Everything we announced at Chainguard Assemble London 2026

The rapid integration of AI agents into the software development lifecycle has accelerated both the creation of code and the discovery of vulnerabilities. In response, a new suite of tools aims to secure the environments where AI agents and developers operate. These tools focus on isolating execution via microVMs, automating the hunt for exploitable logic flaws using frontier models like Claude Mythos 5.1, and providing verified, "hardened" dependency catalogs to prevent the injection of malware.
The strategy shifts security "left" by integrating checks into pre-commit hooks and development workspaces, ensuring that AI-generated code is verified before reaching production. While several of these offerings are in beta, the ecosystem seeks to create a closed loop of discovery and remediation, where vulnerabilities found by AI are patched by engineers and redistributed through secure repositories. The effectiveness of this approach depends on the ability of these isolated environments to accurately mirror production without introducing new attack vectors.

* Chainguard announced new products at Chainguard Assemble London 2026.
* Chainguard Harness uses Anthropic’s Claude Mythos 5.1 to hunt for exploitable vulnerabilities in isolated microVMs.
* Chainguard Harness is available via the Claude Marketplace by Anthropic.
* Chainguard Workspaces provides microVM-based development environments using zero-CVE Chainguard OS packages.
* Chainguard Checks provides pre-commit verification in isolated microVMs with signed attestations.
* Chainguard MCP consists of four servers (cg-oci, cg-apk, cg-versions, cg-api) for dependency verification.
* Chainguard Libraries added support for Go and .NET in early access.
* CVE remediation is generally available for Java and Python, and in beta for JavaScript and Go.
* Design partner Cuscal has utilized Chainguard Checks since the alpha phase.
* Harness, Workspaces, and Checks are currently in beta with open waitlists.

The strongest version of this narrative is that AI has fundamentally altered the "arms race" of cybersecurity, necessitating a shift from static scanning to dynamic, agentic defense. By providing isolated "sandboxes" for both development and vulnerability hunting, the goal is to neutralize the risks of AI-generated code and malicious packages before they enter the supply chain.
However, the persuasive weight of this narrative relies on a specific decision frame: that the only way to manage the "flood" of AI-generated vulnerabilities is through a proprietary, vertically integrated ecosystem. The framing leverages the urgency of "frontier models" and "zero-days" to position the vendor's own tools as the primary shield. This creates a loop where the vendor identifies the threat (via their own research) and provides the sole remedy.
Patterns detected: ARC-0043 Fear Appeal, ARC-0031 Authority Game
The underlying paradigm is "Security as a Service," where trust is outsourced to a curated provider rather than verified through open, decentralized standards. This echoes the historical shift from local security controls to cloud-managed security postures. The implication is a potential decrease in human developer autonomy; as agents handle more of the loop, the developer becomes a "supervisor" of snapshots rather than a primary architect of the security model.
Bridge Questions:
1. If vulnerability discovery becomes fully agentic, how does the industry define "trust" beyond a vendor's attestation?
2. What are the risks of creating a "single point of failure" by consolidating the development, testing, and patching environment under one provider?
Counterstrike Scan: A coordinated campaign would manufacture a sense of inevitable crisis ("AI is breaking security") to force rapid adoption of a specific toolset. While this content uses high-urgency language, it describes a logical expansion of a supply-chain security product; however, it aligns structurally with the "threat-then-solution" vendor playbook.