Follow on Twitter using the following hashtags or usernames: #NFD41
Watch on YouTube
Watch on Vimeo
Bassam Khan of Cisco outlined the company’s secure networking strategy and argued that being AI ready and being secure are inseparable. He examined how agentic AI traffic changes network behavior, how the post-Mythos surge in reported vulnerabilities raises the stakes, and how Cisco’s three pillars of trusting the network, protecting every flow, and assuring every experience fit together. Kiran Ghodgaonkar opened Cisco’s Networking Field Day 41 appearance by previewing a day that brought together Cisco’s security and networking teams, then handed off to Khan.
Khan described agents as the new workforce and inference as the new workload, and he pointed to a recent Cisco research report on production traffic. AI flows last about twice as long as non-AI flows because agents hold sessions open while they wait on inference, which puts pressure on stateful devices like firewalls and NAC gateways. Agentic traffic also sends far more data upstream to frontier models as prompts and context grow, and it runs at machine speed. On security, he showed reported CVEs spiking in the post-Mythos era. He noted that Cisco was one of 12 vendors in Anthropic’s Project Glasswing and reviewed 1.8 billion lines of code across more than 25 languages, which Cisco’s security leadership called eight years of work in eight weeks. A demo showed a high-severity DNS alert, unresolved for nine days and flagged for possible hijacking, that an operator could ask about in plain language.
Khan called the network a must-have layer in post-Mythos defense in depth, because it sees everything while other tools come and go. Trusting the network depends on hardened IOS XE, runtime checks, hardware verification, Live Protect shielding, and post-quantum readiness. Protecting flows relies on identity, zero trust, and SASE, and assurance keeps rapid patches and segmentation changes from breaking the user experience. When asked about assurance, he acknowledged that it detects malicious activity rather than preventing it, but said it gives the SOC additional telemetry. On patching, Live Protect shields vulnerabilities until customers can upgrade. Ghodgaonkar added that engineering teams are removing unused code and disabling unneeded default features to reduce future CVE exposure. Khan’s conclusion was blunt: the network configurations that got organizations here won’t carry them into the post-AI world.
Personnel: Bassam Khan, Kiran Ghodgaonkar
Watch on YouTube
Watch on Vimeo
Saurabh Goel and Kush Shah of Cisco demonstrated how AgenticOps brings AI-driven operations to Cisco Cloud Control. Goel showed deep reasoning in the AI Assistant handling a long-horizon wireless capacity planning task. Shah walked through the agentic loop of sensing, diagnosing, and remediating network issues, using live incidents from Cisco’s own network and early results from beta customers. Goel and Shah, both product managers in Cisco Networking, presented at Networking Field Day 41.
Goel cited Cisco survey data showing that more than half of IT teams’ daily alerts are network related and that 90% of respondents face cross-domain issues, a load that exceeds what people alone can handle. Cisco Cloud Control addresses this with a single front door for login, inventory, topology, and alerts. From there, operators can use the AI Assistant, delegate work to agents, or bring in Cisco intelligence through MCP servers. In Goel’s demo, deep reasoning assessed whether a London office with six APs could support a 200-person event. It pulled data from Meraki dashboard APIs, scored AP health, channel utilization, and connection success, and returned an overall readiness score of four with prioritized fixes for DHCP and authentication failures. Shah then showed the agentic loop on Cisco’s RTP network. An incident for RADIUS rejections on a corporate SSID arrived with a briefing, recommended actions, and a full reasoning log, and a PoE alert led to a cable test that found a faulty cable.
Both presenters described autonomy as a journey of trust that moves from diagnosis and recommendations toward constrained operation and, eventually, automation under policy. Actions entered beta the week before the event. RBAC governs agent changes, and an audit trail records them. Cisco is also embedding ThousandEyes hop-by-hop data so that root cause analysis can look beyond Cisco’s own gear. Beta customers have already found a failing WAN link, an AP on an access port that affected 166 clients, and a retail guest circuit with no cellular failover. Delegates pushed back on calling statistical output “reasoning,” and Shah conceded that the model sometimes pulls in unrelated issues, adding that human feedback is how it improves. With roughly 150,000 queries a month, Goel said Cisco is still tuning guardrails to block misuse without shutting out users who are simply curious.
Personnel: Kush Shah, Saurabh Goel
Watch on YouTube
Watch on Vimeo
Hamzah Kardme and Nitin Kumar of Cisco explained how Cisco SASE is evolving to secure an agentic workforce. Kardme covered identifying and steering AI traffic at the edge, distributed security enforcement, and unified management in Cisco Cloud Control. Kumar detailed AI Access controls, agent identity, MCP server governance, and semantic inspection of agent tool calls. Both presented as part of Cisco’s appearance at Networking Field Day 41.
Kardme noted that a single question to a personal agent now fans out into many tool calls, API calls, and TLS sessions. Cisco SASE appliances use deep packet inspection with signatures fed from the cloud, so they can classify new AI tools without upgrades. That lets them steer enterprise AI to the best WAN path and limit consumer AI. One fulfillment-center customer’s robots became confused during latency spikes, which forced a 35-minute floor reset each time. Segmentation context from VPNs and security group tags now extends to SASE PoPs, so policy can follow a group tag rather than an IP address. In the demo, Kardme onboarded a London site through a guided workflow with automatic PoP selection, then used AI Canvas to trace Trojan traffic to a roaming client. Kumar described AI Access, which covers more than 3,800 AI applications. He demonstrated blocking unsanctioned tools, enforcing the corporate ChatGPT tenant, endpoint DLP, and guardrail alerts on prompt injection.
Kumar described the next step as zero trust for agents. An integration with Astrix pulls in signals from AI platforms, EDR tools, browsers, and non-human identity fingerprinting. Duo’s identity views showed 73 discovered agents, including a Cursor agent rated high risk because it used a deprecated MCP tool. Agent Gateway, now in beta, applies access policy and segmentation to the MCP servers customers build. Semantic inspection, which is already live, examines every agent tool call. It caught a weather MCP server being asked to harvest user data, and it also surfaces shadow MCP servers. Delegates asked about securing cloud-hosted agents and about forcing deployments through change control. Cisco deferred the first question to Security Field Day and acknowledged that it can’t yet enforce the second. Kumar closed by describing the combined networking and security stack as a unified platform rather than a point product, with zero trust for every agent as the goal.
Personnel: Hamzah Kardame, Nitin Kumar
Watch on YouTube
Watch on Vimeo
Sarvjeet Kaur and Claire Gobran of Cisco IT shared how Cisco became customer zero for Cisco Secure Access and moved its global workforce to SSE in phases. They covered the shift from on-premises VPN to zero trust access and generative AI protection, the architecture and support model behind the rollout, and the lessons they learned along the way. Both presented as part of Cisco’s appearance at Networking Field Day 41.
Gobran explained that Cisco’s roughly 120,000 employees connect through 12 global hubs, nine of which now run SSE. A few hubs will stay on-premises because of regulatory requirements. The team started with VPN because it was the most mature use case and the legacy hardware was due for replacement, then moved on to zero trust access and generative AI protection. Leadership made a go/no-go decision at every phase. Today the entire workforce uses the SSE VPN, 155 applications are enabled, and the team has submitted about 115 features and enhancements to the product group. Replacing 24 location profiles with automatic nearest-location selection cut incidents dramatically. Kaur outlined the architecture: SD-WAN peers with SSE over BGP, DNS security has moved from Umbrella to Secure Access, and weekly reviews have extended generative AI protection to roughly 3,000 applications.
The rollout began with engineering users on the product team, who expected hiccups, then moved to Cisco IT staff and then to wider waves. The VPN track alone took a year. Kaur named the biggest problems as applications that hadn’t allowlisted SSE public IP addresses, man-in-the-middle issues during decryption for DLP, and authentication failures. When asked about TAC, the presenters said Cisco IT opens and pays for cases like any other customer and needs leadership approval to escalate them. Splunk handles alerting, ThousandEyes monitors all nine regions, and an in-house AI chat tool now resolves about 84% of support cases without a human. Gobran closed with lessons on executive buy-in, a willingness to roll back changes, and communication campaigns that eventually required users to migrate. She noted that the same experience now follows employees from campus to home to the road.
Personnel: Claire Gobran, Sarvjeet Kaur
Watch on YouTube
Watch on Vimeo
Sumant Mali and Mike Ketabforoosh of Cisco explained how Cisco is building resilient infrastructure for the AI and quantum era. Mali covered post-quantum cryptography across IPsec, SD-WAN, MACsec, and secure boot, including hybrid key exchange and live demos. Ketabforoosh introduced Live Protect, which uses eBPF-based shields to block exploits of known vulnerabilities until customers can patch. Both presented as part of Cisco’s appearance at Networking Field Day 41.
Mali explained that AES-256 remains safe, but public key exchange is exposed to harvest-now, decrypt-later attacks, and authentication and secure boot face real-time quantum risk. CNSA 2.0 will require ML-KEM, ML-DSA, and LMS for national security systems after January 2027. Cisco’s answer is hybrid key exchange under RFC 9370, which layers ML-KEM on top of a classical exchange. An “optional” keyword lets it fall back to classical cryptography, which supports a phased migration. In IOS XE 26.1 and 26.2, enabling it takes one CLI line for IPsec or three clicks for SD-WAN. SD-WAN now moves key exchange to IKEv2 and protects its control plane with TLS 1.3 and ML-KEM. Bringing up 15,000 FlexVPN tunnels takes about 21 minutes instead of 20. Mali showed interoperability with Cloudflare’s IPsec gateway and, for the first time in public, quantum-safe secure boot on the 8600 using LMS and ML-DSA signatures.
Mali advised customers to inventory their cryptography, classify long-lived data, and start with the WAN. Ketabforoosh said the time to exploit has shrunk from 2.3 years in 2018 to roughly 20 hours in 2026. Live Protect is built on the Tetragon agent from Cisco’s 2024 Isovalent acquisition and places eBPF programs in kernel hooks without changing binaries. PSIRT triages vulnerabilities, Talos builds the shields, and product teams test them. Shields run in monitor or enforce mode and are supported for six months, and Cisco currently recommends running no more than ten at once. Ketabforoosh stressed that a shield is not a substitute for patching. A demo showed a shield for last week’s SD-WAN control component API authentication flaw shipping alongside the fix. A delegate pointed out that monitor mode doesn’t stop a breach, while an operator in the room said Live Protect eases the race against a 20-hour exploit window on Cisco gear.
Personnel: Mike Ketabforoosh, Sumant Mali
Watch on YouTube
Watch on Vimeo
Mirko Grabel and Jubil Mathew of Cisco introduced Cisco Multi-Cloud Fabric, a network-as-a-service offering in Cisco Cloud Control that connects sites, regions, and AWS, Azure, and Google Cloud directly. They covered its Cisco-operated virtual points of presence, zero trust routing, ThousandEyes integration, and consumption pricing, and they demonstrated onboarding cloud networks and a Meraki site live. Grabel, the lead product manager, and Mathew, a technical marketing engineer, presented at Networking Field Day 41, and Kiran Ghodgaonkar closed Cisco’s appearance.
Grabel said that backhauling cloud traffic through data centers adds latency and fragments operations, and the problem compounds for AI workloads that repeatedly call data sources. In one example, a Los Angeles branch reaching cloud apps through a Dallas data center crossed eight hops, and connecting directly cut latency by about 75%. Multi-Cloud Fabric places VPoPs in each cloud. They run Cisco’s elastic routing stack in Cisco’s own account, where SRE teams handle sizing and scaling. Zero trust routing prunes routes and denies everything by default until users define who can talk to whom. Stateful firewalls are available through service chaining, and ThousandEyes agents are hosted inside the VPoPs. Customer VPCs connect over IPsec from a VPN gateway, limited at first to 1.25 Gbps per VPC, while the backbone offers hundreds of gigabits. Pricing combines per-attachment fees, data transfer charges, and premium services.
In the live demo, Mathew supplied cloud credentials and discovered VPCs across all three clouds. He onboarded them to the nearest VPoP, connected a new San Francisco Meraki site to three cloud applications, and showed health views for IPsec tunnels and BGP. Grabel contrasted the service with Cisco Cloud OnRamp, which suits customers who want to run their own routers. He said brownfield customers can migrate CIDR by CIDR alongside their existing networking. Delegates pressed on the per-VPC throughput limit and on security that stops at Layer 4 ACLs in the MVP, and Grabel said both will expand. The service runs in North America and Europe today, with Oracle Cloud as the next cloud, and it remains in public beta ahead of general availability.
Personnel: Jubil Mathew, Mirko Grabel
Thank you for being part of the Tech Field Day community! Our mailing list is a great way to stay up to date on our events and technical content, and we appreciate your signup.
We promise that we’ll never spam you, send ads, or sell your information. This list will only be used to communicate with our community about our events and content. And we’ll limit it to no more than one message per week.
Although we only need your email address, it would be nice if you provided a little more information to help us get to know you better!