You're reading the Friday, November 20, 2026 edition. Today's briefing →
Toronto
Friday, November 20, 2026No. 159

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. Telco·Industry news

    Telcos are writing their own network autonomy code

    Finland's Elisa trained its engineers to write the code that runs its Level 4 network.

    Why it matters As more carriers build their own automation, they're deciding which calls the software can make for itself — moving operational decisions in-house rather than relying on vendor platforms.

  2. Automation·Industry news

    Itential open-sources AI skills library for agentic NetOps

    Itential released open-source AI Skills library providing integrations with AI platforms, permissions controls, execution safeguards, and visibility for agentic network operations.

    Why it matters Provides NetDevOps practitioners with vendor-backed skills framework to safely incorporate AI agents into operations; shifts from monolithic automation to governed, AI-augmented workflows with built-in audit trails.

  3. DC networking·Industry news

    Arista launches R4 routers for edge AI inferencing

    Arista unveiled R4 routers—compact, hardened access routers, coherent transport router, and 25.6-Tbps pluggable optical line system—designed for distributed AI inferencing at hyperscalers and neocloud operators needing 1.6T links with liquid cooling.

    Why it matters Extends AI networking from data-center spine-leaf to metro/edge; enables neoclouds and hyperscalers to link 100k-1M accelerator clusters across multiple facilities with native 1.6T interconnect.

Today's briefing

What happened, and why it matters

20 stories · 10 topics · Updated 1:36 PM ET

Itential open-sources AI skills library for agentic NetOps

Futuriom · Nov 17, 2026 · Industry news

Itential released open-source AI Skills library providing integrations with AI platforms, permissions controls, execution safeguards, and visibility for agentic network operations. Release enables teams to move faster on network automation while introducing governance and controls around AI-driven changes.

Why it matters Provides NetDevOps practitioners with vendor-backed skills framework to safely incorporate AI agents into operations; shifts from monolithic automation to governed, AI-augmented workflows with built-in audit trails.

Itential says its new AI Skills release helps teams move faster and broaden participation in network automation by providing integrations with AI platforms along with permissions, execution controls, and visibility. This includes vendor-specific skills for a wide variety of network infrastructure environments. To safely incorporate AI into operations, network and infrastructure managers need specialized platforms that introduce governance and controls. Itential is taking part in a huge wave of enterprise automation in which infrastructure teams are adopting AI tools to help monitor and make changes to networks. The company believes the next step for infrastructure operations is for teams to describe the outcome they need and have agents translate that intent into action across their environments. Customers want the network to be as deterministic as possible—this is the ultimate in intent-based networking. Itential is careful to draw a line between guidance and execution, with the platform showing the library at AutoCon 6 in Tucson this week.

Read the original at futuriom.com ↗

Nautobot alpha loosens primary IP interface assignment requirement

Nautobot (GitHub/PyPI) · Nov 13, 2026 · Primary source

Nautobot released v3.3.0a2 alpha on Nov 13, 2026, adding PRIMARY_IP_REQUIRES_INTERFACE_ASSIGNMENT configuration setting. When disabled, allows IP addresses not assigned to any interface to be designated as primary IP for Devices, Virtual Machines, or VDCs.

Why it matters Configuration flexibility enables automation of overlay and virtual infrastructure models where primary IPs can float independent of physical interface state; reduces constraints on intent-based network automation patterns.

v3.3.0a2 alpha introduced PRIMARY_IP_REQUIRES_INTERFACE_ASSIGNMENT configuration setting; when disabled, an IP address that is not assigned to any interface can be designated as the primary IP of a Device, Virtual Machine, or Virtual Device Context. This change supports modern network designs using floating IPs, virtual addresses, and overlay networking where primary IP semantics decouple from physical interface membership. Release v3.2.9 (also Nov 9, 2026) fixed Device LLDP Neighbors tab database query performance regression by eliminating per-interface queries when rendering connected endpoints. These are pre-release (alpha) versions available on PyPI; stable v3.2.9 remains the production release. The timing aligns with NautoCon Tucson (Nov 18, 2026) where customer panel from ONEOK, Microsoft, and Fidelity will share production Nautobot operational patterns.

Read the original at github.com ↗

Broadcom pitches disaggregated Ethernet fabric against Cisco Silicon One

IEEE ComSoc Blog · Nov 16, 2026 · Analysis

Broadcom unveiled its open Ethernet fabric portfolio at SC26, contrasting its disaggregated ASIC strategy (Tomahawk Ultra, Tomahawk 6, Jericho 4 paired with Thor Ultra NICs) against Cisco's unified P4-programmable Silicon One architecture for AI workloads.

Why it matters Clarifies architectural divergence in 102.4T switching: disaggregated ASICs versus unified silicon affects tail-latency behavior, fabric programmability, and multi-vendor interoperability for large AI clusters.

Broadcom's SC26 presentation highlighted two competing 102.4 Tbps switching architectures: Broadcom's specialized chipsets (Tomahawk 6 for spine/fabric, Jericho 4 for disaggregated leaf) paired with Thor Ultra NICs, versus Cisco's unified Silicon One G300 architecture. The architectural choice affects congestion management: Broadcom relies on AI-optimized ASICs with cognitive routing, while Cisco embeds Intelligent Packet Flow—absorbing all-reduce micro-bursts directly in Silicon One's on-die packet buffers with real-time fabric telemetry to steer flows. For RoCEv2 deployments, Cisco claims 33% better network utilization; Broadcom emphasizes supplier flexibility and ecosystem choice. Both ship 1.6T-ready silicon in production. For practitioners, the choice maps to procurement philosophy: Broadcom appeals to white-box operators and hyperscalers who want component fungibility; Cisco targets enterprises seeking an integrated platform with unified telemetry and P4 programmability.

Read the original at techblog.comsoc.org ↗

Arista launches R4 routers for edge AI inferencing

Communications Today · Nov 18, 2026 · Industry news

Arista unveiled R4 routers—compact, hardened access routers, coherent transport router, and 25.6-Tbps pluggable optical line system—designed for distributed AI inferencing at hyperscalers and neocloud operators needing 1.6T links with liquid cooling.

Why it matters Extends AI networking from data-center spine-leaf to metro/edge; enables neoclouds and hyperscalers to link 100k-1M accelerator clusters across multiple facilities with native 1.6T interconnect.

Arista positioned its R4 router family to bridge the gap between in-cluster (100k–1M accelerators) and inter-cluster networking. The R4 targets two customer segments: hyperscalers and neoclouds running 100k+ accelerator clusters across multiple data centers requiring 1.6-terabit links with liquid cooling, and enterprises deploying smaller clusters (500–1k accelerators) over air-cooled, lower-speed links. The launch includes Broadcom Jericho-based hardware in form factors suited to metro and carrier deployments, integrated with Arista's EOS operating system. The R4 leverages Arista's cloud routing track record to offer operational consistency for customers already running EOS. This is strategically significant: Cisco, HPE Juniper, and Nokia will pursue the same opportunity, but Arista is claiming first-mover advantage on turning cloud routing into metro/carrier production wins. For practitioners building distributed AI infrastructure, this signals that metro interconnect for AI is moving from proprietary coherent solutions to standard, Ethernet-based, software-driven platforms.

Read the original at communicationstoday.co.in ↗

WACS subsea cable back online after Congo repairs

ITWeb Africa · Nov 19, 2026 · Industry news

The West Africa Cable System (WACS) submarine fibre-optic network was recommissioned on November 16, two days ahead of schedule, following repairs to the cable segment linking the Democratic Republic of Congo and the Republic of Congo after a 29 October disruption.

Why it matters WACS carries critical traffic for Central and West African internet connectivity; rapid repair coordination across multiple consortium members demonstrates effective incident response for submarine cable resilience and geographic redundancy importance.

The West Africa Cable System (WACS) submarine network returned to service after the 29 October fault disrupted traffic, with technical teams isolating the problem to the Congo-to-Congo segment. The repair cycle illustrates operational challenges in submarine cable restoration: coordinating repair vessel positioning, landing station logistics, and multi-party consortium communication across two months of downtime. The system was recommissioned on November 16, ahead of its November 18 target, following interventions at sea supported by teams at Matombi and Muanda landing stations. For operators relying on WACS, the incident reinforces the importance of geographic diversity in international circuits and monitoring landing station health. The two-day ahead-of-schedule completion demonstrates effective consortium coordination but also highlights how single-cable dependencies create extended outage windows when primary routes fail.

Read the original at itweb.africa ↗

SEA-ME-WE 5 fault disrupts Pakistan internet connectivity

Tom's Hardware · Nov 19, 2026 · Industry news

Pakistan's telecommunications authority confirmed that internet connectivity issues affecting the country are related to a SEA-ME-WE 5 submarine cable fault approximately 11 km from Pakistan's shore near the Strait of Hormuz, with traffic rerouted through alternative links while Transworld Associates investigates root cause.

Why it matters Internet connectivity for South Asia depends on submarine cable resilience; near-shore faults are harder to diagnose and repair than deep-ocean breaks, and geopolitical positioning of critical cable routes creates operational vulnerabilities requiring geographic diversity in international circuits.

Pakistan's telecommunications authority (PTA) on November 19 confirmed that the SEA-ME-WE 5 (SMW5) submarine cable reported a fault at approximately 11 km from Pakistan's shore, causing intermittent connectivity issues and service quality decline. The incident illustrates several operational realities: submarine cable faults near shore tend to be harder to pinpoint than deep-ocean breaks, suggesting the possibility of human interaction (anchors, fishing equipment) or infrastructure damage. Transworld Associates was coordinating with the SMW5 consortium to identify the root cause and determine restoration time; internet traffic was being rerouted through alternative international links to minimize impact and ensure service continuity. For Pakistan-dependent operators, the incident underscores the geopolitical vulnerabilities of Strait of Hormuz routing and the importance of geographic diversity in international circuits. The lack of an immediate restoration timeline reflects the challenges of mobilizing repair vessels in contested waters and coordinating across multiple consortium members.

Read the original at tomshardware.com ↗

Critical OpenSSL Base64 overflow waits for December fix

CyberSecurityNews · Nov 20, 2026 · Industry news

A critical CVSS 9.8 heap buffer overflow in OpenSSL's Base64 encoding function (CVE-2026-121481, published November 18, 2026) can corrupt application memory when processing very large inputs. Fixed versions 4.0.4 and 4.1.1 are scheduled for December 1, 2026.

Why it matters No known exploitation or CISA KEV entry exists as of November 19, but widespread OpenSSL usage in TLS implementations across network infrastructure makes this high-priority for pre-patch staging before December release.

The vulnerability is a heap buffer overflow in OpenSSL's Base64 encoding function that can corrupt application memory when the library processes very large inputs. This affects the fundamental cryptographic library underpinning TLS in millions of network devices, load balancers, VPNs, and API gateways. While no known exploitation or CISA KEV entry exists as of November 19, the December 1 release window leaves a narrow patching window before widespread adoption of fixed versions. Network teams running inline encryption, SSL inspection, or VPN termination should audit their OpenSSL versions immediately: affected versions include OpenSSL 3.x and 4.x across multiple release lines. Organizations should plan patch testing now to avoid December emergency patching cycles. The bug resides in the encoding path, so systems performing Base64 operations on large payloads—common in API gateways and content filters—are at highest risk.

Read the original at cybersecuritynews.com ↗

CISA flags exploited LiteLLM and Synology flaws

CISA · Nov 20, 2026 · Primary source

CISA added four actively exploited vulnerabilities affecting Apache Axis, Vanderbilt REDCap, BerriAI LiteLLM, and Synology DiskStation Manager; CVE-2024-10441 in DSM is being used to deploy C3RB3R ransomware on NAS devices, and CVE-2026-40217 in LiteLLM enables LLMjacking for unauthorized LLM access.

Why it matters Federal agencies must patch DSM, REDCap, and Axis by November 21, 2026; infrastructure teams operating internet-exposed NAS or legacy Java frameworks face immediate ransomware and RCE risk.

CVE-2024-10441 in Synology DiskStation Manager is being exploited to deploy C3RB3R ransomware on network-attached storage devices. More than 380,000 Synology NAS appliances remain internet-exposed as of November 2026, with C3RB3R actively scanning for unpatched DSM versions. Apache Axis 1.x is affected by CVE-2026-90436 (CVSS 9.8), an unsafe reflection flaw enabling remote code execution; because this software has reached end of life, users are urged to migrate to alternative frameworks. CVE-2026-90817 in Vanderbilt REDCap allows code injection, which threat actors have used to deploy cryptocurrency miners. CVE-2026-40217 in BerriAI LiteLLM is being exploited for LLMjacking, allowing unauthorized access to large language models across global sectors. Network and infrastructure teams must prioritize Synology DSM patching for internet-exposed devices and migrate off Apache Axis entirely, as the framework is unsupported.

Read the original at cisa.gov ↗

Palo Alto Networks buys Applied Compute for $4.5 billion

SecurityWeek · Nov 18, 2026 · Industry news

Palo Alto Networks agreed to acquire AI infrastructure company Applied Compute in an all-stock transaction valued at $4.5 billion. Applied Compute's platform enables organizations to adapt open-weight AI models to their own data and run them in production across public clouds and sovereign environments.

Why it matters Infrastructure and security practitioners need to track how security vendors now control AI training and deployment infrastructure, shifting the attack surface for enterprise AI operations.

Applied Compute reached $100 million in annualized revenue run rate just 18 months after founding and raised $350 million at a $3.25 billion valuation weeks before this acquisition announcement. The $4.5 billion all-stock deal represents a 38% premium. Palo Alto Networks plans to offer the platform to its security customers with deployment options in public clouds and sovereign environments. This follows Palo Alto's earlier 2026 acquisitions of CyberArk, Chronosphere (observability), Koi (agentic endpoint security), Portkey (AI gateway), and Console (AI agent platform). For NetDevOps and infrastructure teams, this consolidation means observability and control of AI model training pipelines are increasingly embedded in security platforms rather than standalone ops tools, fundamentally changing how infrastructure teams manage AI workload deployment and compliance.

Read the original at securityweek.com ↗

People Moves

2 stories

SCTE chief Maria Popo steps down as CableLabs absorbs operations

Light Reading · Nov 19, 2026 · Industry news

Maria Popo, president and CEO of the Society of Telecommunications Engineers (SCTE), is stepping down at year-end following the organization's operational integration into CableLabs. CableLabs CTO Mark Bridges will become CableLabs CEO, succeeding retiring Phil McKinney.

Why it matters Leadership transitions at standards bodies and cable industry infrastructure organizations signal strategic shifts in how network engineering community and industry collaboration is structured.

Maria Popo is exiting her role as SCTE president and CEO at the end of 2026 as the organization completes its operational integration into CableLabs. Popo has been a long-time cable industry executive and played a lead role in developing the Strategic Threat Response & Infrastructure Knowledge Exchange (STRIKE), a cross-industry initiative addressing network vandalism and national security threats that now includes participation from AT&T, T-Mobile, and Verizon. Separately, CableLabs is also undergoing a CEO transition: Mark Bridges, the organization's CTO, will become CEO, succeeding Phil McKinney, who joined CableLabs in 2012 from HP and is now retiring. The dual transitions reflect consolidation and leadership evolution within the cable and broadband engineering ecosystem during a period of increased focus on network resilience and infrastructure security.

Read the original at lightreading.com ↗

Lutfu Kitapci to lead Telecom Infrastructure Partners

Telco Magazine · Nov 17, 2026 · Industry news

Lutfu Kitapci, Chief Customer Officer and Managing Director of ISP at Hyperoptic, will become CEO of Telecom Infrastructure Partners (TIP) effective December 1, 2026, succeeding interim CEO John Thompson following the passing of founder and former CEO Eric Overman.

Why it matters Leadership change at TIP, a major telecom infrastructure player, reflects continuity planning in carrier and infrastructure development; Kitapci brings ISP operational and customer management experience to the role.

Telecom Infrastructure Partners announced that Lutfu Kitapci will assume the CEO role effective December 1, 2026. Kitapci comes from Hyperoptic, where he served as Chief Customer Officer and Managing Director of ISP operations. He succeeds John Thompson, who has served as Acting CEO while also holding the CFO role since December 2025, following the passing of TIP's founder and former CEO Eric Overman. TIP is a U.K.-based telecom infrastructure company focused on fiber and connectivity partnerships with property owners and tenants. Kitapci's appointment represents a permanent leadership transition after an interim period and brings direct ISP operations and customer management expertise to lead the company's growth strategy.

Read the original at telcomagazine.com ↗

Telcos are writing their own network autonomy code

Fierce Network · Nov 19, 2026 · Industry news

Finland's Elisa trained its engineers to write the code that runs its Level 4 network. Lumen, Vodafone and Orange are also building their own network automation, with carriers setting the rules for what agents can do and when a human needs to step in.

Why it matters As more carriers build their own automation, they're deciding which calls the software can make for itself — moving operational decisions in-house rather than relying on vendor platforms.

Elisa's engineers wrote the software that runs its Level 4 autonomous network, with COO Sami Komulainen confirming "Overall, level four." In TM Forum's framework, Level 4 means a network can anticipate problems, decide how to respond within goals set by the operator, act and check the result. Engineers set the limits. The system can handle routine decisions without waiting for somebody to approve each one. This reflects a broader industry pivot: telcos are no longer outsourcing network autonomy entirely to vendors like Nokia or Ericsson. Instead, operators including Lumen, Vodafone and Orange are investing in internal automation engineering teams. Nokia's Vivek Jaiswal noted "We are shooting for level four," signaling vendor acknowledgment of the shift. For NetDevOps and SRE leads, this matters because it changes the technology stack decision: telcos are building proprietary control planes and agent frameworks rather than adopting off-the-shelf RAN automation platforms, favoring operational sovereignty and predictability over vendor lock-in.

Read the original at fierce-network.com ↗

Asian operators lead surge in Level 4 autonomy certifications

TelecomTV · Nov 19, 2026 · Industry news

The number of Level 4 autonomous network certifications has jumped from just 15 across six service providers to 53 across 20 service providers. Thailand's True, South Korea's KT Corp. and Vietnam's Viettel are the latest to have achieved AN Level 4.

Why it matters Real-world Level 4 deployments are moving from pilot phase to operational scale across RAN, core, transport and service assurance, informing what automation patterns will work at production scale.

Level 4 certifications tripled from 15 to 53 across service providers in a single period, with Thailand's True, South Korea's KT Corp., and Vietnam's Viettel recently achieving Level 4 status. TM Forum leadership challenged delegates to build on these achievements, expressing ambition to see the number of engineers capable of implementing Level 4 use cases grow from 40 to 400 and ultimately 4,000. AI agents are changing telecom network operations from human-led troubleshooting toward machine-led detection and remediation, with safeguards becoming critical as operators allow AI agents to make increasingly consequential network decisions. This acceleration reflects maturation beyond research: telcos in Asia have moved from isolated pilots to multi-domain autonomy across RAN, core, transport and service assurance. For AIOps and SRE practitioners, the expansion signals that closed-loop automation patterns, intent-driven operations, and agent-based decision-making are now proven in production networks, not just benchmarks.

Read the original at telecomtv.com ↗

Operators debate moving agentic AI into live networks

RCR Wireless News · Nov 20, 2026 · Analysis

The Telco Days 2026 event examined AI applications in telecom operations, fraud management and network protection, with emphasis on real-world implementations and operational challenges. Industry participants centered discussion on a fundamental question: how can operators move AI beyond experimentation and deploy it safely in live environments?

Why it matters Agentic AI deployment is moving beyond capability to governance — operational guardrails and safety frameworks are now the blocking issue for scaled production deployment of autonomous agents.

Telco Days 2026, held in Warsaw on November 6, brought together representatives from Software Mind, Amdocs, Alianza, IBM, Next Mobile, Telefónica and Tallence under the theme "Roadmap to Agentic AI and Safer Networks." Agentic AI was a central theme, reflecting industry interest in moving beyond systems that simply analyze information or generate recommendations, with fraud management also a major subject. The focus has shifted from "can we deploy AI agents?" to "how do we govern them safely at scale?" Operators are exploring architectures in which AI processing and generated content remain within the operator's environment — a signal that operational isolation and data sovereignty are becoming non-negotiable prerequisites. For SRE and AIOps leads, this underscores the emergence of safety frameworks as a prerequisite to production agentic automation, not an afterthought. Expected OpEx impact is significant: Bain estimates AI agents and token consumption could account for 20-30% of telco opex under certain operating models.

Read the original at rcrwireless.com ↗

RFC 10056 standardises YANG data types for traffic engineering

RFC Editor · Nov 17, 2026 · Primary source

RFC 10056 publishes standardized YANG data types for traffic engineering, including packet TE types (ietf-te-packet-types module). Provides common type definitions for TE systems across vendors.

Why it matters Enables consistent TE configuration and interoperability across multi-vendor networks; practitioners can standardize TE automation and telemetry collection using these YANG models.

RFC 10056 formalizes YANG module ietf-te-packet-types with revision date 2026-11-17, defining common data types used across traffic engineering implementations. This standardization addresses the fragmentation that occurs when vendors implement TE without consistent type definitions. The module integrates with existing ietf-yang-types and ietf-te-types, allowing operators to build multi-vendor TE orchestration tools that rely on predictable, interoperable data structures. For NetDevOps practitioners, this means you can write Ansible playbooks, Terraform modules, and gNMI collectors that work consistently across Cisco, Juniper, Nokia, and other vendors without custom type mapping layers. The standard covers path constraints, bandwidth definitions, metric types, and other TE-specific constructs. Early adoption of these types in your infrastructure-as-code pipelines reduces technical debt when upgrading or multi-sourcing TE functions.

Read the original at rfc-editor.org ↗

BGP draft would signal multiple next hops per route

IETF IDR Working Group · Nov 19, 2026 · Primary source

Updated draft of BGP MultiNexthop Attribute extends BGP to signal multiple next hops per route with per-nexthop constraints (proximity, transport class, load balance). Version 06 updated November 19, 2026.

Why it matters Enables explicit multi-path steering and nexthop-specific constraints in BGP; operationally relevant for load balancing, segment routing integration, and service function chaining without requiring separate signaling protocols.

Draft-ietf-idr-multinexthop-attribute-06, updated 2026-11-19, formalizes a BGP attribute allowing a single NLRI to carry multiple next-hop addresses with per-nexthop constraints. Key constraint types include proximity checks (single-hop vs multihop flags), transport class ID (color-based routing integration), and load balance factors (percentage-based traffic split). The attribute solves operational problems with multihomed PEs, label oscillation, and forwarding loops in complex topologies. For network operators, this draft enables richer traffic engineering policies directly in BGP without requiring separate MPLS or Segment Routing channels. The mechanism integrates with Flowspec and BGP Label Unicast. Expected to progress toward WGLC at IETF 127 (Nov 2026); early implementation in test labs recommended for operators planning multi-vendor TE deployments.

Read the original at datatracker.ietf.org ↗

New BGP Flow-Spec draft adds payload compression action

IETF IDR Working Group · Nov 16, 2026 · Primary source

New draft proposes BGP Flow-Spec extension allowing headend routers to compress packet payloads based on Flow-Spec match rules. First version published November 16, 2026.

Why it matters Adds compression-as-a-service capability to Flow-Spec; reduces backbone link utilization for compressible traffic (logs, text) without requiring separate middleboxes; operationally useful for cost reduction on expensive leased circuits.

Draft-ietf-idr-flowspec-traffic-compress-action-00 introduces a new BGP Flow-Spec extended community that signals compression algorithms to be applied by headend routers. For packets matching a Flow-Spec rule that are not already compressed, the headend compresses the payload using the indicated algorithm and encapsulates the result. The mechanism applies to uncompressed packets only, avoiding redundant compression. This draft addresses operational scenarios where certain traffic classes (application logs, telemetry, uncompressed video) can be reduced 30–70% in size, directly impacting WAN link utilization metrics. For network operators with constrained backbone capacity, this avoids costly provisioning upgrades by offloading compression to the data plane. Processing overhead and hardware support (must be vendor-specific) are open implementation questions. Early feedback from vendor implementations likely needed before this advances.

Read the original at datatracker.ietf.org ↗

OpenAI agents resolve 600 ChatGPT performance regressions

Rio Rundown, OpenAI · Nov 20, 2026 · Primary source

OpenAI reports that its new Performance Factory utilizes Codex to autonomously identify and resolve performance regressions in ChatGPT; the system has processed over 1,700 investigations and resolved more than 600 cases in three months, replacing manual debugging with agentic triage and human-validated fixes.

Why it matters Demonstrates a production AIOps use case: autonomous systems triage performance anomalies at scale and propose remediation, requiring only human validation before deployment, applicable to any high-traffic service observability model.

OpenAI disclosed the existence of Performance Factory, an agentic system that autonomously investigates performance slowdowns in ChatGPT production and proposes fixes. The system is built on Codex and operates with human-in-the-loop validation: agents identify regressions, trace root causes, suggest remediations, and human operators review and approve before deployment. Over a three-month window, Performance Factory processed 1,700+ investigations and successfully resolved 600+ incidents, suggesting roughly a 35% resolution rate on triage. The system replaces the traditional SRE model of manual performance profiling and root-cause analysis with an agentic workflow. While specifics on latency, false-positive rate, and types of regressions detected are not disclosed, the production deployment at ChatGPT scale indicates technical maturity beyond research prototypes. This is relevant to infrastructure and AIOps practitioners because it demonstrates one pattern for scaling observability and incident response in high-traffic services using reasoning models without proportional SRE headcount growth.

Read the original at riorundown.substack.com ↗

Trump rejects global AI regulation at APEC summit

CASRAI · Nov 19, 2026 · Industry news

President Trump told leaders at the APEC summit in Shenzhen on November 19, 2026 that the United States opposes 'any form of globalist regulation' of artificial intelligence. Xi simultaneously called for AI to remain 'under human control' and for US-China cooperation on AI safety.

Why it matters Signals US preference for domestic AI governance over international frameworks; directly impacts how enterprises navigate compliance across markets and AI supply-chain restrictions.

Trump's November 19 position rejects multilateral AI governance frameworks, stating the US opposes regulation that would 'control or stifle' superintelligence. This comes as a Commerce Department rule on remote access to advanced AI chips was announced November 16, 2026. The move reflects the Trump administration's framing of AI as a strategic competition with China rather than a domain for global coordination. Xi's parallel call for AI remaining 'under human control' highlights the fundamental divergence: the US emphasizes innovation velocity and market competition, while China advocates for cooperative safety frameworks. For enterprises, this geopolitical split complicates data residency, model deployment, and hardware procurement strategies—especially for companies with cross-border operations or international supply chains dependent on advanced chips. The lack of binding international agreements means fragmented compliance landscapes will intensify across jurisdictions.

Read the original at casrai.org ↗

Applied Compute sells for $4.5 billion weeks after raising $350 million

Futurum Group · Nov 19, 2026 · Industry news

Palo Alto Networks announced November 18, 2026 an all-stock acquisition of Applied Compute for $4.5 billion, a platform for post-training and serving open-weight AI models. Applied Compute raised $350M in November with $100M+ ARR.

Why it matters Major security vendor now operating AI infrastructure; directly impacts enterprise architecture for hybrid open/closed model deployments and security posture around locally-deployable AI.

Palo Alto's $4.5B acquisition signals a strategic pivot from security-only positioning toward building and operating AI infrastructure. Applied Compute raised $350 million in November at a $3.25 billion valuation with annualized recurring revenue exceeding $100 million—making this roughly a 1.4x ARR multiple, typical for infrastructure platforms in growth mode. The company claims inference at less than half the cost of frontier alternatives. For infrastructure teams, this acquisition means a major security vendor now operates model-serving infrastructure—blurring traditional boundaries between endpoint protection and AI infrastructure. Enterprises can expect tighter integration of model governance with security controls. The deal closes a gap in the market: large enterprises need somewhere to run non-proprietary models without depending on OpenAI, Anthropic, or cloud providers alone. Applied Compute's platform targets the enterprise segment where cost-efficient, locally-deployable open-weight models compete with vendor lock-in.

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

Listening

Podcasts and talks

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 · MCP · observability · automation · Agentic AI · RAG · LLM · SRE · inference · OpenConfig · AIOps · OpenTelemetry

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.