Nov 3: Attackers exploit Cisco ASA and FTD VPN zero-day
Today's 3 things that matter
- Attackers exploit Cisco ASA and FTD VPN zero-day
Over 51,000 ASA SSL VPN devices remained internet-exposed—firewall teams must immediately patch or disable secondary authentication on internet-facing tunnel groups; no workaround exists except patching. - Marvell to buy Lumentum for $115 billion
Signals consolidation in AI networking semiconductors; optical interconnect becomes table-stakes for data center operations teams planning infrastructure expansion. - AI and automation dominate NANOG 98 and AutoCon 6
Practitioners need to know what operational patterns and automation strategies are being validated at the community level before adopting them at scale.
Full stories below, grouped by topic.
Network Automation
AI and automation dominate NANOG 98 and AutoCon 6
Buttondown · Oct 29, 2026 · Industry news
What happened: NANOG 98 (Miami, mid-October 2026) emphasized network automation lessons learned and software development practices. AutoCon 6 (Tucson, November) features talks on IPv6 deployment automation, manual task transformation, network automation as microservices, and AI-assisted learning.
Why it matters: Practitioners need to know what operational patterns and automation strategies are being validated at the community level before adopting them at scale.
NANOG 98 brought together network operators in mid-October 2026 with Telegeography's Paul Brodsky delivering the opening keynote on global Internet infrastructure architecture shifts. The Network Automation BoF emphasized peer-to-peer storytelling and lessons learned—specifically what's working and what teams wish they'd known when implementing network automation and software development practices. The keynote from John Evans addressed the gap between self-driving cars and autonomous networks, noting that 'driving' in networking means different things: configuration automation, network control, or automated operations.
Automatic 6, scheduled for Tucson November 16–20, will feature production-grade talks including IPv6 deployment automation, transforming manual tasks into automated workflows, network automation as a series of micro-services, and AI-assisted learning for operators. Lightning talks and a two-day hands-on workshop track precede Thursday's longer tutorial-style talks and standard sessions. The agenda reflects continuity with AutoCon 5 (June 2026, Munich), which featured talks on GitOps delivery patterns, ops-led grassroots automation, fully automated backbone networks managing 10,000+ devices, and navigating the LLM landscape for network engineers.
Read the original at buttondown.com
Data Center & AI Networking
CoreWeave AI backbone reaches 51 data centers on four continents
CoreWeave · Oct 27, 2026 · Primary source
What happened: CoreWeave announced expansion of its Global AI Network to 51 active data centers across North America and Europe, with new PoPs in Asia Pacific and Middle East coming online in Q4 2026, plus a 22.5 MW renewable-powered data center in Copenhagen using NVIDIA Blackwell and high-speed networking.
Why it matters: Global backbone connecting AI data centers across four continents with non-blocking multi-plane fabric architecture eliminates bottlenecks across entire path from data to compute, enabling distributed training and inference at hyperscale with integrated resilience.
CoreWeave announced multi-continent expansion of its Global AI Network on October 27. The company now operates 51 active AI data centers across North America and Europe, with eight added during 2026. New Points of Presence are coming online in Q4 2026 across Asia Pacific and Middle East. CoreWeave brought the majority of 22.5 megawatts of IT capacity online at North's DEN01 data center in Copenhagen, powered by renewable energy and using NVIDIA Blackwell systems and high-speed networking, expanding European footprint to nine data centers across five geographies. An additional new PoP will connect three existing data center facilities in Indonesia announced in August, and a new PoP in India will expand capacity to 240 MW in Navi Mumbai. CoreWeave's global backbone uses the same design philosophy as its intra-data-center fabric: eliminate avoidable bottlenecks across the entire path from data to compute. Inside each data center, interconnected fabric operates as a non-blocking, multi-plane architecture in which every leaf has sufficient uplink capacity to sustain full line rate from all nodes beneath it. GPU-to-GPU compute uses NVIDIA Quantum-2 InfiniBand (3.2 Tbps per node capacity) in a one-to-one non-blocking architecture. For infrastructure teams evaluating GPU cloud options, CoreWeave's global backbone represents a shift from point-to-point regional data center clusters toward integrated AI-native fabric spanning continents with built-in resilience and unified orchestration.
Read the original at coreweave.com
Routing & Internet
Cloudflare details the outages that took the internet offline
Cloudflare Blog · Oct 27, 2026 · Primary source
What happened: Cloudflare Radar examined global internet disruptions observed during Q3 2026, analyzing three kinds of events that affected network connectivity and traffic patterns across regions.
Why it matters: Engineering teams must account for connectivity failures outside their own stack—well-built multi-region setups don't protect users when networks between infrastructure and users are down, requiring visibility into internet-level outages.
On October 27, 2026, Cloudflare released its Q3 2026 internet disruption summary, examining disruptions observed by Cloudflare Radar during the quarter. The report synthesizes data across regions and time windows to identify patterns in internet reliability. Practitioners should cross-reference the Q3 2026 disruption windows against their own traffic and error-rate graphs to correlate unexplained dips with internet-layer events. This complements standard application monitoring with internet-level context. The report pattern-matches against regional and temporal outages observed across multiple global vantage points, providing a useful reference for understanding whether customer impact came from infrastructure, traffic engineering, or upstream connectivity.
Read the original at blog.cloudflare.com
Network Security
Attackers exploit Cisco ASA and FTD VPN zero-day
Cisco Security Advisory / CISA · Oct 28, 2026 · Vendor release
What happened: Cisco disclosed CVE-2026-76418, a high-severity zero-day in Cisco Secure Firewall ASA and FTD Remote Access SSL VPN service, after PSIRT became aware of active exploitation in October 2026. The flaw allows unauthenticated remote attackers to trigger device reloads, causing denial-of-service conditions affecting VPN availability.
Why it matters: Over 51,000 ASA SSL VPN devices remained internet-exposed—firewall teams must immediately patch or disable secondary authentication on internet-facing tunnel groups; no workaround exists except patching.
An unauthenticated remote attacker exploits CVE-2026-76418, a logic error in ASA/FTD SSL VPN secondary authentication handling, by sending a crafted request to internet-exposed webvpn interfaces, forcing targeted firewall devices to reload and disrupting availability. Cisco published its advisory on October 28, 2026, assigning the vulnerability a CVSS score of 8.6. CISA added it to its Known Exploited Vulnerabilities (KEV) catalog on October 29, with a federal remediation deadline of November 1, 2026. Cisco has confirmed in-the-wild exploitation during October but has not identified threat actors, targets, or incident count. No workaround is available except disabling secondary authentication or patching immediately. Fixed versions: ASA 9.16.4.104, 9.18.4.96, 9.20.4.50, 9.22.3.200, and 9.23.1.48 for their corresponding release branches.
Read the original at socprime.com
Two critical PHP flaws exploited against Palo Alto firewalls
Palo Alto Networks / CISA / BleepingComputer · Oct 27, 2026 · Vendor release
What happened: CISA added two critical PHP vulnerabilities (CVE-2026-109618, CVE-2026-109619) collectively tracked as PHP2Shell to its Known Exploited Vulnerabilities catalog after evidence of active exploitation. Both bugs rate 9.2 CVSS and allow unauthenticated code execution on vulnerable PHP servers.
Why it matters: Palo Alto Networks confirmed active exploitation targeting management interfaces exposed to the internet; ransomware groups use PHP2Shell against firewalls for initial access—exposed management interfaces must be immediately remediated or air-gapped.
CISA added CVE-2026-109618 and CVE-2026-109619 to KEV on October 27, setting an October 29 deadline for federal civilian agencies to patch. Attackers are exploiting both critical flaws, which affect web shells deployed on internet-facing PHP servers and Palo Alto firewalls with exposed management interfaces. CISA has not disclosed which organizations have been compromised or provided details of attack campaigns. Palo Alto Networks updated its advisory confirming active exploitation of management interfaces exposed to the internet. Security researcher Kevin Beaumont reported awareness of ransomware groups using PHP2Shell against Palo Alto firewalls to gain initial access, and reports indicate Chinese state group exploitation targeting U.S. government systems for espionage.
Read the original at bleepingcomputer.com
Startups & Funding
Marvell to buy Lumentum for $115 billion
CBS News · Nov 2, 2026 · Industry news
What happened: Marvell Technology announced acquisition of Lumentum Holdings for $115 billion to become a category leader in AI connectivity. The deal addresses optical interconnect requirements for AI data center infrastructure as the industry shifts toward optical-based architectures.
Why it matters: Signals consolidation in AI networking semiconductors; optical interconnect becomes table-stakes for data center operations teams planning infrastructure expansion.
Marvell's $115 billion acquisition of Lumentum, announced November 2, 2026, combines Marvell's data infrastructure semiconductors with Lumentum's optical components (lasers, switches) to create an integrated optical-electrical platform for AI data centers. Marvell CEO Matt Murphy stated: "AI infrastructure is undergoing a fundamental architectural shift toward optical connectivity, creating an opportunity that comes along once in a generation." This follows Marvell's February 2026 acquisition of Celestial AI for $3.25 billion, demonstrating aggressive investment in optical interconnect technology. For operations practitioners, this consolidation means fewer independent vendors in critical AI networking infrastructure, with implications for procurement strategies, multi-vendor redundancy, and architectural decisions around scale-out cluster networking. The timing reflects industry consensus that electrical Ethernet alone cannot meet the bandwidth demands of modern GPU clusters; optical solutions are becoming mandatory rather than optional for hyperscale and AI infrastructure operators.
Read the original at cbsnews.com
People Moves
Dropbox cuts 9% of staff and loses CTO
SEC Filing (Form 8-K) · Oct 28, 2026 · Primary source
What happened: Dropbox announced a 9% workforce reduction and the departure of Chief Technology Officer Ali Dasdan effective November 18, 2026. Dasdan led engineering and technical operations for the cloud storage platform.
Why it matters: CTO transitions often precede shifts in infrastructure or platform priorities; operators using Dropbox should monitor engineering leadership continuity and roadmap implications.
Dropbox announced the departure of CTO Ali Dasdan effective November 18, 2026, in a regulatory filing on October 28 disclosing a 9% workforce reduction (approximately 200 employees). Dasdan, who joined as CTO in March 2025 from Atlassian where he led engineering for Confluence Cloud, will receive a pro-rata bonus of $515,000 and lump-sum severance of $1.3 million under a previously disclosed change-of-control agreement. The restructuring, estimated to cost $27–$33 million primarily in severance, affects Engineering, Product, Design and administrative teams. Dropbox expects most restructuring costs in Q4 2026. This follows an earlier leadership transition in May 2026 when co-founder Andrew Houston stepped down as CEO, replaced by insider Ashraf Alkarmi. The company stated the realignment supports long-term strategy focused on key product and growth priorities.
Telco & Cable AI
AI-RAN automation spending forecast to hit $1.1 billion by 2030
TelcoNews · Oct 28, 2026 · Analysis
What happened: SNS Telecom & IT forecasts annual AI-RAN automation and optimization spending will reach $1.1 billion by 2030, driven by operators' expansion of service management and orchestration platforms, non-real-time RAN intelligent controllers, and software upgrades embedding AI directly in RAN platforms.
Why it matters: Quantifies market shift toward AI-driven RAN automation; operators will spend significantly on orchestration layers and intelligence controllers—critical for NetDevOps teams to understand competitive pressure and budget allocation for RAN modernization.
SNS Telecom & IT's latest market research on RAN automation shows annual spending on AI-RAN optimization will climb to $1.1 billion by 2030. The forecast reflects operator investment in three key areas: expanding service management and orchestration platforms that coordinate AI workloads across distributed sites, deploying non-real-time RAN intelligent controllers that make decisions on radio resource allocation and optimization, and rolling out software upgrades that apply AI models directly within RAN platforms—moving intelligence from centralized clouds to the edge. This spending trajectory underscores the industry's commitment to turning radio networks into active intelligence platforms rather than passive bit pipes. The forecast also implies significant opex reduction potential through automated optimization, predictive maintenance and energy management—factors driving near-term adoption even as full commercial deployments remain planned for 2027-2028.
Read the original at telconews.co.uk
Research, Standards & Industry
IETF approves STAMP extension for reflecting IP headers
IETF IESG · Oct 27, 2026 · Primary source
What happened: IESG approved draft-ietf-ippm-stamp-ext-hdr as Proposed Standard on October 27, 2026. Extension enables STAMP to reflect IPv4/IPv6 headers and IPv6 extension headers for end-to-end and hop-by-hop active measurements, including IOAM data field support.
Why it matters: Extends STAMP measurement capability to capture IP header attributes and IOAM telemetry in production networks, improving performance validation and troubleshooting across network paths.
STAMP (Simple Two-Way Active Measurement Protocol, RFC 8762) is the basis for active network performance measurement in modern networks. This extension allows reflectors to capture and return IPv4/IPv6 headers and IPv6 extension headers from test packets, enabling operators to measure and validate IOAM (In Situ Operations, Administration, and Maintenance) data as packets traverse the network. This is critical for troubleshooting path-specific issues and validating hop-by-hop telemetry in segment routing, MPLS, and other modern network architectures. The extension updates RFC 8762 to define procedures for Session-Senders to include IP headers and for Session-Reflectors to reflect them back. This capability is essential for debugging network behavior and validating SLA compliance in complex multi-layer networks by correlating performance metrics with actual packet header contents and IOAM trace data during active measurement sessions.
Read the original at datatracker.ietf.org
VXLAN specification moves into the IETF document stream
IETF IESG · Oct 27, 2026 · Primary source
What happened: IESG approved draft-ietf-nvo3-rfc7348bis as Informational RFC on October 27, 2026. Moves deployed VXLAN protocol from Independent Submission to IETF document stream, enabling standardization of extensions and proper IANA code point allocation.
Why it matters: Opens VXLAN to controlled standardization for forward extensions and interoperability governance through formal IETF standards process rather than vendor-driven approaches.
VXLAN (Virtual eXtensible Local Area Network) has been a de facto standard for overlay networks in data centers since RFC 7348 (2014), but as an Independent Submission lacked formal IETF management of extensions and IANA code points. This approval resubmits VXLAN as an IETF Informational RFC, enabling the NVO3 working group to coordinate extensions. Key technical benefit: allocation of proper IANA code points within the VXLAN header for future work—such as Group Policy Options and other encapsulation enhancements—that requires formal registry management. The 8-byte VXLAN header carries a 24-bit VNI (segment identifier), preserving all existing protocol semantics while establishing the framework for controlled extension development. For operators, this means future VXLAN enhancements will follow transparent standardization rather than proprietary extensions.
Read the original at datatracker.ietf.org
AI Model Providers
Anthropic tells SEC its AI is accelerating its own development
AI Daily Digest · Nov 3, 2026 · Industry news
What happened: Anthropic amended its IPO filing and told the SEC it is close to building AI that speeds up its own development. The S-1/A filing on November 2 provides the first official quarterly look revealing nearly $17B revenue in Q3.
Why it matters: Self-improving AI systems directly impact infrastructure requirements, model iteration cycles, and deployment planning for organizations relying on Anthropic's products.
Anthropic amended its IPO filing and, in the same document, told the SEC it is close to building AI that speeds up its own development. The amended S-1/A filing submitted November 2 marked the first detailed public disclosure of Q3 2026 financial results, showing approximately $17B in revenue. This disclosure is significant because recursive self-improvement in AI models has long been a theoretical goal but remains largely unexplored at scale. If Anthropic achieves meaningful acceleration of its own R&D velocity through AI assistance, it would compress the timeline for frontier model iterations and potentially reshape competitive dynamics in the industry. For infrastructure teams and practitioners deploying Claude models, this signals potential for more frequent model updates and changes to the production roadmap, requiring more dynamic version management and testing protocols.
Read the original at github.com
AI Industry & Policy
Anthropic reports $17 billion third quarter revenue before IPO
StockTitan · Nov 2, 2026 · Industry news
What happened: Anthropic filed an amended S-1 on November 2 reporting $17.0 billion in Q3 2026 revenue, more than 12x year-over-year. The IPO roadshow could begin as early as November 10, with trading targeted before November 26 at a reported $2 trillion valuation.
Why it matters: Validates massive enterprise AI adoption at scale and sets precedent for frontier lab valuations; signals capital availability for inference infrastructure buildout.
Anthropic reported quarterly sales of $17.0 billion for July–September 2026, more than 12 times prior-year levels. The company's updated S-1 filing reflects explosive revenue acceleration ahead of its planned November IPO. The company is expected to command a $2 trillion valuation, roughly double its $965 billion Series H valuation from May 2026. The filing flags AI's catastrophic risks as the company faces growing safety scrutiny. Earliest roadshow date under SEC staff practice is November 10, with pricing as early as November 13 and trading potentially November 16. For enterprise AI infrastructure planning, this signals major capital deployment to frontier model companies and validates the business case for large-scale inference deployments in production environments.
Read the original at stocktitan.net
Canada puts $1.6 billion into Cohere's $7 billion round
Government of Canada · Nov 2, 2026 · Primary source
What happened: Canada announced a $1.6 billion equity investment in Cohere as part of a $7 billion Series E round led by Germany's Schwarz Group, valuing the Toronto-based AI company at $36 billion and creating the largest AI company outside the US and China.
Why it matters: Signals geopolitical AI sovereignty strategy: governments now directly fund critical AI infrastructure as national assets; creates viable alternative for enterprises seeking non-US-aligned AI deployment.
Prime Minister Mark Carney announced Canada's equity stake in Cohere as part of a $7 billion Series E funding round led by Germany's Schwarz Group, with additional backing from NVIDIA and Bell Canada's venture arm. This values Cohere at approximately $36 billion, making it the largest AI company outside the US and China. Canada's $1.6 billion investment will be staged, and as part of the deal Cohere commits to creating at least 1,000 new high-paying Canadian technical jobs and investing at least $1.7 billion in additional computing infrastructure in Canada by 2030. Cohere will also invest over $4 billion in R&D and AI infrastructure through 2031. The federal government will gain an ownership stake and board representation but exact equity percentage was not disclosed. This reflects broader geopolitical alignment: after European AI investments and Middle East compute consolidation, democracies are now directly funding AI infrastructure as critical national assets. For enterprise procurement teams, this signals a viable, sovereignty-compliant alternative for sensitive workloads and multinational governance requirements.
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.