Nov 4: Attackers chain flaws to root 25 exposed PAN-OS firewalls
Today's 3 things that matter
- Attackers chain flaws to root 25 exposed PAN-OS firewalls
CISA added CVE-2026-46300 to its Known Exploited Vulnerabilities (KEV) catalog, requiring federal agencies to apply fixes by October 31, 2026; this vulnerability can be chained with CVE-2026-109618 and CVE-2026-109619 to implant PHP web shells and run commands as root. All internet-exposed PAN-OS management interfaces must be patched immediately. - Itential launches FlowAI agentic orchestration with MCP control layer
Lumen using Itential collapsed 1 billion alerts into 57,000 actionable incidents, runs 350+ live workflows, and cut customer-impacting incidents by 21%—demonstrating concrete operational gains from agentic orchestration in production. - Northstar Security leaves stealth with $7M for security agents
For security ops and AIOps teams, Northstar's Security Context Graph enables AI agents to correlate disparate findings into actionable risk remediation—moving from alert fatigue to continuous autonomous fixing.
Full stories below, grouped by topic.
Network Automation
Only 30% of enterprises automate as AI widens the gap
Itential Blog · Oct 29, 2026 · Primary source
What happened: Itential's Head of AI & Developer Relations addresses the skills gap and adoption barrier: only 30% of enterprises are automating, and those that are already layering AI on top—widening the gap between automation leaders and laggards.
Why it matters: Frames the practical reality for practitioners: AI-aware network engineers have an advantage, but adoption requires automation foundations first, and the gap between leaders and laggards is accelerating.
John Capobianco frames the challenge for network engineers: the 70% not automating are falling further behind as the 30% already automating rapidly adopt AI on top of their foundations. The practical shift for engineers is not replacement but augmentation—AI takes on repetitive analysis, generates automation, and coordinates tools, while engineers define intent, assess risk, validate results, and set operational guardrails. This directly addresses NetDevOps practitioners: having Ansible, Nornir, or Terraform foundations is no longer sufficient if those automation systems aren't AI-aware or agent-ready. The article emphasizes that effective network engineers will be those who use AI as a force multiplier rather than those who resist it. For organizations in the middle of their automation journey, this signals urgency: delay in AI adoption now creates compounding disadvantage as automation leaders implement agentic operations.
Read the original at itential.com
AIOps & Network Observability
ThousandEyes ships Splunk Synthetic Monitoring integration and deprecates OTel attribute
ThousandEyes · Oct 28, 2026 · Primary source
What happened: ThousandEyes made Splunk Synthetic Monitoring integration generally available on October 28, enabling creation and import of HTTP Server tests directly in Splunk Observability Cloud. Deprecated OpenTelemetry attribute (thousandeyes.source.agent.connection.type) for Endpoint Experience Local Network metrics as of October 29.
Why it matters: Network observability teams using Splunk can now correlate ThousandEyes network path data with application performance metrics in a unified view, reducing tool-switching during incident investigation and MTTR.
The Splunk integration addresses the fragmentation problem where network assurance and APM teams work in separate tools. ThousandEyes customers can now create and manage synthetic tests within Splunk's interface, view network measurements alongside application results in shared dashboards, and pivot back to ThousandEyes for detailed path analysis. The deprecation of the legacy attribute reflects OpenTelemetry standardization: teams using Data Model v2 must update their dashboards, alerts, and processing pipelines to use the standard network.connection.type attribute instead. This pushes observability platforms toward vendor-neutral telemetry standards, reducing lock-in and simplifying multi-tool correlation. The integration surfaces a concrete NOC workflow: when application latency spikes, operators can now see network path contribution metrics without context-switching, then drill into ThousandEyes for hop-by-hop diagnosis. OpenTelemetry alignment means teams exporting to other backends (Datadog, Dynatrace) face consistent attribute naming.
Read the original at docs.thousandeyes.com
Routing & Internet
Internal monitoring misses BGP changes beyond an operator's control
LogicMonitor · Oct 28, 2026 · Industry news
What happened: LogicMonitor published a technical guide on how BGP announcements, replacements, and withdrawals shift reachability across networks beyond an operator's control, covering the 900,000+ route entries in the global IPv4 table and why internal monitoring misses these events.
Why it matters: Network operators need outside-in monitoring to catch BGP route changes propagating from distant networks, since internal metrics stay green while reachability fails on paths you don't control.
The piece explains how BGP route propagation works operationally: each network prepends its own AS number as it advertises routes, building the AS path chain that grows one hop at a time across the Internet. Networks reject routes containing their own AS number to prevent loops. The article emphasizes that with over 900,000 route entries in the global IPv4 routing table, manual tracking is infeasible. Critically, internal infrastructure monitoring (server health, cloud resources, network devices) cannot see routing changes on external networks—failures on paths three hops away can break reachability while all local systems report green. The piece highlights the difference between internal metrics (logs, traces) and external connectivity, demonstrating why route flaps, replacements, and withdrawals require edge-based monitoring using outside-in vantage points to catch the real impact before users are affected.
Read the original at logicmonitor.com
Cloudflare halves DNS build delay and cuts memory 85%
Cloudflare Blog · Oct 30, 2026 · Primary source
What happened: Cloudflare reported a major DNS infrastructure optimization that eliminated the need to rebuild entire DNS zones when zone settings change, cutting build delays in half and reducing memory usage by 85%, even for zones with tens of millions of records.
Why it matters: DNS zone mutations are now incremental rather than full rebuilds, directly reducing latency on configuration changes and freeing significant infrastructure headroom for DNS operators running massive authoritative zones.
Cloudflare's engineering post details operational DNS performance improvements that eliminate a major operational pain point: previously, any zone setting change forced a full rebuild of the entire DNS zone database, creating cascading delays for operators managing zones with tens of millions of records. The new incremental approach tracks only the changed records, halving build delay and recovering 85% of memory previously consumed by redundant work. This is operationally significant because DNS zone mutations are frequent (TTL adjustments, record adds/deletes, policy changes) and happen across global infrastructure. The fix is a direct infrastructure efficiency gain—the same DNS capacity now handles more zones and faster configuration propagation. This type of engineering optimization is critical for CDN and DNS providers managing authoritative service at scale, as it directly impacts both operator agility and customer experience during configuration changes.
Read the original at blog.cloudflare.com
Network Security
Attackers chain flaws to root 25 exposed PAN-OS firewalls
Cybersecurity Dive, Dream Research / Palo Alto Unit 42 · Oct 30, 2026 · Industry news
What happened: A suspected China-nexus actor chained flaws in PHP2Shell with a privilege escalation bug in the Linux kernel to gain root access to Palo Alto Networks firewalls. The attacks reached more than two dozen targets, including multiple servers linked to the U.S. government; the threat actor, assisted by AI, gained root access to 25 PAN-OS firewall and management servers in the U.S., Japan, Taiwan and South Korea.
Why it matters: CISA added CVE-2026-46300 to its Known Exploited Vulnerabilities (KEV) catalog, requiring federal agencies to apply fixes by October 31, 2026; this vulnerability can be chained with CVE-2026-109618 and CVE-2026-109619 to implant PHP web shells and run commands as root. All internet-exposed PAN-OS management interfaces must be patched immediately.
On October 30, 2026, researchers at Dream and Palo Alto Networks Unit 42 disclosed an ongoing AI-assisted campaign targeting internet-exposed Palo Alto Networks PAN-OS management interfaces. Attackers exploited PHP2Shell (CVE-2026-109618 and CVE-2026-109619, two use-after-free vulnerabilities rated CVSS 9.2) combined with CVE-2026-46300, a Linux kernel privilege-escalation flaw rated CVSS 9.8 critical, to achieve root-level code execution on affected firewalls. The campaign compromised 25 systems: 24 internet-facing firewalls plus one internal management server. After gaining firewall access, attackers pivoted to Microsoft Active Directory, resulting in directory enumeration and credential theft. The attack surface was limited—approximately 0.02% of Palo Alto's global installed base ran internet-exposed management interfaces—but impact was severe for those exposed. Patches were released for all vulnerable PAN-OS versions (12.2, 12.1, 11.2, 11.1, 10.2) by October 30. Network operations teams must: (1) immediately patch all PAN-OS versions; (2) disable internet-facing management interfaces or restrict to trusted IPs; (3) enable TLS decryption on management traffic if possible; (4) review firewall audit logs and AD changes since mid-October; and (5) assume firewall and AD compromise vectors if internet exposure was confirmed.
Read the original at cybersecuritydive.com
CISA flags VMware ESX and Linux kernel flaws exploited
CISA Known Exploited Vulnerabilities Catalog · Oct 28, 2026 · Primary source
What happened: CISA added two new vulnerabilities to its Known Exploited Vulnerabilities (KEV) Catalog on October 28, 2026, based on evidence of active exploitation: CVE-2026-41709 (Broadcom VMware ESX Insufficient Logging Vulnerability) and CVE-2026-46300 (Linux Kernel Missing Initialization of a Variable Vulnerability). Both are being exploited in the wild against infrastructure workloads.
Why it matters: CVE-2026-46300 is part of the chained exploit chain used against Palo Alto firewalls; CVE-2026-41709 affects virtualization infrastructure. Both trigger Federal FCEB remediation deadlines under BOD 26-04. Network teams running VMware ESX or Linux-based appliances must prioritize patches for these two CVEs alongside the Palo Alto and Fortinet advisories issued this week.
CISA formally added two network-critical vulnerabilities to its authoritative Known Exploited Vulnerabilities catalog on October 28 under Binding Operational Directive 26-04. CVE-2026-41709 is an insufficient logging flaw in Broadcom VMware ESX hypervisors, enabling attackers to perform actions without proper audit trail visibility—particularly dangerous in zero-trust security models that rely on logging. CVE-2026-46300 is a missing variable initialization in the Linux kernel (also known as the Fragnesia kernel flaw), rated CVSS 9.8 critical, and confirmed to be part of the chained attack chain used against Palo Alto PAN-OS systems this week. Both are confirmed actively exploited. Federal civilian agencies face rapid remediation deadlines per BOD 26-04. Network infrastructure teams should: (1) audit ESX host versions for 41709 and apply Broadcom patches urgently; (2) ensure Linux kernel patches are deployed on all edge appliances, load balancers, and management systems; (3) cross-reference attack timelines with firewall/appliance logs to detect CVE-2026-46300 exploitation; and (4) treat these as Tier-1 patch priorities given confirmed active exploitation.
Startups & Funding
CrowdStrike buys Fiddler AI for agent observability in Falcon
TechMonitor · Nov 4, 2026 · Industry news
What happened: CrowdStrike signed a definitive agreement to acquire Fiddler AI, a leader in AI agent observability and explainability. Fiddler's capabilities will be integrated into CrowdStrike's Falcon platform as Falcon Observe, expected generally available in Q4 FY27, including pre-deployment evaluation, continuous monitoring of agent behavior and explainability tools.
Why it matters: As AI security and observability converge, enterprises need to understand their agents' actions, intervene when they go off course, and explain why—critical for ops teams deploying autonomous agents in production.
CrowdStrike signed a definitive agreement on November 3, 2026, to acquire Fiddler AI at its Fal.Con Europe 2026 conference. The Wall Street Journal reported CrowdStrike will pay about $400 million for Fiddler and about $95 million for Dreadnode, an AI security testing company, for a combined $495 million. The integrated Falcon Observe product will include pre-deployment evaluation, continuous monitoring of agent behavior and explainability tools meant to show how and why AI agents reach decisions. For infrastructure practitioners, this acquisition matters because at least six agent observability vendors have agreed to sell in 2026: Galileo to Cisco's Splunk, Arize to Dynatrace, Langfuse to ClickHouse, Traceloop to ServiceNow, Helicone to Mintlify, and now Fiddler to CrowdStrike. Agent observability is the practice of tracing an agent's decisions, reasoning and tool calls, then scoring them for accuracy and policy compliance. The deal closes in CrowdStrike's fiscal Q4 2027.
Read the original at techmonitor.ai
Northstar Security leaves stealth with $7M for security agents
SecurityWeek · Oct 28, 2026 · Industry news
What happened: Northstar Security emerged from stealth with $7M in seed funding led by Wing Venture Capital to develop AI agents for security teams. The startup addresses cybersecurity's biggest challenge: security teams have more data and tools than ever, but still lack the connected context needed to understand risk and take action.
Why it matters: For security ops and AIOps teams, Northstar's Security Context Graph enables AI agents to correlate disparate findings into actionable risk remediation—moving from alert fatigue to continuous autonomous fixing.
Northstar Security emerged from stealth on October 28, 2026, with $7M in seed funding led by Wing Venture Capital. The startup tackles security with a live Security Context Graph that charts the links between assets, identities, vulnerabilities, applications and business systems, allowing both people and AI agents to work from this common picture to answer intricate questions, surface concealed exposures and run autonomous remediation with human sign-off where needed. Northstar's platform connects with more than 70 enterprise security and IT products, including Okta, GitHub, Workday, Microsoft Defender, Microsoft Entra, AWS and Google Cloud. CEO Amer Deeba says security teams are drowning in data but starving for context, noting the industry has spent decades building better tools to find security problems but connecting the dots and fixing them remains painfully manual. Customers can give external AI agents access to the graph through the Model Context Protocol (MCP) and create their own security workflows.
Read the original at securityweek.com
Agentic AI & MCP
AWS Step Functions become agent tools over MCP
AWS News Blog · Nov 2, 2026 · Primary source
What happened: AWS Step Functions, which orchestrates over 220 AWS services, can now be handed to an agent as a tool through Amazon Bedrock AgentCore Gateway, which connects agents to tools over the Model Context Protocol (MCP). The agent decides when to run a workflow and what input to pass; Step Functions runs each step exactly as defined.
Why it matters: Existing workflows teams already trust became callable without a rewrite; operations teams can now expose existing orchestration as MCP-accessible tools for agentic control while preserving safety and auditability.
This release closes a critical gap for NetOps and SRE teams: AI agents on Bedrock AgentCore can now invoke 220+ AWS services through Step Functions workflows using MCP as the protocol bridge. Instead of agents calling APIs directly, they trigger pre-defined, tested workflows—preserving safety and auditability. MoEngage's parallel effort exposed 400 internal REST APIs across 7 regions as MCP tools, demonstrating the pattern: existing operational automation becomes discoverable and callable by agents without rewriting infrastructure. For operations teams, this means agents can reason about which workflow to run, but execution remains within guardrails. The practical advantage: teams can safely empower agents to orchestrate complex, multi-step operations (deploy, validate, rollback) while maintaining human-level control over what actions are possible. Available in all Bedrock AgentCore regions at standard rates.
Read the original at aws.amazon.com
Gartner names IBM leader in AI agent management platforms
IBM Announcements · Nov 2, 2026 · Vendor release
What happened: IBM has been named a Leader in the inaugural 2026 Gartner Magic Quadrant for AI Agent Management Platforms. IBM watsonx Orchestrate provides agent orchestration and operational management; watsonx.governance extends visibility, control, and accountability for AI governance, risk and compliance.
Why it matters: IBM's agentic control plane coordinates and oversees agent operations across heterogeneous environments with broader approach to managing agent operations alongside enterprise AI risk and compliance requirements.
IBM's placement as a leader reflects enterprise demand for agent platforms that handle orchestration, observation, and governance—not just individual agents. The distinction matters: watsonx Orchestrate acts as the execution layer, coordinating multiple agents and observing their behavior in production. Watsonx.governance bridges the operational/compliance gap—teams can see what agents did, audit decision paths, and tie agent outcomes to broader governance requirements. For IT Ops and SRE teams managing production agents, this two-platform approach answers a real need: enable agent autonomy while maintaining audit trails and policy enforcement. The product targets the heterogeneous environments most large enterprises run—multiple clouds, multiple agent runtimes, multiple vendors—requiring a unifying orchestration and governance layer. The MQ leadership reflects that IBM's positioning around accountable agentic AI at scale resonates with risk-conscious enterprises.
Itential launches FlowAI agentic orchestration with MCP control layer
Itential · Nov 4, 2026 · Primary source
What happened: Itential unveiled FlowAI, delivering agentic orchestration for infrastructure operations, with MCP control layer that translates AI intent into governed workflows. Separates reasoning from deterministic execution with policy, audit, and compliance controls.
Why it matters: Lumen using Itential collapsed 1 billion alerts into 57,000 actionable incidents, runs 350+ live workflows, and cut customer-impacting incidents by 21%—demonstrating concrete operational gains from agentic orchestration in production.
FlowAI represents the maturation of agentic orchestration for infrastructure operations. The key insight: Lumen's MCP-based proof-of-concept connects 90+ tools across multiple LLMs with earned autonomy, governed exposure, and agentic readiness built on the operating model that came first. The architecture separates concerns: Selector (observability) identifies issues and proposes actions; Itential (orchestration) executes them under policy guardrails; ServiceNow (ITSM) surfaces work to engineers. For NetDevOps and AIOps teams, this three-layer pattern is becoming the industry standard. MCP is the bridge between reasoning (agents deciding what to do) and execution (Itential deciding how safely). FlowAI explicitly targets the telco use case—millions of devices, multi-vendor stacks, regulatory requirements—where agentic operations at scale require both intelligence and control. At FutureNet (Nov 4-5), Itential is showcasing this as operational reality, not concept.
Read the original at itential.com
83% of firms need infrastructure upgrades for autonomous systems
Google Cloud Blog · Nov 4, 2026 · Research
What happened: Google Cloud survey of 1,402 global infrastructure leaders shows 83% of organizations require infrastructure upgrades to support production-grade autonomous systems. 52% now use hybrid multicloud architecture as agentic workloads move from pilots to production.
Why it matters: Oracle and Google announced TPUs on Oracle Cloud Infrastructure starting 2027, explicitly supporting agentic, long-context, and multi-step reasoning workloads that production agents require—infrastructure modernization is critical blocker for agent operations maturity.
This research, updated Nov 4, quantifies the infrastructure challenge behind agentic operations: 83% need infrastructure upgrades to handle the compute, networking, and storage demands of production agents. The implication is significant for NetOps and SRE teams: agentic AI success depends not just on agent software but on underlying infrastructure modernization—faster interconnects, better observability, higher-throughput APIs, and resilience patterns that handle agent-driven traffic. The 52% hybrid multicloud finding shows infrastructure has become heterogeneous; agents must reason and act across multiple environments. Google and Oracle's TPU partnership (2027 launch) signals vendor recognition that agents require specialized compute for long-context reasoning and multi-step inference—not traditional GPU clusters. For infrastructure leaders, this means agentic operations maturity is constrained by data center architecture, not just software.
Read the original at cloud.google.com
Telco & Cable AI
Verizon rolls out Ericsson intelligent controller for RAN automation
Ericsson · Nov 3, 2026 · Primary source
What happened: Verizon and Ericsson announced nationwide rollout of Ericsson's Intelligent Controller (EIC) to support AI-native radio access network automation, with field tests showing AI cell shaping achieving gains of up to 8.9% on downlink throughput and 8% on uplink UE throughput.
Why it matters: Verizon plans EIC rollout nationally by year-end 2026 with self-designed rApps accelerating toward Level 4 autonomy. Operators must evaluate how closed-loop RAN optimization changes capacity planning and SLA design.
Verizon and Ericsson are advancing autonomous networks by integrating AI into RAN automation through partnership on the Ericsson Intelligent Controller and AI-native rApps. Field testing demonstrated AI cell shaping of Massive MIMO and traffic steering in live environments, showing the agility of an open, programmable network enabling Verizon to deploy self-designed rApps. Fully autonomous closed-loop rApps running round the clock, including Ericsson's Advanced Antenna System Cell Shaper, raised downlink data radio bearer throughput by up to 8.9% and uplink UE throughput by up to 8%. This moves Verizon closer to Level 4 autonomy while maintaining operator control of intent and guardrails. Building its own rApps on the open platform lets Verizon automate more of its network as it prepares for 6G. The nationwide rollout by year-end 2026 signals the shift from AI-RAN pilots to production-scale autonomous RAN optimization in North America.
Read the original at ericsson.com
Bell and Telus plan 450 MW of AI data centres
TelecomLead · Oct 28, 2026 · Industry news
What happened: Bell and Telus are shifting capex toward AI infrastructure and network automation. Bell plans a 300 MW AI facility near Regina (operations 2027), while Telus invests CAD66 billion over five years with 150 MW expansion across three British Columbia sites by 2032 including 60,000 GPUs.
Why it matters: Canadian carriers are positioning as regional AI infrastructure providers, not just connectivity. Network architects must account for interconnection strategies, edge placement, and multi-cloud peering tied to carrier-operated AI compute resources.
Bell, Rogers, and Telus are shifting investment priorities toward 5G standalone, fiber, fixed wireless access, enterprise connectivity, and AI infrastructure as traditional mobile revenue slows. Telus announced CAD66 billion investment over five years in network infrastructure, operations, and AI capacity. Bell is building a 300 MW AI facility near Regina with operations starting in 2027, while Telus operates a Sovereign AI Factory in Rimouski and plans 150 MW expansion across three British Columbia sites by 2032 with 60,000 GPUs when fully operational. Bell plans to use its national fiber backbone to connect AI computing infrastructure with enterprise customers and has deployed AI-powered network operations using Google Cloud for fault detection and incident resolution. Rogers introduced AI capabilities in its radio access network via partnership with Ericsson. These moves position Canadian carriers as AI infrastructure anchors, directly affecting multi-cloud interconnection, edge compute placement decisions, and enterprise network routing strategies.
Read the original at telecomlead.com
AI Model Providers
OpenAI tests GPT-6.2 Astra on a fifth of Codex traffic
Kucoin, CometAPI, 36Kr · Nov 2, 2026 · Industry news
What happened: On November 1, developers discovered gpt-6.2-astra in Codex API responses during gray release testing, with the new model already consuming 19.8% of traffic. OpenAI skipped GPT-6.1 due to security issues; Anthropic's Claude 5.5 emerged via code leak the same day, split into Mythos 5.5 for cybersecurity and Fable 5.5 for general use.
Why it matters: GPT-6.2 Astra reportedly surpasses Claude Opus 5.5 across multiple dimensions, directly affecting model selection criteria for production deployments and cost-per-task calculations.
GPT-6.2 Astra appeared undocumented in OpenAI's model catalog and API changelog as of early November, indicating a backend canary rollout rather than formal launch. OpenAI took a month to address safety, controllability, and engineering gaps from the canceled GPT-6.1. The concurrent discovery of Anthropic's Claude 5.5 variants—supporting up to 1 million token context across security-focused and general-use tiers—signals both labs accelerating releases in response to benchmarking pressure. Token economics diverge sharply: Claude Opus 5.5 burns ~119,000 output tokens per task at peak performance, while Astra uses ~27,000, making Astra more cost-efficient for deterministic tasks. Neither model is officially announced; practitioners should expect formal product pages and pricing tiers in coming days. The gray release pattern suggests OpenAI is stress-testing at scale before public announcement.
Read the original at kucoin.com
AI Industry & Policy
OpenAI raises $11 billion after cutting fundraising target
Wall Street Journal / TechCrunch · Nov 2, 2026 · Industry news
What happened: OpenAI cut its $30 billion fundraising target to $20 billion, securing only $11 billion in commitments at a $1.25 trillion valuation. Major investors showed tepid interest amid tech sector sell-off concerns over AI infrastructure spending sustainability.
Why it matters: Signals cooling investor appetite for heavy AI infrastructure capital, affecting access to frontier model training and inference scaling funding across the sector.
OpenAI lowered its target from $30 billion to $20 billion after securing ~$11 billion in commitments. The financing values the company at $1.25 trillion, down from the previously sought $1.4 trillion. OpenAI's Q1 2026 cost of revenue and R&D expenses totaled $12.1 billion, and the company held over $73 billion in cash and securities as of March 31. Major investors showed tepid response amid a broader tech sell-off triggered by weak quarterly results from Google, Amazon, and Intel. The company plans to approach family offices and high-net-worth individuals to close the remaining funding gap. Rival Anthropic is pushing toward an IPO as early as November, potentially securing well over $100 billion in capital. This contrasts sharply with aggressive capital-raise strategies of early 2026, signaling a market inflection in AI funding availability as investor concerns about the financial sustainability of AI infrastructure spending intensify.
Read the original at techcrunch.com
CrowdStrike pays $495M for Fiddler AI and Dreadnode
TechCrunch / Dealroom · Nov 3, 2026 · Industry news
What happened: CrowdStrike announced acquisition of Fiddler AI and Dreadnode for combined $495 million (~$400M for Fiddler, ~$95M for Dreadnode) to add AI agent observability, behavior monitoring, and automated security-testing capabilities to its Falcon platform.
Why it matters: Addresses critical governance gap as enterprises deploy autonomous agents with legitimate credentials; reflects urgent market need for agent traceability, behavioral audit trails, and pre-deployment security validation.
Fiddler AI specializes in AI agent observability and explainability; acquisition integrates it into Falcon as "Falcon Observe," available Q4 FY27, adding pre-deployment evaluation, behavioral accountability, and traceability so enterprises can trace agent actions and verify stay-within-task compliance. Dreadnode tests AI models, applications and agents for vulnerabilities before deployment, running over 100 known attack strategies including jailbreaks and prompt injection, mapping findings to OWASP Top 10 for LLMs, MITRE ATLAS, and NIST AI Risk Management Framework. CrowdStrike simultaneously launched AI Kill Switch in Falcon Guardian, enabling immediate agent halt if behavior drifts. Fiddler's purpose-built models reduce AI evaluation costs by up to 98%. Acquisition reflects enterprises' urgent need to audit and govern autonomous agent behavior in production—agents can hold valid credentials yet ignore instructions, hallucinate, or execute unintended actions. Deal signals acceleration in market race to make AI agents observable and controllable as agentic AI moves from pilot to production.
Read the original at techcrunch.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.