Oct 21: RPKI now protects 70% of internet routes after China surge
Today's 3 things that matter
- RPKI now protects 70% of internet routes after China surge
Validates RPKI deployment momentum while exposing that 16.2% of covered IPv4 routes use loose maxLength settings, leaving room for attacks like the Hetzner hijack. - Ransomware crews exploit PAN-OS GlobalProtect authentication bypass since June
Despite warnings since May, systems remain unpatched; multiple threat groups are actively targeting internet-exposed GlobalProtect deployments across finance, healthcare, and manufacturing sectors. - Cisco ships Silicon One G300 and unifies Nexus One fabric management
Full-stack approach (G300 silicon + Nexus One + NVIDIA interop) reduces operational friction for enterprises building multi-gigawatt AI data centers; v1 production shipment signals multi-vendor Ethernet fabric maturity for 100k+ GPU scale deployments.
Full stories below, grouped by topic.
Network Automation
NANOG 98 adds network automation BoF and autonomous networks keynote
NANOG · Oct 19, 2026 · Primary source
What happened: NANOG 98 (October 19-21, 2026 in Miami) features a Network Automation Birds-of-a-Feather session led by Ryan Hamel (Zayo) and a keynote by John Evans asking why autonomous networks remain aspirational despite self-driving cars becoming reality. The BoF covers practical automation implementation, lessons learned, and what's working in network automation and software development.
Why it matters: Peer-led forum for practitioners to share real-world automation challenges, solutions, and operational models; keynote addresses the gap between automation tooling and autonomous operations—a key concern for NetDevOps teams.
NANOG 98 brings together network operators and engineers for three days of peer-reviewed technical content. The Network Automation BoF, hosted by Ryan Hamel from Zayo, is specifically designed for practitioners curious about network automation and software development. Attendees swap war stories, share lessons learned, and discuss what's actually working—and what they wish they'd known—when implementing network automation at scale. The opening keynote by John Evans confronts a central challenge facing the industry: self-driving cars progressed from concept to deployment while autonomous network operations remains an industry aspiration. Evans examines what actually enables operational autonomy: configuration automation, network control primitives, automated operations, operational models, machine-actionable telemetry, causal reasoning, and guardrails. This directly addresses the maturity gap many organizations face when moving from scripted automation (Ansible, Terraform) to truly autonomous closed-loop operations—a core concern for SRE and NetDevOps leads scaling beyond manual playbooks.
Read the original at nanog.org
netlab used to test Cisco's open-source OpenEIGRP with FRR
ipSpace.net Blog · Oct 19, 2026 · Analysis
What happened: Ivan Pepelnjak at ipSpace.net published a guide on testing OpenEIGRP (Cisco's new open-source EIGRP implementation) using netlab, a labs-as-code framework. The post demonstrates how to integrate OpenEIGRP containers with FRR daemons and use netlab automation to simplify topology deployment and testing workflows.
Why it matters: Shows practical labs-as-code patterns for CI/CD network testing; demonstrates how container-native network emulation (netlab) integrates with open-source routing stacks for reproducible validation before production.
Ivan Pepelnjak documented how to use netlab—a labs-as-code framework—to test Cisco's new open-source OpenEIGRP implementation alongside FRR (Free Range Routing). Donnie Savage, who announced OpenEIGRP's open beta, delivered a ready-to-use container with OpenEIGRP integrated into the FRR daemon ecosystem and CLI. Pepelnjak's post shows how netlab automates the container orchestration, topology declaration, and test execution, making it practical to spin up reproducible network labs for validation. This exemplifies the CI/CD-friendly approach to network testing: declare topology in code, verify behavior, tear down—all without manual device access or complex lab infrastructure. The pattern is directly applicable to pre-deployment validation pipelines where Batfish checks config semantics and containerlab+netlab verify actual control-plane behavior before production rollout.
Read the original at blog.ipspace.net
AIOps & Network Observability
Inngest traces outage to network partition and PgBouncer pool collapse
Inngest Blog · Oct 15, 2026 · Primary source
What happened: Inngest's production incident report documents a data-center provider network partition (October 9) and subsequent PgBouncer connection pool collapse (October 10), revealing cascading failures across distributed systems and shared event pipelines.
Why it matters: Practitioners need concrete examples of how network partitions propagate failures through dependency graphs; this shows why topology awareness and correlation are essential for NOC incident triage.
On October 9 at 21:05 UTC, Inngest experienced a network partition at one data center provider, losing bidirectional connectivity to other data centers and outbound internet access. Accounts with run state stored in the affected data center could not create, load, or save state. More critically, accounts in other data centers using shared event-processing pipelines were also impacted because failed work from the disconnected accounts built a large backlog, delaying event processing for unaffected accounts. Service was restored at 21:49. On October 10, both PgBouncer connection poolers (running in AWS) simultaneously lost all connections to the primary database at 14:00 and again at 15:30, unable to rebuild connections. A mitigation route through NAT gateways introduced silent connection drops. This incident exemplifies why observability platforms must correlate network topology changes with application-layer failures: the NOC needed to distinguish between a network problem (partition), a database problem (connection pool exhaustion), and cascading secondary impacts on systems without direct dependencies on the affected infrastructure.
Read the original at inngest.com
Data Center & AI Networking
Broadcom co-packaged optics ship as vendors launch 102.4T systems
Dell'Oro Group · Oct 15, 2026 · Industry news
What happened: NVIDIA Spectrum-X Ethernet Photonics and Broadcom Tomahawk 6 Davisson CPO reached production milestone with early customer deployments. Celestica DS6005 delivers 102.4 Tbps in 1.5RU with 64 × 1.6 Tbps co-packaged ports; Micas Networks announced three 102.4T platforms spanning 800G air-cooled, 1.6T liquid-cooled, and 1.6T Davisson CPO variants.
Why it matters: CPO production enables escape from transceiver density and per-link power constraints; Tomahawk 6 Davisson and purpose-built systems provide operators options for scaling beyond 100k GPU deployments with 3.5x power efficiency improvement and simplified optics supply chains.
Co-packaged optics transitioned from lab validation to production deployment at OCP Global Summit 2026. Broadcom's Tomahawk 6–Davisson integrates 16 × 6.4-Tbps optical engines on-die, eliminating external transceiver cages and delivering material efficiency gains: channel loss drops from 22 dB (pluggable) to 4 dB, and per-link power falls from ~30 W to ~9 W. Celestica and Micas Networks each announced 1.6T systems using Davisson, with Micas offering a portfolio spanning 800G air-cooled (4RU, 128 × 800G ports), 1.6T liquid-cooled (2RU, 64 × 1.6T ports), and Davisson CPO (1RU, 64 MMC DR8 optical interfaces). NVIDIA Spectrum-X Photonics also entered production with early customer deployments. The shift addresses thermal and space constraints in hyperscale clusters: dense 1.6T port counts require either aggressive liquid cooling or CPO integration to remain operationally feasible. Pluggable 1.6T transceivers face supply bottlenecks and per-port cost penalties; CPO sidesteps both by pooling lasers and moving optics inside the chassis. Operators targeting 2027 deployments can now specify CPO architectures and begin power-budget planning. Traditional pluggable 800G remains production-ready for near-term builds through 2026–2027, providing a technology bridge.
Read the original at delloro.com
Cisco ships Silicon One G300 and unifies Nexus One fabric management
Cisco Newsroom / CRN · Oct 15, 2026 · Vendor release
What happened: Cisco shipping systems powered by Silicon One G300, a 102.4 Tbps switching silicon for large-scale AI deployments with Intelligent Collective Networking and P4-programmable architecture. Nexus One platform unified to orchestrate heterogeneous silicon (Cisco G300 and NVIDIA Spectrum-X) with single management plane and job-aware GPU telemetry correlation.
Why it matters: Full-stack approach (G300 silicon + Nexus One + NVIDIA interop) reduces operational friction for enterprises building multi-gigawatt AI data centers; v1 production shipment signals multi-vendor Ethernet fabric maturity for 100k+ GPU scale deployments.
Cisco announced production availability of Silicon One G300–powered systems at OCP Summit 2026. The G300 102.4-Tbps switching silicon targets distributed AI training and inference via two architectural advances: (1) Intelligent Collective Networking offloads MPI collective operations (all-reduce, all-gather, broadcast) to hardware, improving job completion time by 28% per Cisco benchmarks; (2) P4-programmable packet processing enables per-flow telemetry and custom actions without ASIC redesign. Nexus One was enhanced to provide unified management across Cisco Silicon One G300 systems and NVIDIA Spectrum-X deployments, with option for NX-OS or SONiC operating systems. Job-aware telemetry correlates network queue depth, packet loss, and link degradation with GPU workload stalls, enabling predictive alerting. Cisco's NVIDIA Cloud Partner (NCP) compliance means enterprises can now specify Cisco N9000 and 8000 systems in end-to-end AI infrastructure without vendor lock-in. Liquid-cooled systems with high-density 1.6T OSFP optics support begin shipping in 2026. For network operators, significance is multi-fold: prior constraints (single-vendor networking, proprietary management) are dissolving; operators can now evaluate fabrics on cost and performance; and Intelligent Collective Networking patterns from Cisco provide alternative to InfiniBand offloads for Ethernet-based clusters.
Read the original at msspalert.com
Routing & Internet
RPKI now protects 70% of internet routes after China surge
IPinfo · Oct 15, 2026 · Analysis
What happened: RPKI coverage jumped from 54% to 70.3% of routed prefixes in one year, driven largely by China's adoption surge from 3.7% to 83%. Invalid routes barely propagate in practice, reaching only ~2 collectors vs ~596 for valid ones.
Why it matters: Validates RPKI deployment momentum while exposing that 16.2% of covered IPv4 routes use loose maxLength settings, leaving room for attacks like the Hetzner hijack.
IPinfo's measurement shows RPKI-valid routing now covers 70.3% of global prefixes and 65% of IPv4 address space, up from 54% and 53.1% respectively. China drove most of the 16.4-point gain, jumping from 3.7% to 83% of its IPv4 addresses. Critically, the data reveals that invalid RPKI routes barely propagate: vantage-point analysis shows they reach only ~2 collectors on average versus ~596 for valid routes, meaning raw invalid-route percentages often exaggerate real-world impact. However, one significant gap remains: 16.2% of IPv4 and 20.9% of IPv6 RPKI-covered routes have permissive ROA configurations with no maxLength constraint, creating 'open windows' like the one exploited in the August 2026 Hetzner BGP hijack where an attacker announced a /24 inside a /16 ROA. Of unprotected space, ~60% is in the US, with about one in four addresses belonging to DoD (mostly unused space for monitoring). The report recommends operators: publish ROAs for all announced prefixes, leave maxLength unset to constrain specificity, and audit ROAs periodically as some operators (Saudi Telecom, SEACOM, Orange Maroc) have lost ROA records without noticing.
Read the original at ipinfo.io
Network Security
Ransomware crews exploit PAN-OS GlobalProtect authentication bypass since June
The Register · Oct 20, 2026 · Industry news
What happened: Palo Alto Networks' Unit 42 confirmed that ransomware crews including Qilin and Inc continue to exploit CVE-2026-0257, a high-severity authentication bypass in PAN-OS GlobalProtect portals and gateways. Attackers are leveraging this vulnerability to gain initial footholds into corporate networks, bypassing MFA to establish unauthorized VPN sessions.
Why it matters: Despite warnings since May, systems remain unpatched; multiple threat groups are actively targeting internet-exposed GlobalProtect deployments across finance, healthcare, and manufacturing sectors.
CVE-2026-0257, disclosed May 13, 2026, continues active exploitation by multiple ransomware groups nearly five months later. Unit 42 investigated attacks since early June across critical sectors. The vulnerability affects PAN-OS GlobalProtect portal and gateway components, allowing unauthenticated attackers to forge authentication override cookies and establish VPN sessions without credentials or MFA. Affected products include PA-Series and VM-Series firewalls with GlobalProtect enabled, and Prisma Access. CISA added this to KEV on May 29 with a June 1 federal deadline, yet organizations continue operating unpatched systems. Recent reporting indicates Settra ransomware operations also targeting the flaw. Attack chains typically involve initial VPN foothold followed by pivot to internal infrastructure, with observed lateral movement to virtualization management appliances like vCenter.
Read the original at theregister.com
Startups & Funding
Cohere to acquire vulnerability-hunting security startup AISLE
PYMNTS · Oct 15, 2026 · Industry news
What happened: Cohere announced a letter of intent to acquire AISLE, a Prague-based cybersecurity startup founded in 2025 that uses AI to discover and help fix software vulnerabilities. AISLE discovered over 500 previously unknown vulnerabilities in its first year, positioning itself as critical infrastructure defense for open-source ecosystems.
Why it matters: As enterprises move AI to production, security integration becomes operationally critical; this acquisition signals AI vendors consolidating security capabilities for regulated environments and on-premises deployments.
Cohere, a Canadian AI developer, signed a letter of intent to acquire AISLE to strengthen security capabilities for government and business customers in regulated industries operating on private infrastructure or air-gapped systems. AISLE, founded in 2025 with offices in San Francisco and Prague, has built autonomous systems that discover previously unknown vulnerabilities, validate findings, and help develop fixes. The company claimed to have found over 500 vulnerabilities across more than 100 codebases in its first year, outperforming competing systems. Cohere CEO Aidan Gomez stated AISLE has defended critical open-source infrastructure better than many well-known efforts. Following the deal, AISLE's CEO would become senior vice president of a new Cohere Cyber unit selling cybersecurity tools to government and enterprise customers. This acquisition expands Cohere's European footprint following its September 2026 agreement to combine with German AI firm Aleph Alpha. The deal reflects broader industry consolidation around AI security as enterprises grapple with production-readiness governance.
Read the original at pymnts.com
People Moves
Optiv names Michael Saintcross president of federal business
GovConWire · Oct 21, 2026 · Primary source
What happened: Optiv + ClearShark has appointed Michael Saintcross, most recently senior vice president of revenue, as president. In his new role, Saintcross will head the company's federal business and serve on Optiv's executive leadership team.
Why it matters: Saintcross will focus on expanding the company's footprint in U.S. public sector, providing government agencies with technical expertise and engineering capabilities, with focus areas on managed services and localized engineering.
Optiv + ClearShark promoted Michael Saintcross from Senior Vice President of Revenue to President on October 21, 2026. Saintcross is a technology executive with over 25 years of leadership experience across commercial and public sectors; he previously headed the U.S. federal team at BeyondTrust, oversaw U.S. federal and Eastern regions at Illusive Networks, led FireMon's U.S. federal and managed services practices, and held various roles at Sun Microsystems. As President, he will head the company's federal business on Optiv's executive leadership team and advance the company's mission to provide government agencies with technical expertise, engineering capabilities and partnerships, with focus on managed services and localized engineering. For federal security and infrastructure practitioners, this reflects Optiv + ClearShark's continued investment in federal market penetration, elevating a revenue-focused executive to operational leadership to drive government sector expansion.
Read the original at govconwire.com
Telco & Cable AI
Nokia and NVIDIA push AI-RAN platform into live operator trials
Manila Times · Oct 21, 2026 · Vendor release
What happened: Nokia and NVIDIA announced commercial AI-RAN platform enhancements applying AI within radio processing and extending governed AI agents for cross-domain network operations. BT, Deutsche Telekom, Elisa and Orange are piloting the platform on live networks.
Why it matters: AI-RAN moves from lab trials to live network integration; operators gain practical path to improve 5G service quality and begin evolution toward AI-native 6G infrastructure.
Nokia announced next steps in bringing AI-native networks to commercial reality with NVIDIA, introducing new capabilities applying AI within radio processing, supporting intelligence close to data generation, and extending governed AI agents for cross-domain network operations. Building on the industry's first commercial AI-RAN platform announced in July, AI-RAN is being integrated into live networks and moving closer to deployment across multiple operators. BT, Deutsche Telekom, Elisa and Orange provide growing evidence of AI-RAN progress among European operators, with pilots running on live networks. Together, Nokia and NVIDIA are creating the real-time network foundation for physical-AI applications, giving operators a practical path to improve today's 5G and 5G Advanced networks while evolving toward AI-native 6G architecture.
Read the original at manilatimes.net
AT&T avoids 3.1 million dispatches using AI predictive maintenance
TelecomLead · Oct 21, 2026 · Industry news
What happened: Telecom operators are deploying AI agents for predictive maintenance and fault diagnosis. AT&T avoided 3.1 million unnecessary technician dispatches; NTT DOCOMO reduced complex-fault response time by 50%; China Mobile reduced mean time to repair by 30%; Deutsche Telekom reduced incident management from hours to one minute.
Why it matters: AI-driven predictive maintenance delivers immediate, measurable ROI in operations: fewer dispatch costs, faster resolution, reduced downtime. Core AIOps use case with proven business case across multiple operators.
AT&T avoided 3.1 million unnecessary technician dispatches, NTT DOCOMO reduced complex-fault response time by more than 50 percent, and China Mobile reported a 30 percent reduction in mean time to repair. Deutsche Telekom demonstrated autonomous AI agents reducing network incident management from hours to approximately one minute. NTT DOCOMO began commercial deployment of its AI-agent network maintenance system in February 2026, with the platform analyzing traffic and alarm information from over one million network devices including mobile base stations and core-network infrastructure. Multiple AI agents cooperate to identify abnormalities, locate potential failure points and recommend corrective actions. Faster diagnosis improves service availability and reduces maintenance workload. The strongest opportunities extend across radio networks, fiber infrastructure, power systems and network operations centers.
Read the original at telecomlead.com
Research, Standards & Industry
RIPE Atlas study finds IP geolocation providers often inconsistent
arXiv · Oct 19, 2026 · Research
What happened: Longitudinal study (January–June 2026) comparing RIPE Atlas RTT measurements against IP geolocation provider location claims for 5.6M+ unique IPs using multilateration distance constraints. Finds consistency variance across four major providers (IPinfo, DB-IP, MaxMind, IP2Location) and distinct degradation patterns by IP type and address family.
Why it matters: NetOps teams using IP geolocation for traffic steering, DDoS mitigation, or compliance must validate provider consistency empirically against measured RTT constraints rather than assuming provider accuracy across all address ranges.
Paper arXiv:2610.27318 (submitted 19 Oct 2026) presents the first large-scale empirical validation of geolocation providers using RIPE Atlas's global probe network. Methodology: analyzed 2.6+ billion measurement results from January–June 2026, testing whether geoprovider location estimates violate speed-of-light distance constraints given measured RTT values. Key findings: consistency levels remain mostly stable longitudinally but show distinct provider-specific degradation patterns. IPinfo leads DB-IP by 5.4 percentage points for IPv4 and 11.5 percentage points for IPv6 consistency rates. Different providers degrade differently on distinct IP address subsets (mobile vs. backbone, IPv4 vs. IPv6, routers vs. endpoints). For operations, the practical implication is clear: geolocation provider accuracy is not uniform. If you rely on geolocation for steering or analytics, run your own RIPE Atlas measurements to profile your chosen provider's behavior in your specific topology. The paper provides methodology and benchmarks for doing so at scale.
Read the original at arxiv.org
AI Model Providers
Anthropic, Google and OpenAI line up new models for 2026
Emergent · Oct 19, 2026 · Industry news
What happened: Three leading AI research organizations—Anthropic, Google, and OpenAI—are simultaneously preparing major new model releases expected later in 2026, signaling coordinated pushes toward more capable systems across reasoning, multimodal understanding, and real-world task performance.
Why it matters: For infrastructure teams evaluating LLM platforms and inference costs, this signals imminent capability shifts and a competitive cycle that will affect model selection and workload planning through year-end.
According to The Information report published October 19, 2026, all three organizations are advancing next-generation systems with anticipated improvements in complex reasoning, multimodal capabilities, and reduced hallucination rates. Industry observers expect staggered announcements as each completes safety testing and deployment preparation. The timing reveals coordinated competitive pressure suggesting organizations building on current models (GPT-6, Claude Opus 5, Gemini 4 Argon) face obsolescence risk within months. For infrastructure practitioners, this underscores the need to architect inference systems with flexibility for rapid model swaps rather than deep vendor lock-in, and to plan refresh cycles for compute and memory provisioning as capability jumps arrive.
Read the original at emergent.sh
AI Industry & Policy
AI regulators shift from warnings to active enforcement and penalties
Origin Brief · Oct 19, 2026 · Analysis
What happened: Coordinated state AG litigation against OpenAI, EU AI Office compliance requests to frontier labs, and the first OISP civil penalty signal the end of voluntary compliance era. The US rebranded its AI safety apparatus as 'super intelligence' promotion with explicit mandate to resist foreign regulation, while the EU activated enforcement powers, creating compliance asymmetry for multinational developers.
Why it matters: Organizations without operationalized AI governance programs face immediate legal exposure across multiple jurisdictions simultaneously.
The AI governance landscape crossed a threshold from regulatory threat to active enforcement this week. Coordinated state AG litigation against OpenAI, EU AI Office compliance requests to frontier labs, and the first OISP civil penalty for an AI investment collectively signal that the era of voluntary compliance and self-regulation is ending. The US rebranded its AI safety apparatus as 'super intelligence' promotion with explicit mandate to resist foreign regulation of American AI, while the EU activated enforcement powers and MEPs called for binding AI liability legislation. This creates a compliance asymmetry that multinational AI developers must navigate: what satisfies US policy expectations may actively conflict with EU legal obligations. The convergence of accelerating AI adoption (EU strategic sector use doubled to 18%), active enforcement (EU AI Office requests, US OISP penalty), and public skepticism (55-77% of Americans say AI moves too fast) creates a governance pressure cooker that will accelerate regulatory action globally.
Read the original at originbrief.app
Illinois governor creates cabinet to draft state AI policy
Capitol News Illinois / Insurance Journal · Oct 19, 2026 · Industry news
What happened: Gov. JB Pritzker signed an executive order establishing an AI Cabinet of experts and state agency representatives to formulate policy proposals through 2027 and analyze risks posed by AI to Illinois residents and critical infrastructure. State-level regulatory momentum continues amid federal inaction.
Why it matters: State-level AI governance frameworks proliferate amid federal inaction, creating regulatory fragmentation that enterprises must navigate across jurisdictions.
Illinois is making a move to tackle state-focused AI regulation as the federal government continues to take a hands-off approach. Gov. JB Pritzker set up a new group within state government to explore more regulations for artificial intelligence as calls grow across the country for new restrictions. Pritzker signed an executive order to establish an AI Cabinet made up of experts and representatives from several state agencies; the unpaid group will formulate policy proposals through 2027 and analyze and assess the various risks posed by AI to the safety of Illinois residents and the critical infrastructure residents rely on. This continues the pattern of state-led governance initiatives following federal hesitance to establish comprehensive regulation.
Read the original at insurancejournal.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.