You're reading the Thursday, November 12, 2026 edition. Today's briefing →
Toronto
Thursday, November 12, 2026No. 151

Digital Plumber

Plumbing the information age

AI-curated intelligence for people who run networks. Daily coverage of AIOps, network automation, agentic operations, AI infrastructure, security and the vendors shaping them.

Today's 3 things that matter

Picked by the AI editor
  1. Routing·Industry news

    BGP hijack redirects Chinese traffic to gambling sites

    Malicious actors exploited BGP to announce false routes through AS 402425, redirecting traffic from multiple Chinese provinces to unauthorized gambling sites using longest-prefix-match manipulation and DNS tampering without device infection.

    Why it matters Demonstrates that route hijacking remains a practical threat despite RPKI deployment; operators must filter suspicious ASNs and validate BGP announcements at the origin level.

  2. Startups·Industry news

    Bay leaves stealth with $50M for AI agent policing

    Tel Aviv-based Bay emerged from stealth with $50 million funding ($37M Series A led by Battery Ventures, $13M seed) to provide runtime policy enforcement for AI agents.

    Why it matters Enterprises deploying AI agents need runtime policy enforcement; Bay's agent discovery and session-level control addresses infrastructure security and governance gaps operations teams must manage.

  3. DC networking·Industry news

    Arista splits AI fabrics between ESUN and UEC Ethernet

    Arista announced dual Ethernet standards for scale-up and scale-out: ESUN for scale-up (supporting up to 144 accelerators per rack, 1,024 cross-rack) and UEC-based Ethernet for scale-out.

    Why it matters Scale-up networking is now treated as a first-class citizen with full-featured network stack parity, enabling tighter GPU/XPU clustering and heterogeneous acceleration without proprietary lock-in—crucial for operators building multi-vendor AI fabrics.

Today's briefing

What happened, and why it matters

20 stories · 12 topics · Updated 1:12 PM ET

Zabbix 7.0 LTS monitors multi-site Cisco IOS-XE lab

pinglabz.com · Nov 8, 2026 · Analysis

Zabbix 7.0.32 LTS deployed against Cisco IOS-XE 17.18.2 in a multi-site home lab with SNMPv3 collection, interface discovery, loopback polling, and timed HTTP outage testing. Practical configuration using official Cisco IOS SNMP templates plus HTTP, SMTP, and NTP simple checks.

Why it matters Demonstrates real Zabbix deployment patterns for network observability in multi-site infrastructure; relevant context for teams evaluating OSS monitoring integration with network automation.

The article describes Zabbix 7.0.32 LTS configured in a home lab on November 7, 2026, built against Cisco IOS-XE 17.18.2, using the official Cisco IOS by SNMP template plus ICMP ping and HTTP, SMTP, and NTP simple checks. Interface discovery successfully found ifHCInOctets and operational-status items on all 15 Cisco devices after name-filter correction; Zabbix trigger for an HTTP outage was raised 1 second after the actual failure, showing sub-second detection latency. Zabbix 8.0.0 LTS shipped on November 5, 2026. The setup represents practitioner-level monitoring configuration relevant to NetDevOps pipelines that need network observability without commercial licensing.

Read the original at pinglabz.com ↗

Red Hat network observability 1.12 adds TLS tracking

Red Hat · Nov 5, 2026 · Primary source

Red Hat OpenShift Network Observability 1.12 adds TLS tracking, Kafka compression, FLP shared cache, and automatic secondary network indexing for eBPF-based cluster monitoring. Reduces CPU/memory overhead and simplifies multi-network pod enrichment without manual configuration.

Why it matters Reduces observability collector overhead in Kubernetes environments; enables zero-touch secondary network indexing for hybrid and multi-tenant clusters without manual configuration updates.

Red Hat released OpenShift Network Observability 1.12 with three operational improvements for production Kubernetes clusters. Security enhancement: TLS tracking between eBPF agents and Flow Log Pipeline (FLP) processors prevents unencrypted telemetry transit. Performance: Kafka compression reduces message overhead in high-volume environments; FLP shared cache component prevents kube-apiserver slowdown when network observability starts (addresses common scaling pain point in large clusters). Ease of use: Automatic secondary network indexing eliminates manual configuration of network names in FLP config. Prior versions required operators to maintain a list of secondary network interfaces; 1.12 auto-discovers and enriches all pod flows across Multus networks without explicit specification. For network operators managing OpenShift at scale, this reduces operational toil and human error in multi-network pod deployments. Tradeoff: Shared cache requires careful memory tuning in extremely large clusters; TLS overhead adds latency but improves security posture for air-gapped deployments.

Read the original at developers.redhat.com ↗

ServiceNow adds AIOps AI Specialist for alert correlation

ServiceNow Community Blog · Nov 6, 2026 · Primary source

ServiceNow released AIOps AI Specialist in November 2026 release to correlate and triage multi-source alerts without manual sorting. Reduces alert volume climbing from siloed service, tool, and integration sprawl.

Why it matters Addresses alert fatigue in NOC workflows by automating alert grouping and correlation across monitoring, observability, and integration signals at platform level.

ServiceNow announced general availability of the AIOps AI Specialist as part of its November 2026 platform release. The feature targets a core NOC pain point: as alert volume climbs with each new service, monitoring tool, and integration, operations teams manually sort through increasingly noisy signal streams. The AI Specialist automates correlation of alerts across heterogeneous sources—monitoring platforms, observability tools, and third-party integrations—without operator intervention. This shifts the NOC workflow from reactive triage (engineer opens 50 alerts, manually groups related ones) to proactive grouping (AI clusters and deduplicates before paging). ServiceNow positions this as a platform-native layer preventing vendors from requiring separate AIOps middleware. For NOC teams already invested in ServiceNow ITSM, this eliminates tool sprawl for alert correlation. Limitation: The feature is architecture-specific to ServiceNow's incident management data model; enterprises with non-ServiceNow ITSM backends or those using best-of-breed observability (Datadog, Dynatrace) still need middleware for cross-platform correlation.

Read the original at servicenow.com ↗

Arista splits AI fabrics between ESUN and UEC Ethernet

Fierce Network / NextPlatform · Nov 5, 2026 · Industry news

Arista announced dual Ethernet standards for scale-up and scale-out: ESUN for scale-up (supporting up to 144 accelerators per rack, 1,024 cross-rack) and UEC-based Ethernet for scale-out. Arista has surpassed 100 Etherlink customers and forecasts providing networking for 10 gigawatts of AI cluster capacity by end of 2027, linking 10+ accelerator types.

Why it matters Scale-up networking is now treated as a first-class citizen with full-featured network stack parity, enabling tighter GPU/XPU clustering and heterogeneous acceleration without proprietary lock-in—crucial for operators building multi-vendor AI fabrics.

Arista's Q3 2026 earnings call (November 4) revealed two major technical moves. The Etherlink SU-144 architecture supports up to 144 accelerators per rack and enables scaling to 1,000 XPUs through cross-rack topologies. This addresses scale-up (tightly coupled, within-rack) networking where Arista previously had limited play. For scale-out, Arista uses UEC standards with Multipath Reliable Connection (MRC) fabric resiliency, multi-plane routing, and intelligent Dynamic Load Balancing (DLB) and Cluster Load Balancing (CLB). The architectures are co-developed with AMD, Arm, Broadcom, Meta, Microsoft, and Qualcomm. The rack design features multiple liquid-cooled Arista 7060EX7 switches with 102.4 Tb/s capacity per switch, fiber patch panels, integrated liquid-cooling manifolds, and drip trays with active leak detection. Arista's $3.66B Q3 revenue (up 58.5% YoY) shows strong adoption with Etherlink customer base surpassing 100 customers in 2026. CEO Jayshree Ullal noted scale-up trials will start in 2027 with no material revenue expected that year, but this is the first time a major switch vendor has committed rack-scale designs across both scale-up and scale-out in one cohesive strategy.

Read the original at fierce-network.com ↗

BGP hijack redirects Chinese traffic to gambling sites

Tribune Online · Nov 11, 2026 · Industry news

Malicious actors exploited BGP to announce false routes through AS 402425, redirecting traffic from multiple Chinese provinces to unauthorized gambling sites using longest-prefix-match manipulation and DNS tampering without device infection.

Why it matters Demonstrates that route hijacking remains a practical threat despite RPKI deployment; operators must filter suspicious ASNs and validate BGP announcements at the origin level.

On November 11, 2026, China's CNCERT reported a sophisticated BGP hijacking attack where malicious actors used foreign autonomous system AS 402425 (registered in April 2026 by US-based Qorexa Labs) to announce false routes targeting multiple Chinese provinces. The attack leveraged the longest-prefix-match rule to override legitimate routes, redirecting traffic to unauthorized gambling destinations or malicious DNS servers without infecting end-user devices. The hijacked route flapped continuously throughout the approximately 22-hour incident, creating intermittent diversions. Route measurements showed roughly 72% of networks seeing the diverted traffic at peak moments, making it a broadly visible (though intermittent) hijack. CNCERT urged international registries to review and revoke malicious AS resources and recommended operators filter routes containing AS 402425. The incident underscores that BGP origin validation alone (RPKI) does not prevent path manipulation—ASPA (Autonomous System Provider Authorization) deployment is needed to validate AS relationships in the complete path.

Read the original at tribuna.com ↗

eMudhra revokes eight more CAs after key exposure

The Hacker News · Nov 11, 2026 · Industry news

eMudhra revoked 8 additional certificate authorities after discovering a Python script exposed RSA private keys to process memory during HSM transfer. Six were for TLS, two for code signing, with 195 total certificates issued including 29 from eMudhra's own domains.

Why it matters Network teams relying on eMudhra certificates must urgently audit certificate chains before browser root programs remove distrusted roots; applications using WebPKI CAs for mTLS face forged-certificate authentication bypass risks.

eMudhra discovered its key-wrapping ceremony script left RSA private key material—private exponents and prime factors—resident in process memory on the ceremony computer. The company had placed four of the affected intermediate CAs into production in May 2026. This compounds an October discovery that eight eMudhra intermediates had cryptographically weak keys generated by a broken random-number mechanism. Microsoft is removing eMudhra root certificates from all products starting November 16, 2026; Google Chrome and Mozilla Firefox have set SCTNotAfter removal to the same date. Network teams must identify all services using eMudhra-issued certificates and plan replacement before browsers distrust them. Applications using WebPKI roots in mutual TLS scenarios now face direct authentication bypass: attackers holding either of the two compromised code-signing CAs can forge certificates to authenticate to mTLS endpoints that trust the web PKI.

Read the original at thehackernews.com ↗

Barracuda email gateway bug exploited for root access

The Decryption Digest · Nov 10, 2026 · Industry news

CVE-2026-116606 (CVSS 9.3 Critical) is a pre-authentication authorization bypass in Barracuda Email Security Gateway that allows unauthenticated remote attackers to restore crafted configuration backups. When chained with CVE-2026-111560 (CVSS 8.6), attackers gain unauthenticated root access. Barracuda confirmed active exploitation on November 9, 2026.

Why it matters Organizations with internet-facing Barracuda ESG appliances face immediate unauthenticated root compromise; patching to 9.4.0.030+ is mandatory under BOD 26-04 forensic triage requirements; CISA deadline is November 13 for federal agencies.

CVE-2026-116606 exposes a configuration import function through an alternate path that bypasses authentication, allowing an unauthenticated attacker to restore a malicious ESG configuration. CVE-2026-111560 is an OS command injection flaw in the timezone field that reaches root privileges when chained. Barracuda confirmed exploitation in the wild as of November 9. The fix is a single upgrade to ESG version 9.4.0.030 or later, which addresses both flaws. Any internet-exposed ESG appliance can be fully compromised by an unauthenticated attacker. Network teams must treat this as an immediate priority: check Barracuda support documentation for in-appliance forensic indicators of prior exploitation and apply patches emergently. Conduct 30-day retrospective search for unexpected admin UI posts from non-management sources.

Read the original at decryptiondigest.com ↗

SonicWall SSLVPN MFA bypass exploited with no workaround

Help Net Security · Nov 9, 2026 · Industry news

CVE-2026-19868 (CVSS 7.1 High) in SonicWall SonicOS SSLVPN allows users with valid credentials for an alternate authentication method to bypass SAML-enforced MFA. CISA added it to the Known Exploited Vulnerabilities catalog on November 6, confirming active exploitation in the wild. No workarounds exist.

Why it matters Organizations running SonicWall SSLVPN with SAML-enforced MFA must patch by November 20 or face active MFA bypass exploitation; attackers holding any valid alternate authentication credential can skip SAML controls entirely.

CVE-2026-19868 is an authentication enforcement flaw (CWE-308: Use of Single-Factor Authentication) in SonicOS SSLVPN that only manifests when SAML is enabled for SSLVPN logins. An attacker with valid credentials for any non-SAML authentication method can bypass SAML controls and any MFA enforced through SAML. Affected versions: Gen7 firewalls running SonicOS 7.3.3-7015 or earlier; Gen8 firewalls running 8.2.2-8015 or earlier. Patches available immediately in SonicOS 7.3.4-7013+ (Gen7) and 8.2.3-8015+ (Gen8). CISA added the CVE to KEV on November 6 based on evidence of active exploitation, setting the federal remediation deadline for November 20. Private organizations should treat this as equally urgent. Conduct immediate audit of all user accounts with active SSLVPN sessions and alternate authentication credentials; cross-reference with MFA logs to identify potential compromise activity.

Read the original at helpnetsecurity.com ↗

Bay leaves stealth with $50M for AI agent policing

SecurityWeek · Nov 11, 2026 · Industry news

Tel Aviv-based Bay emerged from stealth with $50 million funding ($37M Series A led by Battery Ventures, $13M seed) to provide runtime policy enforcement for AI agents. The platform automatically discovers agents and enforces policies by evaluating proposed actions in milliseconds to allow, deny, or redirect them.

Why it matters Enterprises deploying AI agents need runtime policy enforcement; Bay's agent discovery and session-level control addresses infrastructure security and governance gaps operations teams must manage.

Bay was founded in February 2026 by three ex-IDF Unit 8200 officers: Yuval Zacharia (CEO), Boris Danilovich (VP of R&D), and Aner Morag (CPO). When installed, the platform automatically builds a full inventory of AI agents, coding assistants, MCP servers, skills and extensions in use, then follows each agent session from first prompt to final action. Its Enforcer component evaluates every proposed action in the context of the full session and decides in milliseconds whether to allow, refer, redirect or deny it. Bay learns from how a company's agents behave, so security teams do not have to write rules for every new agent or tool. The startup has 35 employees, mostly in Israel, and plans to spend funding on product, engineering, and sales/marketing teams in the US and Europe. For ops teams, uncontrolled agent deployment creates compliance and incident-response challenges that traditional security and observability tools don't address.

Read the original at securityweek.com ↗

Vanta acquires Pillar Security for AI runtime protection

Unite.AI · Nov 9, 2026 · Industry news

Vanta signed a definitive agreement to acquire AI security startup Pillar Security, integrating its Secure AI Lifecycle (SAIL) framework into Vanta's trust and compliance platform. Pillar's framework has 50,000+ downloads and maps 91 AI risks across seven lifecycle phases.

Why it matters Integration of AI agent testing, monitoring, and runtime protection into governance platforms extends compliance and audit coverage to autonomous systems—operations teams now treat agent behavior as an auditable infrastructure component.

The planned acquisition connects AI agent testing, monitoring, and runtime protection with Vanta's governance platform, specifically its Trust Graph, which links business policies, security controls, risk ownership, and compliance evidence. Pillar's SAIL framework gives development, ML operations, security, and governance teams a common approach to AI risk from initial development through live operation, covering seven lifecycle phases. This builds on Vanta's AI governance offering introduced in July 2026. Vanta has grown to over 20,000 customers, up from 16,000 in Q1 2026. For infrastructure and SRE teams, this signals that compliance frameworks and audit trails are expanding to cover agent behavior—treating autonomous systems as production components that must be monitored, logged, and governed alongside traditional infrastructure. The acquisition consolidates testing (pre-deploy), monitoring (runtime), and compliance-reporting capabilities in a single platform.

Read the original at unite.ai ↗

People Moves

2 stories

Jensen Huang resigns as Nvidia CEO, Kress takes over

Nvidia Newsroom · Nov 12, 2026 · Primary source

Nvidia's board announced that Jensen Huang resigned as President, CEO, and board member effective immediately. Colette Kress, Executive Vice President and Chief Financial Officer, was appointed interim CEO, and Harvey C. Jones was appointed the company's first board Chairman.

Why it matters Leadership change at the world's largest AI infrastructure company reshapes executive oversight of chip roadmaps, strategic partnerships with hyperscalers and telcos, and governance of critical national security and export controls around advanced AI compute.

Nvidia's board announced on November 12, 2026, that Jensen Huang resigned as President, Chief Executive Officer, and board member, effective immediately. Colette Kress, current Executive Vice President and Chief Financial Officer since September 2013, was appointed interim CEO and will continue in her CFO role until a permanent successor is named. The board initiated a search for a permanent CEO considering internal and external candidates. Harvey C. Jones, a board member since Nvidia's founding in 1993 and former Synopsys CEO, was appointed as the company's first board Chairman. Kress has led Nvidia's finance organization and played a key role in the company's growth and evolution. Huang had led Nvidia as CEO for 33 years since co-founding it; Nvidia shares fell as much as 13% after news of the resignation before rebounding somewhat.

Read the original at nvidianews.nvidia.com ↗

Meta infrastructure chief steps down, Dave Brown replaces him

Bloomberg / TechCrunch · Nov 11, 2026 · Industry news

Santosh Janardhan, Meta's head of infrastructure and co-head of engineering, will step down from leading the infrastructure organization. Dave Brown, who joined Meta as vice president of infrastructure earlier in 2026 and spent 19 years at Amazon on compute, cloud, and AI, will replace him.

Why it matters Major leadership transition in Meta's infrastructure organization directly impacts engineering decisions on AI data center architecture, compute platform investments, and operational strategy for supporting Meta's multi-hundred-billion-dollar AI infrastructure buildout.

Santosh Janardhan, Meta's longtime head of infrastructure and co-head of engineering who joined the company in 2009, is stepping down from his role. Dave Brown, who joined Meta as vice president of infrastructure earlier in 2026, will replace him; Brown previously spent 19 years at Amazon focused on compute, cloud, and AI. Janardhan is expected to leave in early 2027 and will remain at Meta for transition, moving into an advisory role on Meta Compute. The departure represents a major leadership change as Meta continues to invest hundreds of billions of dollars in AI and the data centers needed to support it. Janardhan's memo cited the need to pass the baton but did not detail specific reasons for his departure.

Read the original at bloomberg.com ↗

Red Hat extends Kubernetes and llm-d for AI agents

SiliconANGLE · Nov 10, 2026 · Industry news

Red Hat extends Kubernetes infrastructure to handle enterprise-scale AI agents with security boundaries and inference requirements. The company is advancing Kubernetes-native security for agentic AI and extending llm-d to address inference infrastructure demands specific to agent workloads.

Why it matters Kubernetes operators and platform teams need to understand how agents change resource management, security models, and inference infrastructure sizing for on-prem and hybrid deployments.

Red Hat's approach addresses three distinct operational challenges agents create: security boundaries between agent contexts, inference infrastructure that differs from traditional container workloads, and the need for enforceable access controls as agents move into production. The company's announcement signals a shift in how Kubernetes is being extended—not just as a compute platform for agents, but as security and observability infrastructure. Red Hat's contributions include work on OpenShift for agent management, llm-d for inference optimization, and OpenShell for agent orchestration. This matters because enterprise teams are moving agents from experimental to production workloads but lack established patterns for running agentic workloads securely alongside traditional applications. The Kubernetes-native approach allows ops teams to apply existing skills (RBAC, networking policies, resource quotas) to agent lifecycle management. One key caveat: these are early-stage extensions to Kubernetes, so teams should expect API and operational patterns to evolve significantly as the ecosystem stabilizes.

Read the original at siliconangle.com ↗

AWS ships local emulator written entirely by AI agents

Rio Rundown · Nov 12, 2026 · Industry news

AWS released AWS Emulator, a free local tool covering 200+ services, entirely written by autonomous AI agents running on Amazon Bedrock AgentCore without human code review. Agents wrote, tested, and fixed all emulator code.

Why it matters Operators and SRE teams can now test infrastructure code and agent behaviors offline without cloud costs; this signals agent-generated production code is reaching maturity in AWS infrastructure tooling.

AWS's claim that autonomous agents wrote production-grade software without human review marks a significant milestone in agent-based development. The AWS Emulator covers 200+ AWS services, allowing developers and operators to test applications and infrastructure changes locally before deployment, reducing iteration time and cloud costs. This is particularly valuable for SRE teams testing disaster recovery scenarios, compliance changes, or agent-based workflows without incurring network latency or API costs. The agents used Amazon Bedrock AgentCore to iteratively write and validate code, suggesting a multi-turn reasoning and testing loop. Caveat: AWS's statement of "no human review" likely means no traditional code review, but the infrastructure and testing guardrails used by the agents themselves constitute a form of review. Operators should treat this as a reference implementation for how autonomous agent code can reach production quality, though teams should validate agent output for their own use cases before running in production. The tool's value for NetOps and infrastructure teams is immediate—agents can now test infrastructure-as-code changes in a sandbox before applying them.

Read the original at riorundown.substack.com ↗

Cisco bundles ThousandEyes agents into Wi-Fi 7 access points

Cisco Blogs · Nov 9, 2026 · Primary source

Cisco includes ThousandEyes agents and synthetic testing with every Wi-Fi 6E and Wi-Fi 7 access point at no extra cost. When powered on, access points automatically provision agents, discover critical services, and intelligently deploy tests.

Why it matters Network ops teams get built-in autonomous testing and troubleshooting for wireless access points at no extra cost, shifting from manual synthetic monitoring to agent-driven continuous discovery and validation.

Cisco's announcement comes as AgenticOps has moved from vision to day-to-day necessity for NetOps. The inclusion of ThousandEyes agents directly in access points addresses a practical problem: traditional wireless networks require manual configuration of monitoring tests, and coverage gaps often go undetected until outages occur. By embedding agents that automatically discover services and deploy intelligent tests, Cisco is reducing the operational overhead of network visibility. The intelligent network telemetry helps pinpoint network problems sooner, and agents automatically provision and intelligently deploy tests the moment you onboard a new access point. For operators managing large campus or branch networks, this reduces the time spent configuring synthetic monitoring and provides continuous validation that new access points are behaving correctly. The integration also signals how vendors are embedding agentic capabilities into the infrastructure itself rather than requiring separate agent management platforms. Teams should note this is currently bundled into Wi-Fi 6E/7 access points; older 802.11ac infrastructure won't have this capability.

Read the original at blogs.cisco.com ↗

GSMA launches AI challenges for RAN and telco standards

TelecomTV · Nov 11, 2026 · Primary source

The GSMA's Open Telco AI initiative announced two competitions—TelcoRetrieve and AI Agents for RAN Challenges—developed with ETSI, IEEE, ITU and Huawei to enable AI agents to solve complex network engineering problems and improve accuracy of AI systems working with 3GPP technical standards.

Why it matters AI agents grounded in 3GPP specs and capable of reasoning across network domains are prerequisites for L4 autonomy; practitioners need common interfaces and vendor-neutral standards for cross-vendor orchestration.

The GSMA's Open Telco AI initiative launched two competitions addressing critical gaps in production AI deployment. TelcoRetrieve uses an open, vendor-neutral retrieval-augmented generation (RAG) pipeline that connects AI-generated answers to 3GPP-specific source documents, solving the problem that standard large language models lack telecom-specific domain knowledge and fail when applied to network standards. AI Agents for RAN Challenges addresses the need for agents to solve complex radio network engineering problems. The competitions target real deployment blockers: operationalizing AI requires grounded reasoning tied to standardized technical documents, not generic language models. A dedicated webinar on November 13 supports innovator engagement. Winning solutions could accelerate operator deployment of trustworthy, interoperable AI agents that work across multiple vendor platforms—a prerequisite for true multi-domain autonomy.

Read the original at telecomtv.com ↗

OpenTelemetry Collector replaces batch processor with queue_batch

OpenTelemetry Project · Nov 7, 2026 · Primary source

OpenTelemetry Collector 0.165.0 released November 7, 2026. Adds queue_batch processor to replace legacy batch processor, ECS receiver for Amazon ECS task metrics, and baseline hardening for systemd services.

Why it matters Queue batching migration path moves away from legacy processor; ECS instrumentation reduces vendor lock-in; systemd hardening improves operational security.

OpenTelemetry Collector 0.165.0 released on 7 November 2026 introduces workflow improvements for telemetry pipelines. New components include queue_batch processor (eventually to replace batch processor), the ECS receiver for Amazon ECS task metrics in the otelcol-prometheus distribution, and baseline hardening for packaged Collector systemd services. The release includes compressed-request memory bounds and protobuf recursion hardening. For infrastructure operators, the queue_batch processor introduces a new migration path away from the legacy batch processor with a documented RFC (batching-migration.md). The ECS receiver enables native telemetry collection from container task metadata without vendor-specific exporters. Systemd hardening addresses operational security posture in containerless deployments.

Read the original at github.com ↗

Anthropic cuts Claude Fable prices and speeds inference

Anthropic official documentation · Nov 9, 2026 · Primary source

Claude Fable 5.2 released November 9, 2026, with improved coding, reasoning, and vision. The model costs 20% less than Fable 5.1, generates output up to 30% faster, and maintains 1M-token context and 128K output maximum.

Why it matters A team of 1,000 agents running Fable 5.2 formalized a proof of Fermat's Last Theorem in 30 hours and built a prototype operating system from scratch—demonstrating significant improvements in agentic reasoning for infrastructure automation and complex DevOps workflows.

Claude Fable 5.2 released November 9, 2026, priced at $8 per million input tokens and $40 output (down from $10/$50 for Fable 5.1), maintaining 1M token context and 128K output maximum. Fable 5.2 has approximately 55 percent fewer refusals than Fable 5.1, with outputs generated up to 30 percent faster. A team of 1,000 Fable 5.2 agents formalized the proof of Fermat's Last Theorem in 30 hours. On the RoboDojo robot-control benchmark, Fable 5.2 scored 64.9%, more than double Fable 5.1's 27.6%. The model coordinated over 1,000 sub-agents for eight days and built a prototype operating system from scratch. Size estimates place active parameters per token at approximately 150B centrepoint, with ~250T training tokens seen. Fable 5.2 has cybersecurity and biology safeguards with approximately 55 percent fewer blocks than Fable 5.1. Default effort is High in the API and Medium in Claude apps; adaptive thinking is always on with effort parameter controlling depth.

Read the original at platform.claude.com ↗

EU AI Act obligations take force with million-dollar fines

iTech Guides · Nov 10, 2026 · Industry news

EU AI Act prohibitions, general-purpose AI obligations and Article 50 transparency requirements entered into force as of 10 November 2026, with registration beginning that month. Reported penalties reach $1 million for a first violation and $3 million for repeat violations.

Why it matters The EU has the broadest economy-wide, risk-based AI statute as of 10 November 2026; this affects enterprises deploying AI systems in European markets immediately.

The EU AI Act enforcement phase has begun in early November 2026, marking a critical regulatory inflection point. Prohibitions, general-purpose AI (GPAI) obligations and Article 50 transparency requirements—requiring disclosure when users interact with AI systems and marking of synthetic content—became enforceable on 2 August 2026, with full registration requirements kicking in by November. Violations carry steep penalties: €1 million ($1M) for first breaches and €3 million ($3M) for repeat violations, creating immediate compliance urgency. The scope is broad: any organization deploying or selling AI systems in EU jurisdictions must now maintain technical documentation, copyright policies, training-data summaries, and incident notification within 72 hours. Additional milestones loom: 2 December 2026 brings a ban on non-consensual intimate imagery generators, and 9 December 2026 introduces the Product Liability Directive covering AI software with no-fault liability. This creates a hard enforcement calendar that enterprises operating across EU and non-EU markets must track closely for vendor procurement, internal controls, and incident response procedures.

Read the original at itechguides.com ↗

Australia creates AI safety regulator for frontier models

The Conversation · Nov 9, 2026 · Industry news

Prime Minister Anthony Albanese announced a new artificial intelligence safety regulator that will enforce penalties for frontier AI companies breaching their obligations. The government is also considering criminal law changes to deal with serious misconduct by frontier AI companies.

Why it matters Measures target frontier AI companies building the largest, most advanced models, requiring security controls, risk identification, mandatory incident notification, independent model evaluation, and published safety frameworks.

Australia's Albanese Government is establishing dedicated regulatory machinery for frontier AI safety and security, signaling a more aggressive enforcement posture than many peer nations. The new regulator will have explicit authority to impose penalties on frontier AI developers—defined as companies building the largest, most advanced models—that breach safety and security obligations. Six mandatory requirements emerged: maintaining security controls across AI development and deployment phases; identifying and managing serious risks linked to models; rapid incident reporting to the Australian government; enabling independent third-party evaluation of models; maintaining and publishing transparent safety and security frameworks for governance and risk management; and responding to evolving technical standards. The government is additionally exploring criminal law amendments to prosecute serious misconduct, suggesting enforcement teeth beyond administrative penalties. This parallels EU and UK efforts but focuses specifically on frontier capabilities rather than broad AI adoption. For vendors and enterprises, this creates another enforcement jurisdiction with distinct notification and documentation requirements; notably absent are the transparency/marking requirements central to EU and UK frameworks, suggesting Australia is prioritizing upstream developer accountability over downstream user notification.

Read the original at theconversation.com ↗
Nothing in today's briefing matches that.

Listening

Podcasts and talks

Vendor Radar

Last 7 days · arrows compare with the 7 before

What changed this week

Last 7 days vs the 7 before

Biggest moves

Trending topics

agent · automation · observability · Agentic AI · OpenTelemetry · RAG · AIOps · Kubernetes · inference · MCP · LLM · SRE

Get the daily briefing

Today's 3 things that matter and every story with why it matters, in your inbox each morning. Free, and you can unsubscribe at any time. Prefer a reader? Follow the RSS feed.