diff --git a/.changeset/caip-parts-strict-validation.md b/.changeset/caip-parts-strict-validation.md
new file mode 100644
index 0000000..28c119a
--- /dev/null
+++ b/.changeset/caip-parts-strict-validation.md
@@ -0,0 +1,27 @@
+---
+"@agentcommercekit/caip": patch
+---
+
+Fix: `caip2Parts` and `caip10Parts` now validate the full input string
+instead of silently truncating it.
+
+Both functions parsed their input with `caip.split(":")` and destructured
+the first two (or three) resulting segments, only checking that those
+segments were non-empty. `split` does not enforce an upper bound on the
+number of segments, so a malformed string with extra trailing `:segments`
+(e.g. `"eip155:1:extra"` for `caip2Parts`, or
+`"eip155:1:0xabc...:extra"` for `caip10Parts`) would parse "successfully",
+silently discarding everything after the expected number of colons instead
+of being rejected.
+
+`caip10Parts` is used by `@agentcommercekit/did`'s `did:pkh` method
+(`createVerificationMethod`, reached via
+`createDidPkhDocumentFromCaip10AccountId`) to build a DID document's
+verification method from a caller-supplied CAIP-10 account ID. That call
+path does not otherwise re-validate the account ID against the CAIP-10
+pattern, so a malformed account ID could previously produce a DID document
+with silently truncated/incorrect account data instead of throwing.
+
+Both functions now test the input against their full `RegExp` pattern
+(`caip2ChainIdRegex` / `caip10AccountIdRegex`) before parsing, so malformed
+input is rejected up front.
diff --git a/packages/caip/src/caips/caip-10.test.ts b/packages/caip/src/caips/caip-10.test.ts
index 7e688ea..7ca6833 100644
--- a/packages/caip/src/caips/caip-10.test.ts
+++ b/packages/caip/src/caips/caip-10.test.ts
@@ -1,6 +1,6 @@
 import { describe, expect, it } from "vitest"
 
-import { createCaip10AccountId } from "./index"
+import { caip10Parts, createCaip10AccountId } from "./index"
 
 describe("createCaip10AccountId", () => {
   it("creates a caip 10 account ID for EVM address", () => {
@@ -37,3 +37,44 @@ describe("createCaip10AccountId", () => {
     )
   })
 })
+
+describe("caip10Parts", () => {
+  it("parses a valid EVM account ID", () => {
+    const result = caip10Parts(
+      "eip155:1:0x1234567890123456789012345678901234567890",
+    )
+    expect(result).toEqual({
+      namespace: "eip155",
+      reference: "1",
+      accountId: "0x1234567890123456789012345678901234567890",
+    })
+  })
+
+  it("parses a valid Solana account ID", () => {
+    const result = caip10Parts(
+      "solana:5eykt4UsFv8P8NJdTREpY1vzqKqZKvdp:FNoGHiv7DKPLXHfuhiEWpJ8qYitawGkuaYwfYkuvFk1P",
+    )
+    expect(result).toEqual({
+      namespace: "solana",
+      reference: "5eykt4UsFv8P8NJdTREpY1vzqKqZKvdp",
+      accountId: "FNoGHiv7DKPLXHfuhiEWpJ8qYitawGkuaYwfYkuvFk1P",
+    })
+  })
+
+  it("throws for an account ID with a trailing extra colon segment", () => {
+    // Regression test: a naive `split(":")` would silently drop everything
+    // after the third colon and happily return a truncated
+    // { namespace, reference, accountId } for this malformed input instead
+    // of rejecting it.
+    expect(() =>
+      caip10Parts("eip155:1:0x1234567890123456789012345678901234567890:extra"),
+    ).toThrow("Invalid CAIP-10 account ID")
+  })
+
+  it("throws for a string missing the account address", () => {
+    // oxlint-disable-next-line typescript/no-unsafe-type-assertion -- intentionally passing invalid input to exercise runtime validation
+    expect(() => caip10Parts("eip155:1" as never)).toThrow(
+      "Invalid CAIP-10 account ID",
+    )
+  })
+})
diff --git a/packages/caip/src/caips/caip-10.ts b/packages/caip/src/caips/caip-10.ts
index 12db7ab..2e87e8b 100644
--- a/packages/caip/src/caips/caip-10.ts
+++ b/packages/caip/src/caips/caip-10.ts
@@ -48,6 +48,15 @@ export function createCaip10AccountId(
 }
 
 export function caip10Parts(caip: Caip10AccountId): Caip10AccountIdParts {
+  // Validate the full string against the CAIP-10 pattern rather than just
+  // splitting on ":". A naive split silently drops any segments after the
+  // third colon (e.g. "eip155:1:0xabc:extra" would parse as if it were
+  // just "eip155:1:0xabc"), so malformed input could pass through
+  // undetected instead of being rejected.
+  if (!caip10AccountIdRegex.test(caip)) {
+    throw new Error("Invalid CAIP-10 account ID")
+  }
+
   const [namespace, reference, accountId] = caip.split(":")
   if (!namespace || !reference || !accountId) {
     throw new Error("Invalid CAIP-10 account ID")
diff --git a/packages/caip/src/caips/caip-2.test.ts b/packages/caip/src/caips/caip-2.test.ts
index ea5ca71..8a21dbb 100644
--- a/packages/caip/src/caips/caip-2.test.ts
+++ b/packages/caip/src/caips/caip-2.test.ts
@@ -41,6 +41,15 @@ describe("caip2Parts", () => {
       "Invalid CAIP-2 chain ID",
     )
   })
+
+  it("throws for a chain ID with a trailing extra colon segment", () => {
+    // Regression test: a naive `split(":")` would silently drop everything
+    // after the second colon and happily return { namespace: "eip155",
+    // reference: "1" } for this malformed input instead of rejecting it.
+    expect(() => caip2Parts("eip155:1:extra" as `${string}:${string}`)).toThrow(
+      "Invalid CAIP-2 chain ID",
+    )
+  })
 })
 
 describe("caip2ChainIdRegex", () => {
diff --git a/packages/caip/src/caips/caip-2.ts b/packages/caip/src/caips/caip-2.ts
index a943ed7..4a7b295 100644
--- a/packages/caip/src/caips/caip-2.ts
+++ b/packages/caip/src/caips/caip-2.ts
@@ -38,6 +38,15 @@ export const caip2ChainIds = {
 } as const
 
 export function caip2Parts(caip: Caip2ChainId): Caip2ChainIdParts {
+  // Validate the full string against the CAIP-2 pattern rather than just
+  // splitting on ":". A naive split silently drops any segments after the
+  // second colon (e.g. "eip155:1:extra" would parse as if it were just
+  // "eip155:1"), so malformed input could pass through undetected instead
+  // of being rejected.
+  if (!caip2ChainIdRegex.test(caip)) {
+    throw new Error("Invalid CAIP-2 chain ID")
+  }
+
   const [namespace, reference] = caip.split(":")
 
   if (!namespace || !reference) {