diff --git a/server/verifyTokenAndRateLimit.test.ts b/server/verifyTokenAndRateLimit.test.ts
index dc5ca7aa..1fc9ac08 100644
--- a/server/verifyTokenAndRateLimit.test.ts
+++ b/server/verifyTokenAndRateLimit.test.ts
@@ -105,6 +105,41 @@ describe("verifyTokenAndRateLimit", () => {
     expect(result.error).toBe("Too many requests.");
   });
 
+  it("should rate-limit invalid tokens instead of skipping the limiter", async () => {
+    mockArgon2VerifyResult = false;
+    mockRateLimiterShouldFail = true;
+    vi.resetModules();
+    const { verifyTokenAndRateLimit } = await import(
+      "./verifyTokenAndRateLimit"
+    );
+    const result = await verifyTokenAndRateLimit("invalid-token");
+    expect(result.statusCode).toBe(429);
+    expect(result.error).toBe("Too many requests.");
+  });
+
+  it("should not run argon2 verification when the request is already rate limited", async () => {
+    mockRateLimiterShouldFail = true;
+    vi.resetModules();
+    const { verifyTokenAndRateLimit } = await import(
+      "./verifyTokenAndRateLimit"
+    );
+    const hashWasm = await import("hash-wasm");
+    const result = await verifyTokenAndRateLimit("some-token");
+    expect(result.statusCode).toBe(429);
+    expect(hashWasm.argon2Verify).not.toHaveBeenCalled();
+  });
+
+  it("should rate-limit requests with a missing token", async () => {
+    mockRateLimiterShouldFail = true;
+    vi.resetModules();
+    const { verifyTokenAndRateLimit } = await import(
+      "./verifyTokenAndRateLimit"
+    );
+    const result = await verifyTokenAndRateLimit(null);
+    expect(result.statusCode).toBe(429);
+    expect(result.error).toBe("Too many requests.");
+  });
+
   it("should key rate limiter on the socket address by default (untrusted proxy)", async () => {
     mockRateLimiterShouldFail = false;
     vi.resetModules();
diff --git a/server/verifyTokenAndRateLimit.ts b/server/verifyTokenAndRateLimit.ts
index 1335f7fe..2a6d5eef 100644
--- a/server/verifyTokenAndRateLimit.ts
+++ b/server/verifyTokenAndRateLimit.ts
@@ -59,6 +59,22 @@ export async function verifyTokenAndRateLimit(
   statusCode?: number;
   error?: string;
 }> {
+  // Rate-limit before anything else. An invalid or missing token used to skip
+  // the limiter entirely, and every rejected request still pays for a full
+  // argon2 verification, which is expensive enough to be a DoS lever on a
+  // publicly reachable instance: a caller could send an endless stream of
+  // bogus tokens and pin the server's CPU without ever hitting the limiter.
+  const rateLimitKey = request ? getClientIp(request) : (token ?? "anonymous");
+  try {
+    await rateLimiter.consume(rateLimitKey);
+  } catch {
+    return {
+      isAuthorized: false,
+      statusCode: 429,
+      error: "Too many requests.",
+    };
+  }
+
   if (!token) {
     return {
       isAuthorized: false,
@@ -91,17 +107,5 @@ export async function verifyTokenAndRateLimit(
   // Records a new session or refreshes an active one's last-seen time.
   addVerifiedToken(token);
 
-  const rateLimitKey = request ? getClientIp(request) : token;
-
-  try {
-    await rateLimiter.consume(rateLimitKey);
-  } catch {
-    return {
-      isAuthorized: false,
-      statusCode: 429,
-      error: "Too many requests.",
-    };
-  }
-
   return { isAuthorized: true };
 }