Nov 11: Attackers exploit SonicWall SSL-VPN flaw to bypass MFA
Today's 3 things that matter
- Attackers exploit SonicWall SSL-VPN flaw to bypass MFA
An attacker with valid alternate-auth credentials can bypass the organization's identity provider (Okta, Entra ID) MFA check, allowing VPN login with password alone. This breaks zero-trust network access for organizations relying on SAML MFA as primary VPN defense; patch deadline is November 20 for federal agencies. - Bay leaves stealth with $50M to police AI agents
With dozens of enterprise customers including Micron, Peloton, and Wix already deployed, this signals enterprise demand for agent execution controls beyond prompt filtering—critical for infrastructure and security teams deploying autonomous systems at scale. - Oracle ships OCCL plugin for resilient AI training networks
OCCL supports multi-planar RDMA networks in vendor-agnostic fashion, directly addressing network-layer resilience as a bottleneck in multi-thousand-GPU training clusters where collective communication reliability determines job completion time.
Full stories below, grouped by topic.
Network Automation
Network to Code sets NautoCon Tucson alongside AutoCon 6
Network to Code · Nov 7, 2026 · Primary source
What happened: Network to Code confirmed NautoCon Tucson as a half-day event on November 18, 2026, co-located with AutoCon 6, featuring platform roadmap updates and customer panels from ONEOK, Microsoft, and Fidelity discussing production Nautobot deployments.
Why it matters: Direct insights from enterprises running Nautobot at scale on compliance, automation workflow execution, and operational validation—practical reference for evaluating network source-of-truth platforms in production.
NautoCon Tucson 2026 is the third official Nautobot Community Event, running 8:00 AM to 12:00 PM MST on November 18, 2026 at The Westin La Paloma Resort & Spa in Tucson, Arizona, immediately before AutoCon 6's main conference sessions. The half-day agenda includes a look at where the Nautobot vision is headed, latest platform updates, and a customer panel featuring network automation leaders from ONEOK, Microsoft, and Fidelity sharing how they run Nautobot in production. This positions Nautobot as a Network Source of Truth and Automation Platform designed for enterprise teams who need extensibility, native job scheduling, and automation-first architecture—as opposed to NetBox's focus on pure documentation and infrastructure management. The event signals Network to Code's confidence in Nautobot's production adoption and continues the company's strategy of positioning Nautobot as the automation-centric fork of NetBox with first-class Jobs engine, GraphQL, and Git integration.
Read the original at networktocode.com
Data Center & AI Networking
Oracle ships OCCL plugin for resilient AI training networks
HPCwire · Nov 10, 2026 · Primary source
What happened: Oracle Cloud Infrastructure released Oracle Collective Communications Library (OCCL), a network plugin for distributed AI workloads that handles congested paths, failed links, and NIC failures. Internal benchmarks show 2-10% increased average NCCL bus bandwidth for AllReduce operations on NVIDIA GB300 systems.
Why it matters: OCCL supports multi-planar RDMA networks in vendor-agnostic fashion, directly addressing network-layer resilience as a bottleneck in multi-thousand-GPU training clusters where collective communication reliability determines job completion time.
Oracle's OCCL addresses a concrete pain point for practitioners running large-scale distributed AI training: network path failures and congestion that cause collective communication operations to stall entire training jobs. The library operates at the collective-communication transport layer, automatically rerouting traffic around link or NIC failures and restoring connectivity when hardware recovers. According to Oracle's benchmarks on 16-128 node NVIDIA GB300 clusters with four GPUs per node, OCCL increased average NCCL bus bandwidth by 2-10% for AllReduce operations. The plugin uses a vendor-agnostic network plugin interface, supporting both single-planar (traditional spine-leaf) and multi-planar RDMA topologies. This fills a gap in the collective-communication stack: while frameworks continue using NCCL/RCCL, OCCL injects networking intelligence for congestion control and failure recovery directly into the transport layer. For operators running large-scale training, this reduces operational friction of keeping distributed clusters healthy across hundreds or thousands of nodes where any single path failure cascades to all coordinating GPUs.
Read the original at hpcwire.com
Broadcom says network performance now limits trillion-parameter AI training
HPCwire · Nov 11, 2026 · Vendor release
What happened: Broadcom announced at SC26 that network performance has become the limiting factor for trillion-parameter AI models and 100,000+ accelerator clusters. Company showcased open Ethernet fabric architecture using Tomahawk 6 and Jericho 4 silicon under ESUN standards as alternative to proprietary InfiniBand.
Why it matters: Broadcom's 102.4 Tbps Tomahawk 6 platform provides open, multi-vendor alternative to NVIDIA's InfiniBand monopoly, enabling 100,000+ GPU cluster deployments at lower cost and without proprietary feature locks like SHARP.
Broadcom's SC26 positioning marks critical inflection point: shift from proprietary single-vendor stacks toward open, multi-vendor Ethernet fabrics for frontier-scale clusters. The company showcased integration of Tomahawk 6 (102.4 Tbps), Tomahawk Ultra (sub-250ns latency scale-up), and Jericho 4 switching silicon across complete rack-to-fabric hierarchies. Key technical contribution: cognitive routing (second generation) providing dynamic packet steering that adapts to traffic patterns, and Global Load Balancing as foundations for AI scale-out. Broadcom is chairing SC26's first workshop dedicated to high-performance fabrics for AI/HPC, signaling industry standardization momentum. Company announced third-generation co-packaged optics (CPO) platforms integrating optical engines directly into switch packages, reducing power consumption versus pluggable-optic approaches. For practitioners: this ecosystem push against NVIDIA's InfiniBand monopoly creates real optionality for teams deploying 100,000+ accelerator clusters. Open Ethernet avoids proprietary feature locks like NVIDIA SHARP (in-switch collective offload). Cost differences are material: Meta benchmarks show ~$650K for Ultra Ethernet vs. ~$1.2M for InfiniBand on 1024-GPU B200 clusters. However, open Ethernet remains less battle-tested at frontier scale compared to InfiniBand's production deployments on xAI Colossus and other verified 100K+ GPU systems.
Read the original at hpcwire.com
Network Security
Attackers exploit SonicWall SSL-VPN flaw to bypass MFA
CISA Known Exploited Vulnerabilities Catalog · Nov 6, 2026 · Primary source
What happened: CISA added high-severity CVE-2026-19868 (CVSS 7.1/7.5) to its Known Exploited Vulnerabilities catalog after SonicWall disclosed evidence of real-world exploitation. The flaw in SonicOS SSL-VPN allows users with valid alternate authentication credentials to bypass SAML authentication controls and SAML-enforced MFA. Fix available in SonicOS 7.3.4-7013 (Gen7) and 8.2.3-8015 (Gen8); SonicWall confirms no workarounds exist.
Why it matters: An attacker with valid alternate-auth credentials can bypass the organization's identity provider (Okta, Entra ID) MFA check, allowing VPN login with password alone. This breaks zero-trust network access for organizations relying on SAML MFA as primary VPN defense; patch deadline is November 20 for federal agencies.
CVE-2026-19868 (CWE-308) fails to enforce SAML controls when users supply credentials for an alternate authentication method; if MFA depends on SAML, that enforcement is also bypassed. This is not unauthenticated—attackers need valid credentials first, but valid local or alternate-auth passwords are sufficient to bypass MFA. SonicWall published the advisory November 6; CISA added it to KEV the same day, confirming active exploitation. Federal agencies must patch by November 20. This directly undermines zero-trust architecture where VPN MFA is the perimeter control, making it a critical priority for any organization running SonicWall SSL-VPN with SAML-based MFA.
Startups & Funding
Bay leaves stealth with $50M to police AI agents
SiliconANGLE · Nov 10, 2026 · Industry news
What happened: Bay, a Tel Aviv AI security startup, raised $50M combining a $37M Series A led by Battery Ventures and $13M seed to intercept and evaluate every action an AI agent proposes before it executes. Bay's Lighthouse system automatically generates policies from 30 days of agent activity, maintaining false-positive rates under 1%.
Why it matters: With dozens of enterprise customers including Micron, Peloton, and Wix already deployed, this signals enterprise demand for agent execution controls beyond prompt filtering—critical for infrastructure and security teams deploying autonomous systems at scale.
Bay emerged November 10 as the first startup to shift AI security focus from prompt/response filtering to agent action enforcement. The company intercepts requests at the device level through native hooks before agents execute, applying learned policies that detect risky actions (unauthorized database deletions, infrastructure changes) in milliseconds. Unlike traditional permission-based controls that check identity, Bay evaluates proposed actions against the full session context and user-defined guardrails. The platform integrates with existing endpoint detection and response tools in under 5 minutes, auto-generating policies from 30 days of baseline activity rather than requiring manual rule writing. Founded in early 2026 by Yuval Zacharia, Aner Morag, and Boris Danilovich—security researchers from military and Preempt Security backgrounds—Bay already has ~35 employees and operational deployments across health, finance, and tech sectors. The $50M round combines $37M Series A (Battery Ventures, leading) with earlier $13M seed (YL Ventures, Striker Venture Partners), plus strategic investors including Micron Ventures, M12, and Microsoft's venture fund. For infrastructure teams, this represents a practical model for centralized AI governance without centralized data ingestion—addressing the compliance and safety gaps that block enterprise agent rollouts.
Read the original at siliconangle.com
People Moves
Meta infrastructure chief Janardhan steps down, Dave Brown succeeds
Bloomberg News · Nov 11, 2026 · Industry news
What happened: Santosh Janardhan, Meta's head of infrastructure and co-head of engineering, will step down in early 2027 and transition to CTO of Meta Compute. Dave Brown, recently hired from Amazon Web Services, will succeed him as head of infrastructure during Meta's $125–145 billion annual infrastructure spending cycle.
Why it matters: Infrastructure leadership changes at this scale directly affect network operations teams' capacity planning and AI data-center deployment strategies across the industry.
Santosh Janardhan, Meta's longtime head of infrastructure and co-head of engineering, announced his departure in an internal memo on November 11, 2026, effective early 2027. Dave Brown, who joined Meta earlier in 2026 after nearly two decades at Amazon Web Services where he oversaw compute and machine learning services, will succeed him. Janardhan will remain through the first quarter of 2027 to support the transition before moving to a CTO role focused on Meta Compute. Brown's background in cloud infrastructure and machine learning aligns with Meta's dual focus on AI model training and data center operations. The handoff is significant given Meta's multi-hundred-billion-dollar capital allocation for AI and infrastructure expansion, with CEO Mark Zuckerberg highlighting infrastructure as a key enabler for the company's AI ambitions.
Read the original at bloomberg.com
Netceed names Eric Updyke group chief executive
Business Wire · Nov 10, 2026 · Vendor release
What happened: Netceed announced Eric Updyke as Group CEO effective November 10, 2026, succeeding Alper Turken. Updyke brings 35+ years of infrastructure leadership, most recently as CEO of Spirent Communications (2019–2025) and prior roles as Amdocs Services group president overseeing 10,000+ employees.
Why it matters: Updyke's appointment to lead Netceed's AI infrastructure, broadband, and energy supply-chain operations directly impacts network operators' access to infrastructure solutions and buildout capacity for large-scale deployments.
Netceed, a supply chain solutions provider for broadband, data center, and energy infrastructure, appointed Eric Updyke as Group CEO on November 10, 2026, replacing Alper Turken. Updyke brings more than 35 years of leadership experience in communications and technology infrastructure. Most recently, he served as CEO of Spirent Communications from 2019 to 2025. Prior to Spirent, Updyke held senior leadership roles at Amdocs, including Group President of Amdocs Services, managing a global business of more than 10,000 employees. Turken had joined Netceed in December 2022, became Group CEO in 2024, and led the company through financial restructuring in 2026 while strengthening broadband operations and establishing an AI infrastructure business. Updyke's priorities will focus on building on Netceed's existing infrastructure strategy and accelerating profitable growth across AI, broadband, and energy markets.
Read the original at businesswire.com
Agentic AI & MCP
Runpod unifies MCP server setup across coding agents
Runpod Blog · Nov 9, 2026 · Primary source
What happened: Runpod released infrastructure automation enhancements including unified MCP server setup across Claude Code, Codex, and Cursor; added knowledge base and tool generation from REST v2 API spec; shipped 25 reference tasks for common infrastructure operations run end-to-end on live infrastructure.
Why it matters: Eliminates setup friction for SRE teams deploying AI agents against compute infrastructure; agents access live resource state through versioned API contracts, enabling safer autonomous operations across Pods, volumes, and GPU scheduling.
Runpod's November 2026 update prioritizes infrastructure-as-code patterns for agentic operations. The key innovation is a unified MCP server that agents access either through a Claude plugin or via HTTP—both API-bound means agents and Terraform plans operate against the same REST v2 spec, eliminating translation layers. The hosted MCP server exposes tools labeled read-only or destructive, enabling clients that support these labels to auto-approve reads and require approval before writes. Twenty-five worked reference tasks cover end-to-end scenarios (ComfyUI on a Pod, Whisper endpoints, volume management) run live on production infrastructure, removing the knowledge gap for operators integrating agent-driven provisioning. The knowledge base surfaces Pod types, GPU capabilities, worker counts, and volume topology—data an agent can query to reason about constraint-satisfaction problems (e.g., scheduling a large model on available capacity). This reduces MTTR for infrastructure troubleshooting by giving agents deterministic access to current system state instead of stale documentation.
Read the original at runpod.io
Telco & Cable AI
GSMA opens contests for RAN automation and 3GPP retrieval
TelecomTV · Nov 11, 2026 · Industry news
What happened: The GSMA's Open Telco AI initiative announced two competitions—TelcoRetrieve and AI Agents for RAN—to enable AI agents to solve complex network engineering problems and improve accuracy of AI systems working with 3GPP technical standards through retrieval-augmented generation pipelines.
Why it matters: These challenges address grounding AI agents in authoritative telecom standards (3GPP) to build trust in autonomous RAN operations while fostering an open ecosystem for competing solutions rather than vendor-specific automation.
The GSMA's Open Telco AI initiative announced two competitions inviting telecom operators, AI researchers, developers and startups to advance AI capabilities for telecommunications, developed in collaboration with ETSI, IEEE, ITU and Huawei. TelcoRetrieve addresses the challenging task of connecting AI-generated answers to 3GPP-specific source documents through an open, vendor-neutral RAG pipeline, improving accuracy when engineers need to trace answers to authoritative standards. The AI Agents for RAN Challenge enables participants to develop autonomous solutions for complex network engineering problems using the Giulia simulator to investigate network scenarios, test interventions and develop solutions. GSMA, ITU and IEEE will host a dedicated webinar on November 18 to introduce the AI Agents for RAN Challenge. The aim is to advance AI capabilities needed for more autonomous radio network operations, helping operators improve service reliability, reduce operating costs and deliver better customer experience.
Read the original at telecomtv.com
Research, Standards & Industry
RFC 10082 defines hybrid post-quantum signatures for CMS
RFC Editor · Nov 7, 2026 · Primary source
What happened: RFC 10082 published November 7, 2026 as Standards Track Proposed Standard. Defines combinations of ML-DSA with RSA, ECDSA, and EdDSA for use in Cryptographic Message Syntax (CMS), enabling hybrid post-quantum signatures in S/MIME.
Why it matters: Post-quantum cryptography migration for S/MIME and code signing now has concrete standard; PKI and certificate infrastructure teams must evaluate implications for document and code signing workflows.
Composite ML-DSA pairs the Module-Lattice-Based Digital Signature Algorithm with traditional signature algorithms (RSA, ECDSA, EdDSA). Specifies three ML-DSA variants—ML-DSA-44, ML-DSA-65, ML-DSA-87—each paired with specific traditional algorithms and digest functions. ML-DSA-65 with ECDSA P-256 uses SHA-512; ML-DSA-87 with Ed448 uses SHAKE256. Companion to RFC 10073 (X.509 encoding), enabling hybrid post-quantum signatures in S/MIME, code signing, and document signing. The Composite ML-DSA AlgorithmIdentifier must omit parameters field entirely. Implementation is in early stages; COTS libraries are still adding support. This establishes the first standards-track path for production hybrid PQC in messaging systems.
Read the original at rfc-editor.org
AI Model Providers
Anthropic agents re-derive Fermat proof in 30 hours
AI Daily Digest (GitHub) · Nov 11, 2026 · Industry news
What happened: Anthropic released Claude Fable 5.2 and Mythos 5.2 on November 9, demonstrating ~1,000 agents re-deriving Fermat's Last Theorem formalization in 30 hours versus 11 days with human steering. On November 10, Anthropic launched Claude Sheets, Whiteboards, and Meetings as integrated office productivity tools during its IPO roadshow.
Why it matters: Demonstrates agentic multi-model coordination achieving 9x speedup on complex mathematical tasks; shift toward embedded AI in team workflows signals new deployment patterns for Claude-based automation and orchestration.
Claude Fable 5.2 was released November 9, 2026. The technical milestone: approximately 1,000 Claude agents collaboratively re-derived the Lean formalization of Fermat's Last Theorem in ~30 hours, compared to 11 days for an earlier research version requiring human researcher steering of a custom harness. This represents acceleration through agent cooperation and parallelization rather than single-model capability gains.
Concurrently, Anthropic launched three new products: Claude Sheets (multi-tab spreadsheets generated from prompts with real-time collaborative editing on web, desktop, mobile; beta on paid plans from $20/month), Claude Whiteboards (shared canvas converting planning discussions into timelines and deadlines), and Claude Meetings (Claude joins video calls as a visible participant; beta on Team and Enterprise plans). These integrate Claude into existing workflows rather than replacing standalone chat interfaces.
Context: On November 10, OpenAI opened Businesses in ChatGPT (with claimable brand profiles and per-business Business Agents) and Google moved Gemini 4 Carbon to general availability. Synchronized releases across three labs indicate competitive acceleration toward productized agent and workflow capabilities.
Read the original at github.com
AI Industry & Policy
EU AI Act prohibitions and transparency rules take effect
itechguides.com · Nov 10, 2026 · Analysis
What happened: Prohibitions, general-purpose AI obligations and Article 50 transparency requirements entered into force as of 10 November 2026. Registration of high-risk systems began in November 2026 with a 72-hour incident reporting window and penalties reaching $1 million for first violations and $3 million for repeat violations.
Why it matters: Enterprise compliance with EU AI Act implementation now determines whether systems can stay online and whether enterprises can continue operating within EU jurisdictions.
The EU AI Act's enforcement timeline has moved from theory to mandatory compliance. As of November 10, 2026, three key enforcement layers are now active: prohibited-practice rules (in force since February 2025), general-purpose AI provider documentation and copyright-policy requirements (in force since August 2025), and Article 50 transparency duties requiring disclosure when users interact with AI systems and marking of synthetic content (in force as of August 2026). The Digital Omnibus amendment (July 27, 2026) simplified certain SME and startup compliance burdens. Registration for high-risk systems began in November 2026 with a mandatory 72-hour incident-reporting window. Penalties are substantial: €1 million for first violations, €3 million for repeat violations. By December 2, 2026, machines will require machine-readable marking and new restrictions on non-consensual intimate material and child sexual abuse material generators take effect. Enterprises in scope must have completed conformity-assessment documentation before deadlines; late compliance now carries enforcement risk.
Read the original at itechguides.com
Australia creates regulator to enforce frontier AI obligations
The Conversation · Nov 9, 2026 · Industry news
What happened: Prime Minister Anthony Albanese announced a new artificial intelligence safety regulator that will enforce penalties for frontier AI companies breaching their obligations. The regulator will require companies to maintain security controls across all development and deployment stages, maintain published safety and security frameworks, and report misaligned behavior or cyber incidents.
Why it matters: Measures are targeted at frontier AI companies building the largest, most advanced models, establishing enforcement mechanisms for safety and security compliance.
Australia's government established the first dedicated AI safety regulator outside the EU, signaling a shift from guidance to enforcement. The regulator will have authority over frontier AI companies developing large advanced models and will enforce new obligations: maintaining security controls across all AI development and deployment stages; publishing and maintaining safety and security frameworks for governance, risk management, and transparency; mandatory reporting of misaligned AI behavior or cyber incidents; and greater engagement with international counterparts and frontier AI developers to promote shared safety expectations and accountability. The measures form part of proposed AI framework legislation outlined at a news conference before the prime minister departed for COP in Turkey. This complements parliamentary joint select committee recommendations and reflects Australia's intent to benefit from AI's advantages while mitigating frontier-model risks.
Read the original at theconversation.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.