BUG: python.exe hard-crashes on numpy.linalg calls with NumPy 2.5.3 + default MKL threading on Windows (no traceback) #32570

Describe the issue:

Importing NumPy and running some linear algebra routines can terminate the interpreter process without a Python traceback. The behavior is dependent on MKL_THREADING_LAYER:

• crashes with default/unset
• crashes with OMP
• crashes with INTEL
• succeeds with TBB
• succeeds with SEQUENTIAL

Reproduce the code example:

import numpy as np
import numpy.linalg as la

A = [[1.0, 2.0],
     [3.0, 4.0]]
b = [1.0, 1.0]

print("version", np.__version__)
print("result", la.lstsq(A, b)[0])

Error message:

No Python exception is printed. python.exe exits abruptly with a native crash.

Python and NumPy Versions:

Python: 3.13.15 (packaged by conda-forge)
NumPy: 2.5.3

Runtime Environment:

OS: Windows 10 (10.0.19045), AMD64

How does this issue affect you or how did you find it:

No response