Everything we announced at Chainguard Assemble London 2026

Chainguard Assemble London 2026 brought together developers, platform engineers, and security leaders to talk about what it takes to build software safely now that so much of it is written, tested, and shipped by AI agents.

Agents write code faster than teams can review it, and frontier models are uncovering exploitable vulnerabilities faster than most organizations can patch them. Meanwhile, attackers are using the same capabilities to flood public registries with malicious packages and push daily published CVEs to new highs.

Today's announcements take the secure-by-default approach behind the Chainguard Factory beyond the artifacts you deploy. They reach into the environments where code is written, tested, and verified, and they open up frontier vulnerability hunting that until now has been reserved for a select few. Here's everything we announced.

Chainguard Harness

Today, we're introducing Chainguard Harness, a new offering that lets any organization point Anthropic’s Claude Mythos 5.1 at a live, running copy of its application to hunt down exploitable vulnerabilities. Harness builds your application from source exactly as it ships to production, boots it inside an isolated microVM with no network egress and no credentials, and lets the model probe it for logic flaws, injection, resource exhaustion, and more. Every exploit is replayed in a fresh microVM until it lands every time, and every recommended fix is verified to stop it.

You receive a validated, reproducible report with severity, steps to confirm each finding, and a fix written for your codebase, all reviewed and approved by a Chainguard engineer. If we find a flaw in an open source dependency, we handle coordinated disclosure with upstream maintainers and ship the patched build through Chainguard Repository. Harness will be available through Claude Marketplace by Anthropic, so organizations can put part of their existing Anthropic commitment toward finding and fixing exploitable vulnerabilities. The beta waitlist opens today.

Chainguard Workspaces

Chainguard Workspaces, now in beta, gives developers and AI agents Chainguard-hardened, microVM-based development environments. Agents can install packages, run scripts, and edit configurations without ever touching a laptop that holds corporate credentials, SSH keys, and production access. Each Workspace is built from minimal, zero-CVE Chainguard OS packages and resumes from a warm snapshot in seconds. No long-lived secrets live inside: a host-side proxy injects short-lived credentials in flight, and optional egress policies block and log requests to anything you haven't allowed.

Developers and agents work in Workspaces through chainctl develop, which syncs your local working tree, including uncommitted edits, and snapshots the session when you detach. Teams can fan out parallel Workspaces to explore competing approaches, restore point-in-time snapshots, and rely on a tamper-proof flight recorder of every command, code change, and network call. When an agent hits a decision it shouldn't make alone, the Workspace routes the question to a developer and suspends it to a snapshot, so no compute is spent until someone answers.

Chainguard Checks

Chainguard Checks, also now in beta, turns CI into a fast pre-commit check that developers and AI agents can call from anywhere: the chainctl CLI, a git pre-commit hook, the Chainguard GitHub App, or a tool call through the MCP server or SDKs. Instead of waiting in CI queues while runners re-download the same dependencies, every check runs in a warm, isolated microVM with dependencies pre-baked from your lockfile, protected by the same declarative egress policies and in-flight credential injection used in Workspaces.

Every run produces a signed attestation binding the environment, verdict, and commit SHA, so merge gates can verify a passing result instead of re-running tests. Failing verdicts return structured logs that agents can use to self-correct, and inside Chainguard Workspaces, a passing verdict can serve as the gate for completing an agent loop. Design partners like Cuscal, which powers payment infrastructure for more than 90 banks, fintechs, and corporates, have been running Checks since the alpha.

Chainguard MCP

Agents are great at writing code, but they aren't great at knowing what's trustworthy. Chainguard MCP, now generally available, delivers four supported MCP servers that give agents verified answers at every stage of dependency selection. cg-oci confirms an image tag or digest exists before it lands in a Dockerfile, cg-apk finds exact Chainguard package equivalents by name, cg-versions checks whether a version is current or end of life, and cg-api verifies organization policy, entitlements, and attestation data.

Governance is enforced on the server side rather than through prompt guardrails, with response caps that prevent unbounded queries, lower token consumption, and keep context windows lean during long coding sessions. Chainguard MCP works with any MCP-compatible client, including Claude Code and Cursor, using a standard OAuth login.

Chainguard Libraries for Go and .NET

Chainguard Libraries now covers Go and .NET in early access, giving organizations malware- and greyware-scanned endpoints with a built-in, configurable cooldown, so threats are stopped before a developer, agent, or pipeline can pull a package. During the beta, our scanner returned malware verdicts on more than 5,000 Go and .NET packages, and more than 10% of those verdicts had no match in public sources at the time of detection. None of these packages ever reached a Chainguard Libraries customer.

We're also expanding CVE remediation. Java remediation is now generally available alongside Python, and JavaScript and Go remediation are launching in beta. Chainguard backports critical and high-severity fixes into the older versions that customers can't upgrade from without breaking changes, and every patched artifact is tested and reviewed by a Chainguard engineer. Finally, the first wave of Athena disclosures is live. As these disclosures go live, Chainguard Libraries customers gain access to novel patches for zero-day vulnerabilities discovered by Frontier AI models.

Looking ahead

AI has changed who writes software and how difficult it is to set up an attack. Agents need environments where they can act safely. Security teams need a way to say yes to agentic development without accepting unbounded risk. And every organization needs access to the kind of frontier vulnerability discovery that attackers will soon have on their own.

With Harness, Workspaces, Checks, MCP, and the expansion of Libraries, we're extending the trust we've built into our open source software catalog across the full lifecycle of how software is written, tested, and secured.

Chainguard Harness, Workspaces, and Checks are in beta, and waitlists are open today. Get in touch with our team to learn more.