CAPTCHAs vs Arcjet

How CAPTCHAs work

Looking for a CAPTCHA alternative? Cloudflare Turnstile, Google reCAPTCHA, and hCaptcha challenge a browser form. Arcjet advanced signals detect automation without a widget, layered on server-side bot detection.

They follow the same pattern: a client-side widget runs a challenge, the form submits a token, and the server calls the provider to verify it. If verification passes, then the server allows the request.

Turnstile offers Managed (checkbox only if the visitor looks risky), Non-interactive (spinner while challenges run), and Invisible (no UI). Detection uses proof-of-work, proof-of-space, web API probing, and browser-quirk analysis. Tokens expire after 300 seconds, and you can validate each token only once.

reCAPTCHA v2 is a checkbox or image challenge. v3 is invisible and returns a risk score that you have to turn into a policy. hCaptcha is similar to reCAPTCHA, independent of Google, with visible and invisible modes. Enterprise tiers add bot scoring.

CAPTCHAs are a browser mechanism. They can't protect pure API endpoints, server-to-server calls, native mobile backends without a native CAPTCHA SDK, or any route that can't embed a widget.

How Arcjet advanced bot signals work

Arcjet's answer to the same form-spam problem is advanced client signals, layered on server-side bot detection in your request handlers, not a challenge widget. There's no puzzle and no synchronous verification call on submit.

A small script loads a WebAssembly module on a page that leads into a sensitive route. The module collects environment signals (headless markers, automation fingerprints, browser-quirk anomalies, timing) and stores a continue token in a cookie. On the next server-side request, bot detection reads that cookie alongside user-agent classification, IP reputation, and verified-bot lists. The cookie persists across navigations.

That shape has a few consequences:

• One bot decision. Arcjet can deny a request because signals look automated, because the user agent matches a known scraper, or because the IP has a poor reputation. The layers compose into one bot decision.
• No browser, no cookie. A missing signals cookie is itself a signal. A filter rule can deny clients that never loaded the script.
• Inspectable. Denials distinguish a known bot, a missing-cookie filter, and a signals failure, so your response can differ by failure mode.
• Safe rollout. Live or dry-run applies to signals, bot detection, and the missing-cookie filter. Operational HTTP rules for bots, Shield, filters, and fixed-window or sliding-window rate limits change in the Arcjet dashboard or through MCP and take effect immediately, still inside the application.
• Predictable billing. Arcjet charges for bot-detection calls, not script loads.

Bot detection, filters, and signals are HTTP controls in request handlers. Token-bucket, email, PII, and prompt injection stay in code because they need application or body context. The same SDK can also check tool calls and jobs.

Comparison

AI agent runtime: forms vs the tool path

A CAPTCHA protects a browser-rendered form. That's inbound bot management: stop automated clients that submit a signup or target /login. It isn't a control on the agents that you run.

Arcjet bot detection, filters, Shield WAF, email, and IP checks run in your request handlers. Token-spend budgets, prompt-injection detection, and in-process sensitive-info run on tool handlers, MCP servers, and queue consumers, and you pass the input directly. Sensitive-info detection uses a built-in local engine, plus an optional on-device ML model for names, addresses, and government or financial identifiers. Bot detection is an HTTP control in request handlers.

Labeled tool-call policies let a security team change the policy for a labeled action without changing the tool; that's separate from site-wide HTTP remote rules. The SDK names those two surfaces protect() (HTTP request handlers) and guard() (tool calls and jobs).

When a security check can't finish (a timeout, a transport issue, or an incomplete policy), you choose what happens next. A direct Guard call returns allow with error codes rather than treating the incomplete check as a denial. Vercel AI SDK and LangChain wrappers fail closed unless you opt into continuing on error. HTTP request checks can fail open when your application can't reach the Cloud API; that behavior is configurable too. For more information about the error codes, see the Guards reference.

A buyer question for any runtime control: can it use prior actions in a run when deciding the next one? A correlation ID ties Arcjet decisions to a run so that the sequence is reconstructable for investigation.

If the thing that you're protecting is an MCP tool or a background job, then a CAPTCHA can't see it. If the thing that you're protecting is a public form, then signals plus server-side bot detection are the Arcjet alternative to embedding a widget.

When to use which

Use Arcjet advanced signals if

You need to protect API endpoints as well as forms; you want no visible widget; you want the cookie to persist across requests; you want layered detection in one bot decision; you want dry-run before enforcement; you want inspectable reasons; or you also need rate limiting, Shield WAF, and tool-path rules in the same SDK.

Use a CAPTCHA if

You need a free, client-side-only starting point for form spam; you want your users to see an explicit verification step (account recovery, high-value confirmation); or every protected action goes through a browser form and you have no API surface.

They can run together. A common split is a visible CAPTCHA on account recovery, with Arcjet on APIs, rate limits, and the tool path.

Compare: Cloudflare vs Arcjet · Vercel BotID vs Arcjet · All comparisons