Digital Plumber

OCI is pursuing a multiyear buildout of AI data centers with individual projects requiring hundreds of megawatts, having reported adding 850 MW of data center capacity and deploying more than 300,000 GPUs during fiscal first quarter 2027. Oracle Acceleron combines dedicated network fabrics, programmable networking via AMD Pensando DPUs, and direct data paths. AMD Pensando DPUs power Oracle Acceleron SmartNICs, with the companies co-engineering this foundation to move networking, storage, security and infrastructure functions closer to the data path, preserving host CPU capacity for customer applications. The Collective Communications Library helps customers keep distributed AI workloads running when network paths fail by spreading collective traffic across OCI's backend Multiplanar Network. Transparent front-end multiplanar support brings multi-path resilience to customer-facing networks, reducing tail latency impact when a plane fails. Focus on distributed AI suggests OCI targeting multi-rack, multi-pod cluster scales. Part of October Oracle AI World announcements including OpenAI and Google TPU integrations.

Live VM migration moves running workloads between hypervisors while retaining MAC addresses. When it arrives on the new host, it sends a gratuitous ARP message, allowing the new leaf to learn the MAC locally before BGP updates redirect traffic with a single update, allowing the VM to maintain active sessions. With EVPN, MAC learning moves into the control plane. The MAC Mobility Extended Community is a transitive BGP Extended Community carried within Type-2 routes. Every PE that imports the EVPN instance's Route Target through route reflectors will propagate the update. The risk is that a forged MAC address can redirect traffic across the fabric through BGP. Operators can export routing updates using BGP Monitoring Protocol (BMP) to a collector to gain visibility into these control-plane changes and detect anomalous MAC moves before they impact production workloads. This is especially critical as EVPN deployments scale in modern data centers where rapid workload migration is routine.

In July 2025, LACNIC received a complaint that led investigators to three brief BGP hijacks of its address space, lasting about 20, 15 and 5 minutes respectively. The attacker used forged letterhead and a deceptively similar domain to persuade a multinational transit provider to turn up a BGP connection to an unauthorized ASN. Once inside, the attacker used the stolen AS as transit and announced short-lived routes from other spoofed origin ASNs. The genuine holder reported the fraudulent session to the transit provider, ending the session. The key operational lesson: the attacker used social engineering, forged identity documents and a lookalike domain to convince an upstream provider to enable BGP for an unauthorized ASN, which is why onboarding belongs in the routing playbook. The public records do not measure diverted traffic, but the demonstrated failure is the opening of an unauthorized path. A route can look technically plausible after a provider has accepted a forged corporate identity.

Active exploitation has been observed targeting Palo Alto Networks management interfaces exposed to the internet. Palo Alto Networks confirmed that PAN-OS and Panorama are affected by PHP2Shell. Unit 42 reports the vulnerability being used against internet-exposed PAN-OS management interfaces, chained with Linux kernel privilege-escalation bugs to reach root. As of October 30, 2026, patches have been released for all vulnerable versions of PAN-OS 12.2, 12.1, 11.2, 11.1, and 10.2. Fortinet rated CVE-2026-109618 critical with CVSS 9.8; Cisco lists vulnerabilities critical; Palo Alto rates the issue high with CVSS-B score 8.7. Infrastructure teams managing Palo Alto firewalls must immediately apply patches and restrict management interface access to specific networks.

Cisco published the security advisory on October 28, 2026, after learning of active exploitation during October. The flaw stems from logic errors in the Remote Access SSL VPN service when secondary authentication is configured on a tunnel group. A successful VPN login is not required to reach the affected processing path. This represents the second Cisco firewall VPN zero-day since August 2026. The High-severity (CVSS 8.6) unauthenticated remote denial-of-service vulnerability is caused by logic errors when secondary authentication is configured on a tunnel group. Network teams must check Cisco's advisory for affected versions and immediately apply patches or apply network segmentation controls restricting VPN service access if patching is delayed.

The OpenSSL backend silently accepts revoked client certificates despite configured revocation lists in GoAhead builds 6.1.0 and earlier; revocation list is loaded but never enforced. To be exploited, the build must use OpenSSL backend with client certificates required, trusted CA bundle and issuer checking enabled, and revocation list configured, with attacker holding a certificate signed by trusted authority and listed as revoked. Upgrade to 6.1.1, or enable issuer checking with trusted CA bundle, or use OpenSSL or LibreSSL 4.x. This is particularly critical for infrastructure using mTLS for device authentication, API gateways, and zero-trust implementations where certificate revocation is supposed to block compromised or departed client identities. Silent acceptance of revoked certificates means an attacker holding revoked keys could authenticate indefinitely until the next revocation list refresh.

In this CIO.com DEMO episode, Cisco's James Parnham and Priya Ram showcase AgenticOps capabilities embedded in Cisco Cloud Control. The demonstration covers the core agentic workflow: identify network issues, perform root-cause diagnosis, generate remediation recommendations, and execute fixes. A key technical feature is Deep Reasoning mode, which allows the AI agent to analyze network configurations, perform security audits, and generate detailed reports with explicit reasoning steps visible to operators. The system maintains human oversight throughout—operators set direction and guardrails while the agent executes within bounded autonomy. This addresses a critical operational requirement: enterprises want agent-driven automation but need visibility into agent reasoning and veto capability before critical changes deploy. The demo illustrates that AgenticOps is operationally distinct from earlier AIOps systems, which only alerted operators; this agent actively troubleshoots and remediates in minutes instead of hours.

Telus and Bell Canada are making strategic infrastructure moves to position themselves as AI compute providers beyond traditional telco roles. Telus's CA$70 billion commitment spans network expansion (5G, fiber, rural connectivity), legacy copper retirement, and sovereign AI compute capacity. The company's sovereign AI facilities—Rimouski now operating at full capacity, Kamloops coming online later in 2026—leverage Telus's existing PureFibre backbone and use NVIDIA GPUs with 99% renewable energy. This positions the network operator's fiber footprint as the connectivity layer for AI infrastructure. Bell Canada's competing 'Bell AI Fabric' strategy targets six AI data centers across British Columbia with a 300 MW facility planned near Regina for 2027. Bell aims to connect those compute clusters to enterprise customers via its national fiber backbone. The emerging dynamic: both telcos are bundling network infrastructure (fiber, backbone capacity, power) with AI compute services, treating data center connectivity as a core product line rather than commodity connectivity. This reflects the shift from telcos as pure connectivity providers to infrastructure platforms for AI workloads.

Verizon's CTO Yago Tenorio presented at the 6G Forum a critical architectural insight: AI workloads fundamentally reshape uplink requirements. Current networks assume asymmetric downlink-heavy traffic (roughly 9:1 downlink to uplink), reflecting consumer download patterns. AI agents operating across networks, robotics, and autonomous systems reverse this model. Tenorio frames it as a transition from "transactional" AI (prompt → answer) to "context-aware" AI requiring continuous sensor feeds, video streams, and high-resolution data. This demands symmetric or uplink-heavy capacity—potentially reaching 1:1 ratios. The architectural implications are material: network design must account for agents consuming arbitrary amounts of video, sensor data, and telemetry feeds depending on use case. Verizon references its work on Integrated Sensing and Communication (ISAC) and AI wearables as bottom-up reference architectures. The message to infrastructure planners: 6G cannot be an incremental 5G upgrade focused on peak rate or latency alone. It must be designed AI-native from the start, with Open RAN's standardized, deterministic interfaces as the foundation. This shapes everything from spectrum allocation to fiber backhaul capacity planning.

Mistral Large 4 is a 1-trillion-parameter multimodal model trained on 3,800 NVIDIA Grace Blackwell GPUs in Europe, with 49 billion activated parameters and 52 billion including embedding and output layers. Mistral states this released version was optimized following real-world testing with government and enterprise users. Available under a modified MIT license supporting personal, academic, and commercial use. Large 4 ranks in the top five on the Artificial Analysis Cyber Index and leads open-weight models built outside China; on vulnerability reproduction and patching tasks, Large 4 scored 82%. The weights are available on Hugging Face and via Mistral's API. For practitioners, this represents a milestone: a 1T-parameter model capable of defensive cybersecurity work is now self-hostable, enabling organizations with EU data residency or airgap requirements to avoid cloud-dependent alternatives.

On October 29, 2026, the European Cybersecurity Competence Centre announced three Horizon Europe projects worth approximately EUR 42 million for generative AI cybersecurity applications. The cumulative regulatory activity signals accelerated attention to agentic AI risks. Poland's statutory framework entering force October 28 moves a major EU economy from guidance to binding law. Norway's NSM elevation of cybersecurity risk to high status reflects intelligence agency assessment that autonomous AI agents and language models now pose material national-scale threats. New York State Bill S9051B, signed the same day, introduces restrictions on unsafe chatbot features for minors with attorney general enforcement authority. Spain's National Cryptologic Centre published evaluation methodology and technical templates for AI product security assessment. This cluster—statute, threat elevation, and sectoral guidance—indicates governments moving from framework-setting to enforcement readiness. Organizations should monitor Poland's implementation as precursor for EU statutory AI rule operationalization, and track Norway's threat-level justification to understand what agentic AI capabilities prompted escalation.