JFrog swampUP 2026 brought developers, security specialists, platform engineers, and technology leaders to The Glasshouse in New York City from September 1 through 3. The conference focused on a blunt operational problem. AI agents are creating and consuming software faster than conventional security, remediation, and compliance processes can respond.
JFrog’s answer was “Trusted Intelligence,” backed by new controls for AI assets, autonomous remediation, continuous compliance, and software package traffic.
Key Announcements
Zero-Touch Remediation Takes Aim at the Patch Queue
JFrog introduced Zero-Touch Remediation, which automatically identifies an available fix for a known vulnerability and applies it through the customer’s pipeline. The system is designed to avoid breaking builds, forcing version updates, or interrupting developer workflows.
The capability draws fixes from JFrog’s new Self-Healing Software Supply Chain Security Ecosystem. Initial participants include Broadcom, Chainguard, Echo, IBM/Red Hat, Moderne, TuxCare, and Seal Security.
Eyal Dyment, vice president of security products at JFrog, argued that vulnerability tickets followed by weeks of manual patching have become a liability as attackers use AI to compress exploitation timelines. His framing sets up the test automated remediation has to pass, which is whether organizations can reduce exposure windows while preserving build stability and proving what changed.
AgentSecOps Puts Controls Inside AI Workflows
JFrog announced capabilities for securing what it calls the “agentic workforce.” These include AI Asset Scanning, Agent Package Resolution, AI asset registries, and JFrog Agent Plugins. The controls inspect the models, Model Context Protocol servers, skills, plugins, context files, and packages that agents consume.
Agent Package Resolution provides authenticated access through JFrog so agents retrieve approved packages through established organizational policies. Agent plugins connect coding agents with the JFrog Platform, allowing agent activity across the software supply chain to operate within centralized security and governance controls.
JFrog co-founder and CTO Yoav Landman said established DevSecOps controls were built around the assumption that a human developer sat at the keyboard. Coding agents break that assumption by selecting dependencies and taking actions without waiting for a person to review each step.
DevGovOps Turns Compliance Into a Release Function
New JFrog AppTrust capabilities automate governance through four continuous pillars, Codify, Attest, Enforce, and Monitor. The release includes prebuilt policy templates for frameworks such as the EU Cyber Resilience Act and NIST guidance, automated evidence capture, release gates, and policy-as-code capabilities.
The business case extends beyond faster audits. Machine-generated software can move from planning to deployment in hours, creating a dangerous mismatch when compliance reviews still depend on spreadsheets, manual evidence gathering, and quarterly checkpoints.
JFrog co-founder and CEO Shlomi Ben Haim said governance must keep pace regardless of whether code comes from a person or an agent. That places compliance inside release engineering, where policies can block, approve, or document software before distribution.
Package Traffic Control Moves to the Network Edge
JFrog’s Package Traffic Controller intercepts direct package requests from developers, agents, and other users, then redirects those requests through JFrog for inspection and logging. The company announced work with Cloudflare, Netskope, and Zscaler to apply this control at the network edge.
The announcement addresses a common weakness in software governance. An organization may operate a carefully controlled artifact repository while developers or automated tools still download components directly from public sources, creating blind spots in provenance, malware detection, and license review.
Wiz Connects Runtime Risk With Artifact History
JFrog also announced an integration with Wiz, now part of Google Cloud, that combines JFrog’s artifact information with cloud and runtime findings. Security teams can use that connection to trace an exposed component back to its binary, owner, and development context.
That linkage can reduce the time spent reconciling application-security findings with cloud alerts. It also gives remediation teams added context for deciding which vulnerability needs immediate action based on where the affected software is running.
Strategic Insights
Software Consumption Is Becoming the New Security Boundary
AI agents expand software consumption by selecting packages, models, plugins, skills, and other assets while completing assigned tasks.
JFrog’s announcements consistently targeted the point where those assets enter development. AI Asset Scanning, Agent Package Resolution, registries, and Package Traffic Controller all seek to establish an approved route through which software and AI components can be inspected before use.
An agent optimized to finish a task may treat an unverified public package as a convenient shortcut, which is where that emphasis earns its place. Organizations need technical enforcement that limits available sources and records what the agent selected, because instructions alone leave too much to interpretation.
Remediation Speed Is Becoming a Machine-Scale Contest
The conference treated vulnerability remediation as an automation problem shaped by AI-assisted attack speed.
During a swampUP panel, speakers discussed frontier models that can chain known vulnerabilities and accelerate discovery. Recent open-source auditing work illustrates the scale, with a single model run surfacing more than 500 serious vulnerabilities across widely used libraries in a matter of weeks.
That scale threatens ticket-based operations. Zero-Touch Remediation proposes a pipeline response in which approved fixes are selected and applied automatically, while AppTrust records the action and associated evidence.
How much autonomy security teams are prepared to grant remains the unsettled question. Automated patching needs policy thresholds, rollback procedures, build verification, and clear rules for cases where several technically valid fixes carry different operational risks.
Compliance Is Moving Into Platform Engineering
Continuous evidence capture turns compliance from a periodic project into a property of the delivery system.
JFrog’s DevGovOps message assigns platform teams a larger role in regulatory execution. Policies, attestations, approval conditions, and monitoring become reusable services embedded in pipelines rather than separate work performed shortly before an audit.
This shift can reduce audit preparation, but it raises the bar for policy design. A poorly written automated rule can block safe releases, approve risky ones, or create evidence that looks complete while missing the control an auditor actually needs.
Runtime Context Is Reshaping Prioritization
The Wiz integration reflects a broader shift from counting vulnerabilities toward understanding whether vulnerable software is deployed, exposed, and reachable.
Binary lineage and runtime signals give security teams a stronger basis for remediation decisions. A severe issue buried in an unused package requires different handling from the same issue inside an internet-facing production service, even when both produce identical scanner scores.
The Undercurrent
The announcements shared one premise. Human review cannot remain the traffic controller for every machine-generated action. JFrog repeatedly placed policy, inspection, evidence capture, and remediation directly into the systems that agents use to obtain dependencies and ship software.
Speed and authority pull against each other in that design. JFrog wants organizations to permit more autonomous activity while making the surrounding platform stricter, more observable, and harder to bypass.
The platform becomes the enforcement point for machines that can produce software around the clock.
Why It Matters
Technology leaders should treat agent adoption as a software supply chain redesign. Every coding agent creates questions about approved asset sources, package selection, release evidence, and vulnerability ownership, along with the conditions under which automated remediation may change a build.
The practical response begins with inventory. Teams should identify which agents are operating, which registries and public repositories they can reach, what credentials they use, what assets they create, and whether their actions appear in existing audit records.
Leaders should then test automation against measurable outcomes. Useful measures include vulnerable-component exposure time, the percentage of package requests routed through approved controls, remediation success rate, and rollback frequency. Audit-evidence preparation time and the number of untracked AI assets entering development round out the set.
What’s Next
Roadmaps should connect agent deployment with package governance, binary lineage, runtime visibility, and automated compliance. A coding assistant rollout without those controls can increase delivery speed while multiplying unverified dependencies and undocumented decisions.
JFrog’s announcements give buyers a specific evaluation agenda. Test whether automated remediation protects build integrity, determine whether agent controls cover every asset type in use, and verify that generated evidence satisfies internal and regulatory requirements. One question does more work than the rest. Which actions can the platform prove, block, reverse, and assign to an accountable owner?
