You're reading the Wednesday, October 28, 2026 edition. Today's briefing →
Toronto
Wednesday, October 28, 2026No. 136

Digital Plumber

Plumbing the information age

AI-curated intelligence for people who run networks. Daily coverage of AIOps, network automation, agentic operations, AI infrastructure, security and the vendors shaping them.

Today's 3 things that matter

Picked by the AI editor
  1. Security·Primary source

    Third exploited VeloCloud zero-day lets any user become superuser

    CISA added CVE-2026-114625 (Arista VeloCloud Orchestrator SQL Injection, CVSS 9.9) to Known Exploited Vulnerabilities on October 23.

    Why it matters Attackers are systematically chaining VeloCloud vulnerabilities to achieve persistent high-privilege control over SD-WAN fabrics; SD-WAN operators must treat this as critical priority given the strategic value of orchestrator as central management plane and active exploitation evidence.

  2. Agentic AI·Vendor release

    Itential opens AI skills and MCP server for network automation

    Itential unveiled Itential AI Skills, an open-source portfolio enabling teams to build, operate, and administer the Itential Platform through leading AI coding environments like Claude Code, Codex CLI, GitHub Copilot, and Cursor.

    Why it matters Itential is making its automation reachable from the AI tools engineers already use rather than from its own interface, enabling infrastructure teams to operate agentic workflows through familiar development environments without platform lock-in.

  3. Telco·Industry news

    Comcast self-heals 77% of network alarms with edge telemetry

    Comcast's Octave platform uses billions of daily telemetry data points to make real-time capacity decisions per customer, increasing network capacity by 35% while automating distribution.

    Why it matters 35% capacity gains from data-driven dynamic allocation and 77% of alarms self-healed via AIOps shows concrete ROI metrics for closed-loop automation—directly applicable to sizing optimization and alerting strategy in cable or hybrid wireline networks.

Today's briefing

What happened, and why it matters

15 stories · 8 topics · Updated 12:41 PM ET

Selector launches Loom agent for alert correlation and root cause

Selector · Oct 23, 2026 · Primary source

Selector released Loom, a specialized AI agent within Selector Foundry that automates alert correlation and root cause analysis. Loom ingests scattered alerts from routers, firewalls, cloud gateways, and applications, then reconstructs a single incident with timeline, blast radius, and evidence trail.

Why it matters Reduces manual incident triage from hours to minutes by automating multi-domain alert correlation, directly lowering MTTR for NetOps teams managing complex hybrid environments with tool sprawl.

Loom is part of Selector's Foundry agentic runtime, released October 23, 2026. It solves the core NOC problem: when a network incident occurs, multiple symptoms propagate across independent monitoring tools—BGP changes, interface statistics, application metrics, syslog events—creating alert storms that obscure the actual failure. Loom ingests Selector's correlated telemetry, traces causal relationships across time and topology, and reconstructs what changed, what it touched, and what belongs together as a single incident narrative. The agent outputs timeline, affected resources, topology context, and evidence trail. Downstream agents (TicketMaster, Herald, Artist) inherit this correlated context for investigation, ticketing, and notification. For practitioners managing multi-domain failures, the value is concrete: reducing manual triage, cutting senior engineer escalations, and automating incident context that routes directly to ITSM. Selector customers report 8x improvements in MTTR after deploying Foundry with Loom.

Read the original at selector.ai ↗

Oracle adds congestion management and resilience to Acceleron AI networking

Oracle Official Announcement · Oct 27, 2026 · Primary source

Oracle announced new capabilities in Oracle Acceleron, OCI's suite of networking and infrastructure technologies for running demanding AI workloads reliably, securely, and at scale. Updates include improvements in storage connectivity, congestion management and network resilience.

Why it matters Accelerator performance alone is insufficient at large scale; networking becomes a fundamental constraint where thousands of accelerator systems must exchange data with predictable latency and high availability. Relevant for operators multi-cloud-sourcing AI infrastructure.

Oracle's announcements at AI World 2026 span new high-performance networking through Acceleron AI networking advances, multi-gigawatt NVIDIA Vera Rubin plans, and collaborations with Google, Cerebras, Intel and OpenAI. Oracle Acceleron combines RDMA, multiplanar networking and resilient collective communications to address the fundamental shift that AI infrastructure must be designed as an integrated system. At scale, networking becomes a fundamental constraint requiring thousands of accelerator systems to exchange data with predictable latency and high availability, while Oracle's strategy combines large-scale physical infrastructure with diverse computing architectures including NVIDIA GPUs, Google TPUs, Cerebras inference systems and Intel processors. The announcement reflects growing recognition that infrastructure efficiency must be evaluated across the complete system rather than individual components.

Read the original at oracle.com ↗

Third exploited VeloCloud zero-day lets any user become superuser

CISA Known Exploited Vulnerabilities · Oct 23, 2026 · Primary source

CISA added CVE-2026-114625 (Arista VeloCloud Orchestrator SQL Injection, CVSS 9.9) to Known Exploited Vulnerabilities on October 23. Any authenticated user, including read-only Enterprise tenant accounts, can escalate to superuser operator via SQL injection. Third VeloCloud zero-day since July 2026 signals sustained campaign against SD-WAN management infrastructure.

Why it matters Attackers are systematically chaining VeloCloud vulnerabilities to achieve persistent high-privilege control over SD-WAN fabrics; SD-WAN operators must treat this as critical priority given the strategic value of orchestrator as central management plane and active exploitation evidence.

CVE-2026-114625 requires only network access to orchestrator web interface and a valid authenticated session; minimum role is Enterprise Read Only (lowest built-in tenant role). VeloCloud Orchestrator is exposed by default with no configuration preventing exposure. Fixed in VCO 5.2.3.18, 6.1.3.10, 6.4.2.12; hosted/dedicated orchestrators already patched by Arista. Two attacker-controlled IPs observed exploiting the vulnerability before patches available. This is the third VeloCloud exploited zero-day since July 2026 (CVE-2026-16812, CVE-2026-93952, CVE-2026-114625). The escalation pattern—from unauthenticated flaws to low-privilege user privilege escalation—indicates attackers systematically mapping and chaining weaknesses. Orchestrators patched for July/September flaws remain vulnerable to this flaw. SD-WAN teams must prioritize immediate patching and enforce strict network access controls to orchestrator management interfaces.

Read the original at thehackernews.com ↗

Juniper BGP flowspec overflow can compromise the routing engine

DEV Community / Node.js Security Advisory · Oct 28, 2026 · Primary source

CVE-2026-102178 is a stack-based buffer overflow in Juniper Junos BGP routing daemon triggered by malformed BGP flowspec NLRI during re-advertisement with update-threading enabled, causing daemon DoS or routing-engine compromise. CVE-2026-33789 affects BMP route monitoring. Meanwhile, Node.js CVE-2026-105482 allows NUL-byte truncation in dns.resolve*() functions, enabling bypass of hostname allowlists in SSRF protection code.

Why it matters Juniper BGP crashes affect all peers simultaneously and trigger fabric reconvergence; Node.js DNS bug directly weakens SSRF defenses in API gateways and infrastructure automation, common patterns for hostname validation before resolution.

CVE-2026-102178: Configured BGP flowspec peer can exploit via malformed NLRI during re-advertisement when update-threading enabled. CVE-2026-33789: Stack-based buffer overflow in routing protocol daemon exploitable through BMP by unauthenticated BGP peer (no routing credentials required); requires BMP enabled. Routing daemon handles all BGP sessions, so crash affects all peers and triggers reconvergence. Node.js CVE-2026-105482: Incomplete fix of CVE-2026-48930 (which covered only dns.lookup() and net/tls host options). dns.resolve*() and dns.reverse() truncate names at embedded NUL byte because name passed as C string to c-ares. Example: attacker.example\0.trusted.example resolves to attacker.example without error. Common risky pattern: applications validate hostname against allowlist, then resolve it (typical in SSRF mitigation, API gateways, load balancers). CVE-2026-106072 (TLS verification bypass): Non-canonical IPv4 spellings (0177.0.0.1, 0x7f.0.0.1, etc.) matched against DNS names instead of IP SANs; requires certificate with numeric DNS SAN from trusted CA and glibc-like resolver. Juniper operators should evaluate BGP monitoring necessity and disable BMP where not essential. Node.js users must patch immediately if applications validate hostnames before resolving or connect by IP address.

Read the original at dev.to ↗

Hyperbolic raises $30M to abstract compute across AI clouds

TechStartups · Oct 27, 2026 · Industry news

AI infrastructure startup Hyperbolic raised $30 million in Series A-1 funding led by Walden Catalyst, co-founded by Intel CEO Lip-Bu Tan, to build an abstraction layer for managing compute across multiple AI cloud providers and neoclouds.

Why it matters Addresses infrastructure fragmentation: as AI compute scales across multiple providers with different APIs, ops teams need unified control—directly applicable to multi-provider AI infrastructure management.

Hyperbolic raised $30M Series A-1 led by Walden Catalyst (co-founded by Intel CEO Lip-Bu Tan and former Samsung president Young Sohn), with Dell Technologies Capital, General Catalyst, Vista Equity Partners, and Cambium Capital participating. Total funding reaches $50 million. The startup solves a critical operational problem: as neoclouds, sovereign AI operators, and hyperscalers each offer different control interfaces, infrastructure teams face mounting complexity managing workloads across fragmented providers. Hyperbolic's platform provides unified abstraction similar to how Kubernetes abstracted container orchestration. For SRE and infrastructure operations teams, this represents the next layer of infrastructure abstraction moving from container orchestration into AI-specific resource management. The investor syndicate—combining infrastructure vendors (Dell), semiconductor backing (Walden Catalyst via Intel), and growth-stage VCs—signals this is recognized as a significant operational and market opportunity. Relevant for teams managing heterogeneous AI infrastructure deployments.

Read the original at techstartups.com ↗

Upwind raises $400M as cloud and AI agent security grows

SiliconANGLE · Oct 28, 2026 · Industry news

Cloud and AI security startup Upwind Security closed $400 million Series D at $3.8 billion valuation, with 4x revenue growth in 2026. Platform monitors runtime behavior across cloud workloads, APIs, and AI agents for threat detection.

Why it matters Runtime-based cloud and AI security shifts observability from static posture assessment to live agent behavior monitoring—operationally critical as autonomous systems proliferate in enterprise environments.

Upwind Security raised $400 million Series D at $3.8 billion post-money valuation (reports indicated $300M in early September, additional $100M brought final total to $400M). The company disclosed 4x revenue growth during 2026—significant traction indicating strong enterprise adoption of cloud-native and agentic security. Upwind's architectural approach—gathering runtime telemetry from cloud workloads, API traffic, and AI agent activity—represents a fundamental shift from traditional CSPM (cloud security posture management) toward observable runtime security. For infrastructure, security operations, and SRE teams, this signals the market recognition that static configuration scanning is insufficient for securing autonomous systems operating in dynamic cloud environments. The large Series D round size ($400M) reflects investor conviction that cloud-native and agentic security will command substantial recurring revenue as enterprises operationalize AI agents at scale. Relevant for teams securing heterogeneous cloud infrastructures and deploying autonomous agents.

Read the original at siliconangle.com ↗

Itential opens AI skills and MCP server for network automation

PRNewswire / Itential · Oct 27, 2026 · Vendor release

Itential unveiled Itential AI Skills, an open-source portfolio enabling teams to build, operate, and administer the Itential Platform through leading AI coding environments like Claude Code, Codex CLI, GitHub Copilot, and Cursor. The Itential MCP Server connects external AI applications to platform capabilities, and FlowMCP Gateway brings external MCP tools into workflows and agents.

Why it matters Itential is making its automation reachable from the AI tools engineers already use rather than from its own interface, enabling infrastructure teams to operate agentic workflows through familiar development environments without platform lock-in.

Itential launched Itential AI Skills on October 27, 2026, an open-source portfolio enabling infrastructure teams to build, operate, and administer the Itential Platform through leading AI coding environments, providing a headless way to create platform assets, manage operational work, and administer their environment through plain-language requests. This release extends Itential's agentic infrastructure operations investment. The Itential MCP Server connects external AI applications to platform capabilities, FlowMCP Gateway brings external MCP tools into workflows and agents, and FlowAI provides an agent harness for building and running agents within Itential. AI Skills extends this portfolio with headless development, operations, and administration through leading AI coding tools. For NetOps and infrastructure operations teams, this shifts the operational model from using a proprietary UI to embedding Itential's automation capabilities into tools teams already depend on—Claude Code, GitHub Copilot, and others—lowering friction for adoption and integration with existing development workflows. The open-source architecture also allows teams to extend and customize how agents interact with their infrastructure platform.

Read the original at prnewswire.com ↗

GPT-6.1 Terra reaches general availability in Microsoft Foundry

AI Agents News Brief · Oct 26, 2026 · Industry news

OpenAI's GPT-6.1 Terra is now generally available within Microsoft Foundry, enabling the creation of AI agents capable of reading charts and navigating desktop applications. A 50% API price cut for agentic computer-use signals broader adoption of autonomous desktop automation.

Why it matters Cost reduction on computer-use capabilities and integrated support in Microsoft Foundry lower the barrier for infrastructure and operations teams to build agents that can automate complex visual/interactive tasks—terminal navigation, configuration UI interaction, dashboard analysis.

OpenAI's GPT-6.1 Terra is now generally available within Microsoft Foundry, enabling the creation of AI agents capable of reading charts and navigating desktop applications, accompanied by a 50% API price cut for agentic computer-use. Microsoft is integrating GPT-6.1 Terra into its Foundry platform, enabling agents that can process complex documents and navigate desktop applications. GitHub Copilot is also evaluating GPT-6.1 Terra for enhanced coding capabilities. For operations teams managing heterogeneous infrastructure with proprietary consoles, legacy UIs, and dashboard-based monitoring, this release is significant: agents can now reason about visual content and interact with tools lacking APIs, making automation possible for systems that previously required human intervention. The 50% price reduction on computer-use also shifts the unit economics of desktop automation, making it viable for frequent tasks like security patch validation, configuration verification, and compliance auditing across systems without programmatic interfaces.

Read the original at aiagentsdirectory.com ↗

AT&T and T-Mobile say autonomous networks remain years away

Reuters · Oct 23, 2026 · Industry news

AT&T has deployed AI governance controls including encryption, multi-factor authentication, and human oversight of data access, while automating trouble ticket analysis and implementing some repairs. Though AT&T has automated network aspects, fully autonomous systems remain years away as complex problems still require human intervention.

Why it matters AT&T's explicit deployment of encryption, MFA, and data access oversight signals hardening of AI agent controls—relevant as you scale autonomous workflows in network operations.

AT&T CTO Jeremy Legg acknowledged industry-wide security gaps in open-source AI frameworks and is implementing governance controls including encryption, multi-factor authentication, and human oversight of data access. The company is also verifying third-party AI agents seeking access to customer accounts. While AI has automated network management aspects—such as analyzing trouble tickets to identify fixes and, in some cases, automatically implementing repairs—Legg emphasized that fully autonomous networks remain years away because human intervention is still required to solve complex network problems. This represents a practical acknowledgment that autonomy aspirations announced earlier in the year face real operational and security constraints in production environments.

Read the original at reuters.com ↗

Comcast self-heals 77% of network alarms with edge telemetry

FutureNet World · Oct 22, 2026 · Industry news

Comcast's Octave platform uses billions of daily telemetry data points to make real-time capacity decisions per customer, increasing network capacity by 35% while automating distribution. The operator's AIOps platform has automated pattern identification and self-healing, with roughly 77% of platform alarms now fully self-healed.

Why it matters 35% capacity gains from data-driven dynamic allocation and 77% of alarms self-healed via AIOps shows concrete ROI metrics for closed-loop automation—directly applicable to sizing optimization and alerting strategy in cable or hybrid wireline networks.

Comcast built Octave as a proof-of-concept during COVID when technicians couldn't physically turn up capacity. It recognizes conditions and demands in premises using billions of telemetry data points multiple times daily as the basis for real-time decisions, increasing network capacity by 35% while maintaining customer transparency and reliability. The subsequent AIOps platform layer identifies failure patterns and deploys self-healing automation, resulting in approximately 77% of platform alarms being completely self-healed. Comcast also developed GOLD (Global Outage, Localized Detection), an algorithm for battery backup analysis in locations experiencing power cuts. This represents production AIOps at cable scale—not conceptual pilots.

Read the original at futurenetworld.net ↗

Comcast cuts customer service contacts 20% with agentic AI

TechStory · Oct 26, 2026 · Industry news

Comcast has deployed agentic AI tools to streamline customer service operations, reducing contact rates by roughly 20% and eliminating more than 30 million service interactions year-to-date, as the company undergoes broader restructuring.

Why it matters A 20% reduction in customer service contacts via agentic AI demonstrates force-multiplier economics for ops teams facing margin pressure—but track whether efficiency gains translate to service quality retention.

Comcast has taken a $1 billion restructuring charge for Q4 2026 including layoffs as it addresses broadband and pay-TV pressures. Agentic AI tools have already reduced customer service contact rates by approximately 20% and eliminated over 30 million service interactions year-to-date. This ties directly to Comcast's stated transformation exercise to simplify operations, move faster, and improve decision-making efficiency. The scale (30M+ interactions automated) indicates production deployment, not pilot-phase experimentation.

Read the original at techstory.in ↗

OpenAI ships GPT-6.1 Terra at half the price

BleepingComputer · Oct 26, 2026 · Industry news

OpenAI released GPT-6.1 Terra on October 26 as a fast, cost-effective model excelling in computer use and visual reasoning, handling screen understanding and app actions up to three times faster than GPT-6 Astra at half the price: $1 per million input tokens and $5 per million output.

Why it matters Terra scores 15 percentage points higher than GPT-5.6 Terra on reasoning benchmarks while costing one-third as much, and matches GPT-6 Astra at one-tenth its price per token—a meaningful efficiency shift for automation workloads and cost-sensitive AI deployments.

OpenAI released GPT-6.1 Terra on October 26 as a speedy, cost-effective model excelling in computer use, visual reasoning, and natural writing. It handles screen understanding, app actions, and routine tasks up to three times faster than GPT-6 Astra, at half the price—$1 per million input tokens and $5 per million output. Terra is designed as a faster and cheaper option for tasks such as filling out forms, editing videos, interpreting technical drawings, and automating repetitive desktop work. It replaces GPT-5.6 Terra as the mid-range option between the affordable Luna and the more powerful Sol and Astra models. The model is available from October 26 to ChatGPT Plus, Pro, Enterprise and Edu users, and in Codex, OpenAI's coding tool. For practitioners building agent workflows or agentic automation systems, Terra's combination of speed (3x improvement), lower cost (50% of predecessors), and reasoning parity with Astra makes it a candidate for cost-optimized agent inference at scale, particularly relevant for teams evaluating inference cost structures in production agent deployments.

Read the original at bleepingcomputer.com ↗

DeepMind's Genie 4 generates steerable two-minute 3D worlds

TechCrunch · Oct 27, 2026 · Industry news

Google DeepMind announced Genie 4 on October 27, a research-preview world model generating photorealistic 3D environments with native audio, extending scene duration to two minutes and enabling real-time steering via natural language prompts, with Gemini 4 Argon directing scene planning.

Why it matters Genie 4's native audio generation, improved physics, character interaction, and Gemini integration enable simulation workflows for robotics testing and game prototyping; limited access via waitlist for Google AI Ultra and Gemini API early partners.

Google DeepMind released Genie 4 in research preview on October 27, 2026, generating the most photorealistic and interactive worlds yet with real-time interactive simulation capabilities. The model generates native audio—ambient sound, music, and interactive dialogue—matching the simulated environment, improving upon Genie 3's approach of adding audio post-hoc. Scene duration doubled from one minute to two minutes, with better physics and more humanlike character behavior. Users can type prompts mid-simulation to change weather, camera angle, or add characters. Gemini 3.8 Flash acts as a guiding agent, observing frames and breaking goals into steps. In a robotics research experiment, Gemini 4 Argon (Google's reasoning-optimized flagship model) generated detailed multi-sentence Genie 4 prompts describing robot workflows, with synthetic footage simulated in real time before physical deployment. The rollout is limited to trusted testers, with plans to expand to Google AI Ultra subscribers in Project Genie and early-access partners in Gemini API and Google AI Studio. For practitioners in robotics simulation and game development, this integration of high-fidelity world simulation with large reasoning models provides a testbed for task planning before hardware deployment.

Read the original at techcrunch.com ↗

Treasury proposes rules restricting US capital flows to Chinese AI

Ctrl+AI+Reg (techieray.substack.com) · Oct 26, 2026 · Analysis

The Treasury Department proposed regulations to implement the Comprehensive Outbound Investment National Security Act, extending Washington's technology rivalry with Beijing from chip exports to capital investment flows. AI infrastructure is becoming an instrument of geopolitical leverage with concentrated private-sector influence over national policy.

Why it matters Infrastructure teams must plan for regulatory uncertainty around AI compute provisioning, capital flows, and international data center partnerships as geopolitical controls tighten.

The Treasury's proposed regulations represent a significant expansion of export controls beyond physical semiconductors to capital investment in AI infrastructure globally. This follows earlier restrictions on chip exports and now targets direct investment flows to competitors. For infrastructure operators, this creates a three-layer compliance challenge: understanding which jurisdictions face restrictions, tracking which capital sources are regulated, and modeling how this affects data center siting, chip procurement, and cloud partnership arrangements. The geopolitical dimension extends beyond US-China relations—multiple countries are now negotiating sovereign AI procurement, GPU allocations, and infrastructure partnerships, fragmenting what was once a globally integrated AI compute market. Operations teams deploying AI workloads should expect increasing scrutiny of infrastructure sourcing, security audits tied to geopolitical risk assessments, and potential restrictions on which models, data, and compute resources can coexist in certain jurisdictions.

Read the original at techieray.substack.com ↗

Financial Stability Board sets third-party AI risk practices

The Reliability Layer (thereliabilitylayer.substack.com) · Oct 28, 2026 · Primary source

The Financial Stability Board published final Sound Practices for AI adoption on October 15, establishing Sound Practice 12 requiring firms to manage third-party AI with focus on performance, transparency, data quality, supply chain and concentration risks, and business continuity.

Why it matters Enterprises deploying third-party AI services must now implement contractual visibility, vendor risk assessment, and contingency planning frameworks aligned with FSB guidance—shifting compliance burden to infrastructure and procurement teams.

The FSB's guidance establishes an institutional framework for managing AI vendor risk across the full dependency chain, from foundation-model developers through cloud and compute providers to energy infrastructure. Financial institutions face primary accountability for AI they buy or rely on, requiring contracts that guarantee visibility into AI updates and changes post-deployment. The guidance specifies assessing concentration risk both direct (primary AI vendors) and indirect (subcontracted dependencies), and includes concrete contractual triggers: disclosure of material changes affecting AI behavior, underlying technology dependencies, and contingency scenarios for foundation-model degradation and vendor discontinuation. For infrastructure teams, this means building procurement processes that extract these commitments from vendors, monitoring frameworks that track AI service changes in real-time, and business continuity plans that model provider failure scenarios. The FSB explicitly notes these are "sound practices" not binding standards, but financial regulators in multiple jurisdictions are likely to treat them as de facto requirements, particularly for cross-border AI deployments.

Read the original at thereliabilitylayer.substack.com ↗
Nothing in today's briefing matches that.

Listening

Podcasts and talks
  • Packet Pushers (Network Automation Nerds podcast)

    NAN134: Building Trust in AI-Driven Networking

    Network engineers explore how AI is reshaping operations beyond simple scripts toward self-driving networks, with HPE's Chief AI Officer Bob Friday and VP of Engineering Jose Tellado diving deep into agentic AI workflows and AI-native automation.

  • FastNetMon

    NANOG 98 Event Recap: AI Becoming Part of Everyday Network Operations

    NANOG 98 (October 19–21, 2026) in Miami featured extensive discussion of AI becoming part of everyday network operations and infrastructure planning, with sessions on securing AI infrastructure, designing networks for AI data centers, and making networks more autonomous.

Vendor Radar

Last 7 days · arrows compare with the 7 before

What changed this week

Last 7 days vs the 7 before

Biggest moves

Trending topics

agent · observability · OpenTelemetry · RAG · Agentic AI · AIOps · MCP · Kubernetes · automation · SRE · LLM · GenAI

Get the daily briefing

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.