Chainguard, a US software security company best known for supplying hardened open-source software, announced three new products on 14 October 2026 aimed at teams that let AI agents write code. Two of them, Workspaces and Checks, are available now in beta. The third, Harness, uses Anthropic’s Claude Mythos 5.1 model to find security flaws in a company’s own code.
Chainguard’s claims about what the products do and how well they work come from its own announcements and have not been independently tested.
The problem Chainguard is pitching
AI coding agents can now run for hours on jobs such as moving code to a new system or fixing many bugs at once. In its Workspaces release, Chainguard argues that most agents run with “the same access as the developer who launched them, right next to their credentials, cloud keys, and production systems.” That means a tricked or faulty agent could leak passwords or damage live systems.
Workspaces: a sealed room for each agent
Chainguard describes Workspaces as “hardened microVMs that give every developer and coding agent their own secure environment, with no access to sensitive credentials and optional egress policies that block any connection a team has not approved.”
A microVM is a small, lightweight virtual computer, walled off from the machine it runs on. “Egress policies” control which outside sites a program may connect to. Chainguard says real passwords and keys are injected from outside the workspace, so the agent never holds them, and that every network connection is logged.
The workspaces run remotely but stay in sync with the files on a developer’s own computer. When an agent stops to wait for a person’s input, Chainguard says, “the Workspace stays active for 10 minutes before automatically suspending to a snapshot”, which saves computing costs.
“Every agent gets its own hardened, isolated machine with nothing in it worth stealing, so you can let it run for hours and trust what comes out the other end,” said Dan Lorenc, Chainguard’s chief executive and co-founder.
Checks: testing while the agent works
Continuous integration (CI) is the automated testing that runs when developers submit code. Chainguard argues that agents write code so quickly that these test queues have become a bottleneck.
Checks is what Chainguard calls “a new continuous verification engine optimized to bring continuous integration (CI) into agentic development loops.” In plain terms, the agent can run the tests itself while it is still working. Each passing run produces a signed record, or attestation, that travels with the code. Chainguard says “Merge gates can verify the passing attestation and merge without running tests again.”
“Agents shouldn’t just write code faster. They should be able to prove it works,” Lorenc said.
The release quotes Brad McCoy of Australian payments firm Cuscal, which Chainguard calls an early design partner, and analyst James Governor of RedMonk, who said existing CI tools are “struggling under the massive increase in load generated by agentic software development.”
Harness: AI bug-hunting, checked by people
Harness is described as “a new cyber harness that uses frontier AI models like Anthropic’s Claude Mythos 5.1 to find and fix exploitable vulnerabilities in an organization’s application code.”
Chainguard says Harness builds the customer’s application inside an isolated microVM, has the AI model probe a live, running copy for weaknesses, and replays each finding until it reproduces reliably before offering a fix. The company says “A Chainguard engineer reviews and approves every fix and report before it reaches a customer.” At launch, Chainguard’s own engineers will run Harness for customers rather than customers running it themselves.
Harness can be bought through Anthropic’s Claude Marketplace, so companies can pay for it from their existing Anthropic spending. Chainguard says it is available through a beta waitlist.
“Every time we’ve pointed Harness at code, it has found serious, exploitable vulnerabilities,” Lorenc said. Chainguard did not publish figures to support that.
What this means
- Security companies are building around coding agents. The pitch is to contain agents, rather than trust them to follow rules.
- Two products are in beta and one has a waitlist. No prices were announced.
- The results are Chainguard’s word. None of the speed, cost or security claims have been independently tested.
Sources
- Chainguard via PRNewswire, “Introducing Chainguard Workspaces: Hardened, Hardware-Isolated Environments for Faster Agentic Development”, 14 October 2026. https://www.prnewswire.com/news-releases/introducing-chainguard-workspaces-hardened-hardware-isolated-environments-for-faster-agentic-development-302907716.html
- Chainguard via PRNewswire, “Introducing Chainguard Checks: Reinventing Continuous Integration for the Agentic Era”, 14 October 2026. https://www.prnewswire.com/news-releases/introducing-chainguard-checks-reinventing-continuous-integration-for-the-agentic-era-302907718.html
- Chainguard via PRNewswire, “Introducing Chainguard Harness: Proven Vulnerabilities and Ready-to-Merge Fixes From Frontier AI, No Model Access Required”, 14 October 2026. https://www.prnewswire.com/news-releases/introducing-chainguard-harness-proven-vulnerabilities-and-ready-to-merge-fixes-from-frontier-ai-no-model-access-required-302907725.html
Related stories
- AIR Security Says a Crafted Certificate Can Carry Secrets Past Vercel Sandbox’s Domain Allowlist
- Pine AI Launches Pine Computer, a Sandboxed Cloud Computer for AI Agents, in Private Beta
- Anthropic’s Cyber Verification Program: Three Tiers of Looser Guardrails for Vetted Defenders
- AI Agent Security: A Hijacked SDK Goes After Agent Keys as Copilot, Bitdefender and Citadel Add Guardrails
