Today's 3 things that matter

1. Cloudflare Containers leaked residual disk blocks between tenants
   Exposes multi-tenant isolation risks in serverless platforms where storage optimizations (skip_block_zeroing) trade confidentiality for performance—critical for operators managing shared infrastructure and setting block-wiping policies.
2. Operators redirect capex from broadband to AI interconnect
   Telcos are repositioning from connectivity providers to AI infrastructure operators; network teams must now design for bi-directional, high-capacity traffic and multi-tenant data center interconnection rather than traditional consumer broadband patterns.
3. Komodor ships agentic operations platform with 50 SRE agents
   Addresses the governance gap: 40% of agentic AI initiatives are decommissioned by 2027 due to governance failures; Komodor's platform unifies pre-built and custom agents under common controls to reduce operational chaos from agent sprawl.

Full stories below, grouped by topic.

AIOps & Network Observability

Datadog expands AI observability product set

Kalkineme Media · Sep 28, 2026 · Industry news

What happened: Datadog announced expansion of its AI observability product set, with growing adoption among large enterprises and AI-native organizations. Coverage cited acceleration in enterprise demand for AI-powered observability capabilities integrated with agentic operations.

Why it matters: Indicates market validation that AIOps platforms are shifting from reactive monitoring to agentic automation; relevant for NOC teams evaluating platforms that can autonomously investigate and remediate network-related incidents.

Datadog's expansion of AI observability products reflects a market-wide shift toward agentic operations where AI systems autonomously investigate incidents, propose fixes, and execute remediation. The company's Bits AI SRE agent, announced at DASH 2026, operates as an autonomous on-call responder that triggers on alerts, accesses the same telemetry a human would (metrics, logs, traces, dashboards, change events), and delivers a root-cause hypothesis within 3–4 minutes. This capability is particularly relevant for NOCs managing large distributed systems where human response latency directly translates to customer impact. Datadog's LLM Observability product, now renamed Agent Observability, adds specialized tracking for AI workloads—inputs, outputs, token usage, latency, and hallucination detection—a capability that becomes critical as organizations deploy autonomous agents in production. The speed of execution (from announcement to GA to production-hardened in twelve months) indicates the feature has moved beyond POC into enterprise deployments.

Read the original at kalkinemedia.com

Routing & Internet

Cloudflare Containers leaked residual disk blocks between tenants

Cloudflare Blog · Sep 24, 2026 · Primary source

What happened: A cross-tenant data exposure flaw in Cloudflare Containers allowed one customer to read residual disk blocks (up to 60 KiB per block) from prior tenants' containers due to thin-provisioned disks returning blocks without zeroing via Linux dm-thin skip_block_zeroing. Reported Sept 4, remediated fleet-wide by Sept 19.

Why it matters: Exposes multi-tenant isolation risks in serverless platforms where storage optimizations (skip_block_zeroing) trade confidentiality for performance—critical for operators managing shared infrastructure and setting block-wiping policies.

Cloudflare Containers used thin-provisioned Linux dm-thin disks to optimize storage cost and I/O. When containers exited, deallocated 64 KiB blocks were returned to a shared free pool without being zeroed. New containers could allocate these recycled blocks; writing only 4 KiB left the remaining 60 KiB unmodified, exposing plaintext directory structures, database pages, and SQLite data from prior tenants. Researchers from Accomplish demonstrated recovery of cross-tenant data on 18 of 24 placements by writing 4 KiB then reading the full block via /dev/vdc. The vulnerability affected Containers, Sandboxes (built atop Containers), and Browser Rendering. Exploitation required a paid Workers account; attackers could not target specific tenants. Timeline: vulnerability reported Sept 4 via bug bounty, initial mitigation Sept 7, full fleet remediation with block-zeroing re-enabled and cache/disk clearing completed Sept 19, public disclosure Sept 24. Historical disk telemetry showed no unauthorized exploitation. No CVE was assigned.

Read the original at blog.cloudflare.com

Agentic AI & MCP

Komodor ships agentic operations platform with 50 SRE agents

Komodor · Sep 30, 2026 · Vendor release

What happened: Komodor released an Agentic Operations Platform for SRE and DevOps teams, combining 50+ pre-built specialist agents across AI SRE, software operations, and cost optimization with an extensible backbone for custom agents and centralized governance.

Why it matters: Addresses the governance gap: 40% of agentic AI initiatives are decommissioned by 2027 due to governance failures; Komodor's platform unifies pre-built and custom agents under common controls to reduce operational chaos from agent sprawl.

Komodor's platform targets a critical pain point: SRE and platform teams deploying agents rapidly but struggling to govern them at scale. The product ships with 50+ pre-built specialist agents covering troubleshooting, incident management, alert intelligence, reliability work, cloud cost reduction, observability cost, Kubernetes cost, change intelligence, CI/CD health, and production readiness. Teams can customize these workflows by adjusting steps, changing routing, and inserting their own agents. Custom agents can be created from scripts, runbooks, or existing skills. This addresses the real constraint: not building agents, but operating them safely. Gartner forecasts that 40% of agentic AI initiatives will be decommissioned by 2027 due to governance gaps, unclear returns, or rising costs—a staggering failure rate. Komodor responds by providing centralized control and visibility across all agents. The platform reflects organizational reality: 60% of senior enterprise leaders already deploy agents in production, often ad hoc. Without unified governance, teams end up with fragmented agent implementations, overlapping responsibilities, and no common audit trail. Pre-built workflows accelerate time-to-value while the governance backbone prevents runaway agent systems. This matters especially for SRE teams already drowning in on-call load.

Read the original at itbrief.asia

Telco & Cable AI

AT&T commits $3 billion to Corning fiber supply

AT&T Newsroom · Sep 29, 2026 · Primary source

What happened: AT&T and Corning entered into a multi-year agreement valued at more than $3 billion for fiber and cable supply to expand AT&T's network as AI drives growing data demand. AT&T fiber homes currently consume over 1 terabyte per month—five times 2016 levels—with usage projected to reach 2-2.5 TB monthly by 2030.

Why it matters: The deal secures long-term fiber supply for AT&T's target of 60 million American locations by 2030, addressing supply chain constraints from AI data center buildout competition.

On September 29, 2026, AT&T and Corning announced a multi-year agreement valued at more than $3 billion for fiber and cable supply. Growing demand for data-intensive services—including AI workloads, cloud computing, and connected devices—is increasing the need for faster, more resilient fiber networks. Fiber is critical because it carries equal upstream and downstream capacity, unlike legacy copper or wireless. The deal supports AT&T's target of reaching 60 million Americans with fiber by 2030. This commitment reflects broader competitive pressure: Verizon, Zayo, Lumen, Meta, and Nvidia have all secured major commitments from Corning's manufacturing capacity. Corning will pair U.S.-based advanced manufacturing with AT&T's network leadership to build dense fiber networks while sustaining thousands of union-represented technician jobs. For NetOps practitioners, this signals sustained capex pressure on fiber deployments driven by AI compute infrastructure demand, with long-term supply contracts now central to regional network expansion strategies.

Read the original at about.att.com

Bell Canada and Cisco to build sovereign AI infrastructure

Fierce Network · Sep 29, 2026 · Industry news

What happened: Bell Canada and Cisco signed a memorandum of understanding to collaborate on sovereign AI infrastructure for Canada, combining Bell's data center and network assets with Cisco's AI and security technology to keep sensitive AI workloads secure and data-resident within Canada.

Why it matters: Canadian operators are prioritizing data sovereignty as US hyperscaler relationships evolve; this partnership positions Bell and Cisco to compete in the emerging sovereign compute market.

Bell Canada and Cisco announced a strategic MOU on September 29, 2026, to build sovereign AI infrastructure for Canadian enterprises and government organizations. The partnership addresses regulatory and business pressure around data residency in Canada, combining Bell's existing network assets and operational capabilities with Cisco's AI and security technology stack. This follows Telus's announcement of a sovereign AI factory in Rimouski, Quebec (September 2025) powered by Nvidia GPUs, establishing a competitive landscape among Canadian telcos for AI compute localization. For infrastructure practitioners, this reflects telcos positioning data residency and sovereign compute as service differentiation—moving beyond pure connectivity into managed AI infrastructure. The timing suggests regulatory tailwinds in Canada around data protection and AI governance, particularly as US hyperscalers face increased scrutiny.

Read the original at fierce-network.com

Operators redirect capex from broadband to AI interconnect

TelecomLead · Sep 25, 2026 · Industry news

What happened: AT&T, Verizon, SK Telecom, and Lumen are securing massive fiber volumes and data center capacity as network architecture increasingly connects AI data centers to long-haul fiber, metro networks, and edge computing. The shift reflects a fundamental change from consumer-focused broadband capex to AI compute interconnect infrastructure.

Why it matters: Telcos are repositioning from connectivity providers to AI infrastructure operators; network teams must now design for bi-directional, high-capacity traffic and multi-tenant data center interconnection rather than traditional consumer broadband patterns.

Operators worldwide are shifting capex allocation from spectrum and consumer broadband to fiber, data centers, and AI-optimized network architecture. AT&T, following its Lumen fiber acquisition (closed February 2, 2026), is building interconnect routes between AI data centers. Lumen deployed 17 million intercity fiber miles by end-2025 and targets 47 million miles by end-2028, arguing that real value in the AI era comes from turning fiber into programmable, automated enterprise-grade services via SD-WAN and dark/lit wavelength offerings on long-haul routes. SK Telecom targets 15 GW of AI data-center capacity; Airtel is expanding to 1 GW; SoftBank is combining GPU clouds with AI-RAN. The emerging architecture prioritizes low-latency, high-capacity connections between compute clusters, requiring new optical density standards, programmable wavelength services, and QoS guarantees distinct from traditional ISP operations. For network architects, this consolidation and infrastructure race signals that operators now view fiber as foundational AI compute interconnect—not consumer broadband backbone.

Read the original at telecomlead.com

AI Model Providers

Anthropic releases faster Claude Sonnet 5.5 at same price

Codersera · Sep 28, 2026 · Primary source

What happened: Anthropic released Claude Sonnet 5.5 on September 28, 2026 as its mid-tier model in the Claude 5.5 family, six days after Claude Opus 5.5. Positioned as a faster, lower-cost complement to Opus at $2/$10 per million tokens with 1M context, it scores near Opus 5.5 on most benchmarks at half the price.

Why it matters: Sonnet 5.5 delivers near-Opus performance on coding and knowledge work at half the token cost with 30%+ faster throughput, making it the cost-efficient default for agentic and retrieval workloads.

Claude Sonnet 5.5 shipped September 28, 2026 as the second model in the Claude 5.5 family, following Opus 5.5 on September 22. The model carries a 1M-token context window and 128K max output, with pricing fixed at $2 input/$10 output per million tokens—unchanged from Sonnet 5. Terminal-Bench 4.0 performance jumped from Sonnet 5's 10.3% to 70.6%, and on agentic and office-work evaluations Sonnet 5.5 now approaches Opus 5.5 performance levels. Speed increased over 30% versus Sonnet 5 while maintaining the same per-token rate. On Anthropic's benchmarks it scores within about two points of Opus 5.5 on professional knowledge work and coding at half the token price ($2/$10 vs $4/$20 per million tokens) and 30%+ faster output. For practitioners building agents or knowledge-work applications, Sonnet 5.5 closes the capability gap to Opus 5.5 at the same unit cost, making it the default choice for cost-sensitive workloads; Opus 5.5 remains preferable only for complex, open-ended reasoning requiring sustained judgment.

Read the original at codersera.com

AI Industry & Policy

White House signs voluntary super intelligence safety accord

Business Standard · Sep 30, 2026 · Industry news

What happened: The White House released a voluntary 'White House Accord on Super Intelligence' framework asking companies developing advanced AI models to strengthen safety checks with internal monitoring, external evaluations, and board-level oversight.

Why it matters: Voluntary self-regulatory model creates friction with EU and state-level jurisdictions expecting documented controls and formal governance, forcing enterprises to maintain dual compliance postures.

President Trump signed a 'morally binding' AI document with tech leaders following a White House luncheon, stating he is 'seeing tremendous self-policing' and that the administration is considering a 10-person committee to oversee the AI industry. The voluntary framework sets out four layers of AI safeguards: internal controls, external evaluations, board oversight, and independent audits. This represents the U.S. federal approach contrasting sharply with regulatory enforcement tracks taken by the EU and state-level jurisdictions like California and Colorado, which expect documented AI inventories, risk classifications, third-party due diligence, and model lifecycle controls. The 'morally binding' language and voluntary committee structure indicate continued resistance to mandatory pre-market approval or formal disclosure regimes, creating compliance complexity for enterprises operating across multiple jurisdictions with divergent governance expectations.

Read the original at business-standard.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.