Arcjet Guards

Arcjet Guards is an Arcjet security capability whose point is to run checks inside the agent loop, not only at the edge of an app. The official introduction is https://arcjet.com/blog/introducing-arcjet-guards-security-inside-the-agent-loop . Arcjet's main site is https://arcjet.com/ and the docs are at https://docs.arcjet.com .

This radar database already has Arcjet Himalia, a family of classifiers for different settings. Guards is a different product line: according to the official post, it makes security decisions inside the agent loop, such as prompt-injection checks, tool-call authorization, and stops related to sensitive data. Exact APIs, supported runtimes, and pricing should be taken from the post and docs as they stand when you use them. The previous radar slot already verified this official page in the sitemap and marked it for this slot.

Highlights:

• Introduction: https://arcjet.com/blog/introducing-arcjet-guards-security-inside-the-agent-loop ;
• Product and docs: https://arcjet.com/ , https://docs.arcjet.com ;
• Do not collapse this term with Himalia.

Why it matters

Once an agent can call tools, change data, and reach internal systems, a single WAF-style check outside the gateway is often too late: the harmful action happens mid-loop. Arcjet uses the name Guards to turn in-loop security into a product you can point to, rather than an architecture slogan. The official blog gives the name a primary source and keeps it distinct from the Himalia models already recorded here.

If you are opening tool access for a coding agent or an internal assistant, this is a product line to remember on its own: see which step of the loop it can stop, then decide whether it replaces or sits beside your current policy engine.

Limits

How well a security product works depends on where you hook it in, how you write policies, and which cases you test. A launch blog is not a penetration-test report. This page does not treat unpublished block rates, latency numbers, or claims that it replaces an entire security stack as facts. Guards, Himalia, and other Arcjet modules may be used together; integration details belong in the docs. Not every open-source or in-house agent framework will have a ready adapter, so check runtime support before you commit.

1. Read https://arcjet.com/blog/introducing-arcjet-guards-security-inside-the-agent-loop , then use https://docs.arcjet.com for the integration path;
2. On a test agent that actually calls tools, enable only the checks the docs describe (for example prompt injection or tool authorization) and try both allowed and blocked cases inside the loop;
3. Do not treat Guards and Arcjet Himalia as the same model or the same API;
4. Before production, review missed and false blocks against your own threat model, and confirm what happens on failure — fail open or fail closed — plus how it is logged.