[Bug]: bitwise_and / bitwise_or / bitwise_xor return incorrect values for int8/uint8 when broadcasting

[goyaladitya05]

OpenVINO Version

2026.0

Operating System

Windows System

Device used for inference

CPU

Framework

None

Model used

N/A

Issue description

BitwiseAnd, BitwiseOr, and BitwiseXor silently produce incorrect results for int8/uint8 operands when one operand is broadcast to match the other's shape. Same-shape operations always produce correct results. No exception or warning is raised. The op executes normally but returns wrong values.

Expected: element-wise result identical to NumPy's np.bitwise_and / np.bitwise_or / np.bitwise_xor.

The issue is with Python API (openvino.opset15).

Step-by-step reproduction

import numpy as np
import openvino.opset15 as ov_opset
from openvino.runtime import Model, compile_model

def run(x_np, m_np, label):
    x = ov_opset.constant(x_np)
    m = ov_opset.constant(m_np)
    node = ov_opset.bitwise_and(x.output(0), m.output(0))
    result = compile_model(Model([node.output(0)], []), "CPU")({})[0]
    expected = np.bitwise_and(x_np, m_np)
    ok = np.array_equal(result, expected)
    print(f"{label}: {'OK' if ok else f'FAIL — {(result != expected).sum()}/{result.size} wrong'}")

x = np.arange(64, dtype=np.int8).reshape(1, 64)
run(x, np.full((1, 64), np.int8(0x0F)), "same-shape (1,64)")   # OK
run(x, np.array(np.int8(0x0F)),          "(1,64) vs scalar")    # FAIL
run(x, np.array([[np.int8(0x0F)]]),      "(1,64) vs (1,1)")     # FAIL

x2 = np.arange(32 * 256, dtype=np.int8).reshape(32, 256)
run(x2, np.full((32, 256), np.int8(0x0F)), "same-shape (32,256)")  # OK
run(x2, np.array(np.int8(0x0F)),            "(32,256) vs scalar")   # FAIL

Relevant log output

same-shape (1,64): OK
(1,64) vs scalar:  FAIL — 48/64 wrong
(1,64) vs (1,1):   FAIL — 48/64 wrong
same-shape (32,256): OK
(32,256) vs scalar:  FAIL — 6144/8192 wrong

Issue submission checklist

• I'm reporting an issue. It's not a question.
• I checked the problem with the documentation, FAQ, open issues, Stack Overflow, etc., and have not found a solution.
• There is reproducer code and related data files such as images, videos, models, etc.

[goyaladitya05]

I'll be submitting a PR for this soon.