs (metrics) and safety requirements (auth logs). This is not a simple 'plug in OpenTelemetry' story; it's a lesson in how observability infrastructure must be designed for the criticality and constraints of the systems it observes.
Anthropic released FlashProve research and code for checking whether an AI response genuinely came from the model a provider claims to have used, relying on cryptographic proofs which allow verification without exposing the model's private internal weights, with applications for independent safety evaluations.
Why it matters Enables infrastructure operators to cryptographically verify model provenance without accessing weights—critical for enterprise multi-provider deployments and supply-chain assurance.
Anthropic released FlashProve on November 4, 2026, providing both research documentation and implementation code for a cryptographic proof system that verifies whether a given AI response actually came from a specific model. The system operates through zero-knowledge proofs, allowing users to validate model identity without requiring access to the model's private weights or internal parameters—a significant departure from traditional auditing approaches that require full model inspection. The tool addresses a growing operational concern: in multi-provider AI deployments, infrastructure teams need assurance that API responses claiming to come from a particular model version actually originate from that model, not from a cheaper or less capable substitute. FlashProve enables this verification at runtime without compromising model confidentiality. Anthropic framed the announcement around independent safety evaluation use cases, but the underlying cryptographic mechanism applies directly to SLA verification, compliance attestation, and API integrity checking in production environments. This shifts model verification from a business/sales problem into a verifiable, infrastructure-level security control.
EU AI Act moved into enforcement phase on 2 August 2026. First harmonised standard supporting the Act was published 19 October 2026; no fines announced by 3 November but compliance deadlines now fixed. World AI Regulation Summit (5-6 November, London) brings regulators together to address enforcement challenges.
Why it matters Compliance is no longer theoretical for enterprises serving EU markets—regulators now have technical standards and coordination mechanisms to inspect and sanction non-compliance with high-risk AI system rules.
The EU's AI Act shifted from theoretical framework to active enforcement after the 2 August 2026 applicability date. A November 3 industry report confirms no fines have been levied yet, but publication of harmonised standards and guidelines signals regulators are moving toward enforcement. Key milestones include transparency obligations for AI providers and deployers (published July 2026) and the first harmonised standard (19 October 2026). The World AI Regulation Summit convening in London is coordinating global regulators, legislators, and policymakers on AI Act enforcement challenges, agentic AI liability, and cross-border alignment. For enterprises, this means the implementation window is closing rapidly. Compliance obligations now carry concrete technical standards and regulators across Europe are equipped to audit and sanction violations. Any organization deploying high-risk AI systems in EU markets or serving EU users faces material compliance risk if standards-aligned testing and documentation are not in place.
NetBox Labs and NEYON AI announced a strategic partnership enabling NEYON AI customers to deploy NetBox Labs platform as part of AI factory deployment, combining physical asset topology with real-time diagnostic signals for root cause analysis.
Packet Pushers has been acquired by Meter; guest host Scott Robohn discusses with Ethan Banks and Meter CEO Anil Varanasi how the relationship will work with full editorial independence maintained for Packet Pushers.
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.