Follow on Twitter using the following hashtags or usernames: #XFD16
Watch on YouTube
Watch on Vimeo
Jeff Lombardo of Amazon Web Services demonstrated how AgentCore handles authentication, authorization, and audit for agentic AI. Using a fictional retail company, the session walked through AgentCore Runtime, Gateway, Identity, and Policy, covering JWT-based access to agents, credential translation for legacy systems, delegated OAuth flows, and fine-grained policy enforcement built on the open source Cedar language. Presenting at Security Field Day 16, Lombardo framed agent identity around delegation under human control, strong agent identity, dynamic permissions, and just-in-time issuance and revocation.
Lombardo showed Runtime gating agents with JWTs constrained by client ID or claims, then using each token’s issuer and subject to mark the agent’s own credential so that one user’s agent instance can only reach resources tagged for that user. Gateway exposes tools as MCP endpoints from existing MCP servers, Lambda functions, REST APIs described by an OpenAPI document, or prebuilt connectors for services like Salesforce, BambooHR, Confluence, and DynamoDB. It accepts short-lived OAuth tokens inbound and translates them into the static API keys legacy systems still require. Request and response hooks let teams enrich calls or redact data before it reaches the LLM, which Lombardo warned will try to weaponize whatever the agent learns. Identity stores only references to credentials, including secrets held in 1Password or HashiCorp Vault, so gateway builders never see the secrets themselves, and Gateway supports MCP URL elicitation for user consent.
The policy layer is where the platform becomes truly dynamic. Cedar evaluates principal, action, resource, and context with default deny and overriding forbid rules, and Lombardo stressed that its decisions are mathematically provable even across a million policies. AgentCore Policy runs Cedar as a managed service, generates the schema automatically from a gateway’s tool list, and translates natural language into policy, such as limiting managers to listing products twice per hour. That example relies on Dogwood, a newly released open source temporal extension that conditions access on prior actions, like requiring items in a basket before payment. Because policies are exposed through APIs, observability alarms can drive new restrictions in near real time, turning audit signals directly into tighter agent controls.
Personnel: Brandon Carroll, Jeff Lombardo
Watch on YouTube
Watch on Vimeo
Ram Ramani of Amazon Web Services explained what security at machine speed means as AI coding agents multiply the code and changes reaching production. The session introduced AWS Continuum, covering continuously updated threat models, agentic penetration testing on every deploy, and context-driven vulnerability management that proves exploitability in an isolated sandbox before proposing fixes. At Security Field Day 16, Ramani argued that security teams still triage at human speed while threat models go stale and pen tests happen once a year.
Ramani’s answer is shared security agents at every development phase, drawing on common context like the threat model, code repository, and deployment details. Agents update the threat model as pull requests land and feed it into design reviews, coding agents, scans, and pen tests. AWS Security Agent places a pen test gate before staging, scoped to changes since the last deploy, which teams can schedule, trigger manually, or cap by agent hours. Continuum ingests findings from existing scanners alongside its own, uses a collector built on AWS Config to understand network, container, and security group settings, then validates each finding in a sandbox and opens a remediation pull request. In the demo, Continuum confirmed an insecure direct object reference in an invoices API by exfiltrating three customer PDFs from an unprivileged account.
Rather than raw CVSS scores, Continuum rates issues on likelihood factors like reachability and authentication and impact factors like blast radius and data value, so the same flaw might rate critical when internet-facing or low inside a private VPC. Teams set thresholds that route critical findings to human review. Addressing trust in agentic tools, Ramani warned that models can memorize public benchmarks, and cited Continuum exploiting 33 of 37 live-application targets end to end on vulnerabilities disclosed after model training. He noted that CISOs are already shifting hiring toward high-judgment security engineers, because when exploitable proof replaces noisy findings, the debate moves from whether a flaw is real to whether it matters. His closing point was that findings are now easy and validation is where the value lies.
Personnel: Ram Ramani
Watch on YouTube
Watch on Vimeo
Jesse Lepich of Amazon Web Services took a deep dive into AWS Network Firewall and Route 53 DNS Firewall, explaining how the two services secure both the routed and recursive DNS egress paths in AWS environments. The session covered Network Firewall’s managed architecture, MadPot-powered active threat defense, automated domain list learning, and a console walkthrough of logging, monitoring, and managed rules. After a brief introduction from Brandon Carroll at Security Field Day 16, Lepich pointed to shrinking attack timelines driving demand for preventative guardrails.
Customers steer traffic to a Network Firewall endpoint in each availability zone, rated for 100 Gbps, where Gateway Load Balancer spreads flows across an AWS-managed EC2 fleet that shares state so instances can be replaced without dropping traffic. Egress security is the top use case, and Lepich stressed that attackers use the recursive DNS path independently of the routed path, even holding conversations with LLM chatbots through DNS records. Tighter integration lets DNS Firewall check with Network Firewall before answering a query, enabling domain filtering on any protocol, a checkbox to block direct-to-IP connections, and detection of SNI spoofing. Blocking DNS over HTTPS and public resolvers relies on documented rules from the best practice guide, and a delegate pressed for a managed easy button that Lepich agreed would help.
New capabilities aim to make protection easier to adopt. Active threat defense draws on MadPot, a global honeypot network with thousands of decoys across AWS regions that captures what attackers reach for after compromise. Traffic analysis mode is a free learning feature that lists observed domains with access counts and unique sources, then converts them into an allow list rule group with a click, though drilling into specific sources still requires the logs. Network Firewall also supports rules based on EKS and ECS workload names, URL categories, partner rules through AWS Marketplace, and an explicit proxy in preview, while the monitoring dashboard flags AWS services that should use VPC endpoints instead of crossing the firewall. Lepich’s recurring advice was to start with a loose allow list rather than deny lists and tighten it over time, bringing least privilege from identity to network flows.
Personnel: Jesse Lepich
Watch on YouTube
Watch on Vimeo
Ian Olson of Amazon Web Services introduced AWS Network Security Manager, the newly launched successor to AWS Firewall Manager for centrally managing AWS WAF and AWS Shield Advanced. The session demonstrated how policies, scopes, and deployments let central security teams, application teams, and threat analysts collaborate, including granting targeted WAF exceptions at scale and rolling out emergency mitigations in minutes. At Security Field Day 16, Olson recalled deploying WAF and Shield Advanced to about 5,000 endpoints in an hour for a customer under DDoS attack.
Network Security Manager launched about five weeks before the event. Rules are bundled into policies whose priorities set rule order, scopes define targets by organizational unit, account, tag, ARN, or properties like internet-facing load balancers, and deployments join the two. Enforcement is event driven through AWS Config, so if an application team deletes or replaces a Web ACL, the service restores the baseline, while retrofit mode layers rules onto existing Web ACLs without causing infrastructure-as-code drift. Olson’s exception demo set the AWS common rule set to count and add labels, followed by a low-priority rule that blocks labeled requests unless an exception label is present. An application-specific rule placed between those priorities adds that label, leaving room for more than 9,000 applications. Versioned scopes and rules let teams test updates in dev before promoting them, and an AI assistant generates rule JSON from plain language.
For threat analysts, Olson rolled out a managed PHP rule set at top priority by reusing the central team’s existing scopes, describing a sub-five-minute response to a vulnerability that only patching can fully resolve. Changes typically propagate in seconds to a couple of minutes. AWS has already announced Network Firewall and security group support. WAF now includes a built-in dashboard showing which AI crawlers hit an application, verifies Web Bot Auth signatures for around 400 named bots, and can charge verified bots for access. Olson framed Network Security Manager strictly as a management plane, saying more native insight tools would arrive around re:Invent, and stressed that putting production baselines in dev and sandbox environments eliminates the last-minute security conflicts that stall application teams before launch.
Personnel: Ian Olson
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!