diff --git a/src/resources/cache/hashes.json b/src/resources/cache/hashes.json
index 8c678b1a15ef..98e90430f0b8 100644
--- a/src/resources/cache/hashes.json
+++ b/src/resources/cache/hashes.json
@@ -29,7 +29,7 @@
   "webgpu/shader/execution/binary/f32_multiplication.bin": "40527d93",
   "webgpu/shader/execution/binary/f32_remainder.bin": "d8738967",
   "webgpu/shader/execution/binary/f32_subtraction.bin": "f5453a8",
-  "webgpu/shader/execution/binary/i32_arithmetic.bin": "37440f39",
+  "webgpu/shader/execution/binary/i32_arithmetic.bin": "7fb5b26a",
   "webgpu/shader/execution/binary/i32_comparison.bin": "9679ab05",
   "webgpu/shader/execution/binary/u32_arithmetic.bin": "c9fe0078",
   "webgpu/shader/execution/binary/u32_comparison.bin": "7d0ba49",
diff --git a/src/resources/cache/webgpu/shader/execution/binary/i32_arithmetic.bin b/src/resources/cache/webgpu/shader/execution/binary/i32_arithmetic.bin
index bacd4c0c54ea..381ff7b6d63d 100644
Binary files a/src/resources/cache/webgpu/shader/execution/binary/i32_arithmetic.bin and b/src/resources/cache/webgpu/shader/execution/binary/i32_arithmetic.bin differ
diff --git a/src/webgpu/shader/execution/expression/binary/i32_arithmetic.cache.ts b/src/webgpu/shader/execution/expression/binary/i32_arithmetic.cache.ts
index ca7ca879f52a..2ec09770a2d9 100644
--- a/src/webgpu/shader/execution/expression/binary/i32_arithmetic.cache.ts
+++ b/src/webgpu/shader/execution/expression/binary/i32_arithmetic.cache.ts
@@ -81,6 +81,26 @@ export const d = makeCaseCache('binary/i32_arithmetic', {
   remainder_const: () => {
     return generateBinaryToI32Cases(sparseI32Range(), sparseI32Range(), i32_remainder_const);
   },
+  // Negative dividends across non-power-of-two moduli. The default sparse i32
+  // range only exercises the moduli 10 and 256, so it misses value-dependent
+  // codegen paths (e.g. modulus 768) where some implementations compute signed
+  // `%` as if it were unsigned (e.g. lowering to a poison-prone `OpSRem`),
+  // yielding 255 instead of -1 for `-1 % 768`. See
+  // https://github.com/gfx-rs/wgpu/issues/8191.
+  remainder_negative_non_const: () => {
+    return generateBinaryToI32Cases(
+      [-1, -2, -5, -255, -256, -768, -769, -1000, -1234567],
+      [3, 7, 100, 256, 768, 1000],
+      i32_remainder_non_const
+    );
+  },
+  remainder_negative_const: () => {
+    return generateBinaryToI32Cases(
+      [-1, -2, -5, -255, -256, -768, -769, -1000, -1234567],
+      [3, 7, 100, 256, 768, 1000],
+      i32_remainder_const
+    );
+  },
   addition_scalar_vector2: () => {
     return generateI32VectorBinaryToVectorCases(sparseI32Range(), vectorI32Range(2), i32_add);
   },
diff --git a/src/webgpu/shader/execution/expression/binary/i32_arithmetic.spec.ts b/src/webgpu/shader/execution/expression/binary/i32_arithmetic.spec.ts
index 05eb8ddc10c6..2caba3f586b0 100644
--- a/src/webgpu/shader/execution/expression/binary/i32_arithmetic.spec.ts
+++ b/src/webgpu/shader/execution/expression/binary/i32_arithmetic.spec.ts
@@ -170,6 +170,28 @@ Expression: x %= y
     await run(t, compoundBinary('%='), [Type.i32, Type.i32], Type.i32, t.params, cases);
   });
 
+g.test('remainder_negative')
+  .specURL('https://www.w3.org/TR/WGSL/#arithmetic-expr')
+  .desc(
+    `
+Expression: x % y, with negative dividends across non-power-of-two moduli.
+
+Regression coverage for implementations that compute signed remainder as
+unsigned for negative operands (e.g. lowering to a poison-prone OpSRem without
+VK_KHR_maintenance8), which returns 255 instead of -1 for -1 % 768.
+See https://github.com/gfx-rs/wgpu/issues/8191.
+`
+  )
+  .params(u =>
+    u.combine('inputSource', allInputSources).combine('vectorize', [undefined, 2, 3, 4] as const)
+  )
+  .fn(async t => {
+    const cases = await d.get(
+      t.params.inputSource === 'const' ? 'remainder_negative_const' : 'remainder_negative_non_const'
+    );
+    await run(t, binary('%'), [Type.i32, Type.i32], Type.i32, t.params, cases);
+  });
+
 g.test('addition_scalar_vector')
   .specURL('https://www.w3.org/TR/WGSL/#arithmetic-expr')
   .desc(