Vercel Security Checkpoint: Attack Mode and what crawlers get
Why a crawler, monitor or test script gets the Vercel Security Checkpoint, what Attack Mode and DDoS mitigation let through, and how to tell which one fired.
By Crawlito · 9 min read
In short
- The Vercel Security Checkpoint is the page Vercel serves for any challenge, so the same screen can come from Attack Mode, automatic DDoS mitigation, Bot Protection or a custom challenge rule.
- Attack Mode is free on all plans, runs for 1, 6 or 24 hours and then switches itself off, and lets verified bots from Vercel's directory, such as Googlebot, through without a challenge.
- Clients missing from the verified bots directory, including a curl test that copies a crawler's user agent, get the checkpoint while Attack Mode is on.
- Vercel's automatic DDoS mitigation is always on and can challenge or block traffic on its own; Pro and Enterprise teams can pause it for 24 hours or add System Bypass Rules by IP.
- Vercel does not document the x-vercel-mitigated response header, so use it as a hint and confirm the source in the Firewall overview, Traffic page or the vercel firewall CLI.
On this page
- What serves the Vercel Security Checkpoint?
- How does Attack Mode work?
- What do crawlers get while Attack Mode is on?
- What does automatic DDoS mitigation do to crawlers?
- How to tell which feature showed the checkpoint
- Does a reverse proxy in front of Vercel change this?
- When should you turn Attack Mode off?
- Frequently asked questions
- Check what your Vercel site returns now
The Vercel Security Checkpoint is the page Vercel serves when its firewall decides to challenge a request. If your crawler test, uptime monitor or an AI fetcher gets that page, the challenge came from one of four places: Attack Mode, Vercel's automatic DDoS mitigation, the Bot Protection managed ruleset in challenge mode, or a custom rule with a challenge action. Attack Mode lets the verified bots in Vercel's directory through, Googlebot and the major AI search crawlers among them, so for most sites the real risk is to clients Vercel cannot verify, not to Google.
This post covers Attack Mode and automatic DDoS mitigation, and how to work out which feature produced the checkpoint. For the AI Bots ruleset, Bot Protection and how custom rules are ordered, read the Vercel AI Bots ruleset guide. For why no crawler can pass a JavaScript challenge, see why a WAF challenge page shows AI bots nothing.
What serves the Vercel Security Checkpoint?
Vercel's firewall concepts page describes the challenge action: visitors see a Vercel Security Checkpoint screen, the browser runs JavaScript to prove it is a real browser, and a passed challenge creates a session that lasts 1 hour. Non-browser clients such as bots and scripts cannot pass it. Every feature that challenges uses this same page, so the screen alone does not tell you which one fired.
| Source | When it challenges | Who skips it | How it ends |
|---|---|---|---|
| Attack Mode | Every visitor, while it is on | Verified bots, your own Functions and Cron Jobs | Switches off after 1, 6 or 24 hours, or when you disable it |
| Automatic DDoS mitigation | Traffic Vercel's signals flag as an attack | IPs in System Bypass Rules (Pro and Enterprise) | Automatic; can be paused for 24 hours |
| Bot Protection, challenge mode | Clients that do not behave like a browser | Verified bots, custom bypass rules | When you set it to Log or Off |
| Custom challenge rule | Requests matching your conditions | Earlier custom bypass rules | When you edit, disable or remove the rule |
The CLI's vercel firewall status prints the order Vercel evaluates these layers: bypass, then system mitigations, attack challenge mode, IP blocks, custom rules, and the Bot Protection, AI Bots and OWASP managed rulesets. Attack Mode and DDoS mitigation run before any custom rule you write.
How does Attack Mode work?
Attack Mode is the feature Vercel used to call Attack Challenge Mode, and the CLI and REST API still use the older name in places. Per Vercel's documentation at the time of writing:
- It is available on all plans, and users with the member role can access it.
- While it is on, visitors must complete a security challenge before they reach your site.
- Known bots, such as search engines and webhook providers, are allowed through automatically.
- Requests from your own Functions and Cron Jobs are allowed. Projects in the same Vercel account can call each other without a challenge, while other Vercel accounts cannot bypass it.
- It is free. Requests blocked by Attack Mode do not count towards your usage limits.
Vercel's April 1, 2025 changelog announced that verified webhook providers, including Stripe and PayPal, are automatically allowed and that Cron Jobs in the same account are exempt from challenges, so advice written before then may describe different behaviour.
Turning it on and choosing a duration
In the dashboard, open the project's Firewall page, select the ⋯ menu at the top right, choose Enable Attack Mode, then pick Enable for 1 Hour, or open the menu beside that button for Enable for 6 Hours or Enable for 24 Hours. Attack Mode turns off by itself when the period ends, and you can enable it again if the attack continues.
The CLI does the same thing, and the change takes effect without a publish step:
vercel firewall attack-mode enable --duration 6h --yes
vercel firewall attack-mode disable --yesThe --duration flag accepts 1h (the default), 6h or 24h. The REST endpoint is POST /v1/security/attack-mode, with projectId, attackModeEnabled and an optional numeric attackModeActiveUntil in the body.
What do crawlers get while Attack Mode is on?
Attack Mode lets through the bots listed in Vercel's verified bots directory. Vercel verifies a bot by IP range, reverse DNS or cryptographic signatures with Web Bot Auth, so the decision rests on who sent the request, not on what its user agent says. Vercel states that search engine crawlers like Googlebot pass without a challenge and that Attack Mode does not affect search indexing, even over long periods.
| Client | In the directory at the time of writing | What it gets |
|---|---|---|
| Googlebot, Bingbot, DuckDuckBot | Yes, search_engine_crawler | Your page |
| OAI-SearchBot, Claude-SearchBot, PerplexityBot | Yes, ai_assistant | Your page |
| ChatGPT-User, Claude-User, Perplexity-User | Yes, ai_assistant | Your page |
| GPTBot, ClaudeBot, CCBot | Yes, ai_crawler | Your page, unless a custom rule blocks them |
| Checkly, Better Stack, UptimeRobot, Datadog synthetics | Yes, monitor | Your page |
| A crawler or monitor not in the directory | No | The checkpoint |
| Your own curl or script with a crawler's user agent | No, it is not verified | The checkpoint |
Three practical consequences follow:
- Your tests mislead you. A
curl -A "OAI-SearchBot"from your laptop is not the verified crawler, so it is challenged even when the real OAI-SearchBot is not. Check the Firewall traffic for the real bot instead of trusting the curl result. - Training crawlers pass too. Attack Mode does not separate training from search. If you want GPTBot kept out, Vercel's advice is a custom rule that matches its user agent, and the AI Bots ruleset covered in the ruleset guide is the managed way to do it.
- Standalone APIs and unrecognized services can break. Vercel warns that standalone APIs, other backend frameworks and automated services it does not recognize may fail the challenge. Browser traffic, including a Next.js frontend calling its own API routes, works.
The directory changes over time, so look up any bot you depend on before you rely on this table.
What does automatic DDoS mitigation do to crawlers?
DDoS mitigation is always on, on every plan. Vercel says it uses hundreds of signals to fingerprint request patterns and challenges or blocks requests that look illegitimate, across L3, L4 and L7 attacks. You are not charged for traffic it blocks, but usage still accrues for requests served before a mitigation starts and for traffic not recognized as an attack, which can include bots and crawlers.
Vercel also says these system mitigations can block trusted sources such as proxies or shared networks. Two controls exist for that, both on Pro and Enterprise:
- Pause System Mitigations from the ⋯ menu on the Firewall page, or
vercel firewall system-mitigations pause --yes. It lasts 24 hours unless you resume sooner, and you pay for any attack traffic that gets through. - System Bypass Rules exempt specific IPs or CIDR ranges, optionally per domain, with
vercel firewall system-bypass add <ip>. A custom rule with a bypass action does not skip system mitigations; the CLI reference says so directly.
A System Bypass Rule only helps when you know the client's IP, which suits your own monitoring or a partner's fixed egress. It is not a sensible way to handle AI crawlers, whose operators publish their own ranges and which Vercel already verifies.
How to tell which feature showed the checkpoint
Start with the response headers. Vercel documents server: Vercel and x-vercel-id in its response headers reference, and notes that a proxy such as Cloudflare can override server. You may also see an x-vercel-mitigated header on a firewall response, but it is not listed in that reference at the time of writing, and Vercel does not document the status code of the checkpoint page, so treat both as hints, not a contract.
curl -s -o /dev/null -D - https://example.com/ | grep -iE '^HTTP|x-vercel-mitigated|x-vercel-id|^server'If server is not Vercel and there is no x-vercel-id, another layer in front probably answered, and which WAF blocked the crawler shows how to read other vendors' headers.
Then confirm the source in Vercel's own data:
- Firewall overview. The Firewall Observability page describes a Rules section that breaks traffic down by the rule that applied, an Events section for actions taken by the platform-wide firewall, which is where DDoS mitigation shows up, and an Alerts section. Vercel raises a DDoS alert when malicious traffic reaches 125,000 requests in 15 minutes.
- Traffic page. Filter by rule and by action, then group by user agent to see which clients were challenged.
- CLI. These commands answer the same questions from a terminal:
vercel firewall overview
vercel firewall traffic list --since 1h --dimension rule --dimension user-agent
vercel firewall alerts list --since 24h
vercel firewall persistent-actions listoverview reports whether attack challenge mode is on, and the CLI reference's sample traffic list output shows DDoS Mitigation as a named rule. The last 24 hours work on every team plan; how far back you can go otherwise depends on your plan's firewall data retention, which Observability Plus extends to 30 days.
| What you see | Most likely source |
|---|---|
| Every visitor challenged, Attack Mode shown as on | Attack Mode |
| An alert or a DDoS Mitigation rule entry, Attack Mode off | Automatic DDoS mitigation |
| Only non-browser clients challenged, Bot Protection set to Challenge | Bot Protection |
| Only certain paths, countries or user agents challenged | A custom rule |
| One IP challenged for a while after a single match | A persistent action on a custom rule |
Does a reverse proxy in front of Vercel change this?
Yes, and Vercel is blunt about it. The Bot Management page says Bot Protection does not work behind a reverse proxy such as Cloudflare, Azure or another CDN, because the proxy masks the signals it relies on. Vercel's reverse proxy page does not recommend a proxy at all, since the CDN loses visibility and real client IPs cannot be identified accurately. For a listed set of providers, including Cloudflare, Fastly, CloudFront and Akamai, Vercel's Verified Proxy Lite is enabled automatically on all plans and reads the client IP from the provider's header, such as CF-Connecting-IP. Some providers need setup first: CloudFront's CloudFront-Viewer-Address must be enabled in an origin request policy, and Akamai's True-Client-IP in the property manager.
Vercel does not document how Attack Mode verifies bots behind a proxy. Since verification depends partly on IP and reverse DNS, test it: if Googlebot or OAI-SearchBot shows up as challenged in the Traffic page, the proxy is the first suspect. And if both layers can challenge, check which one did before you change settings on either.
When should you turn Attack Mode off?
Vercel says to use Attack Mode mainly for highly targeted attacks, because automatic DDoS protection already runs. It is safe for search indexing, but it challenges every client Vercel cannot verify, including smaller AI crawlers, link checkers and your own scripts. Turn it off once the alert has cleared and traffic on the overview is back to normal, or let the timer end it. If one path keeps getting attacked, a custom challenge rule scoped to that path does less collateral damage than challenging the whole site.
Frequently asked questions
Does Vercel Attack Mode hurt SEO?
Vercel says no: search engine crawlers like Googlebot pass without a challenge and indexing is not affected, even over extended periods. Crawlers that are not in Vercel's verified directory are challenged.
Can ChatGPT search and Perplexity reach my site during Attack Mode?
At the time of writing, OAI-SearchBot, ChatGPT-User, PerplexityBot and Perplexity-User are in Vercel's verified bots directory, so they are let through. A request that only copies their user agent is challenged.
Why does my curl test get the checkpoint when Googlebot does not?
Vercel verifies bots by IP, reverse DNS or signature, and your request comes from your own network. Read the Firewall traffic for the real crawler instead.
Can I exempt my uptime monitor from Attack Mode?
Vercel does not document a per-client exemption for Attack Mode. Monitors in the verified directory pass; for others, rely on the timer or disable Attack Mode, and use System Bypass Rules only for DDoS mitigation.
Do requests blocked by Attack Mode or DDoS mitigation cost money?
No. Vercel says traffic blocked by Attack Mode does not count towards usage and traffic blocked by DDoS mitigation is not charged, though requests served before a mitigation starts still incur usage.
Check what your Vercel site returns now
After an attack, or after you change any of these settings, run the free check on an important page. It is a single-page check with no account, and CrawlitoBot is not in Vercel's verified bots directory at the time of writing, so the result reflects what an unverified client receives. For a whole site with evidence and prioritized fixes, an audit covers up to 500 pages.
Filed under CDN and bot protection. Topics: bot-protection, waf, nextjs, web-bot-auth, troubleshooting