Out-of-bounds read in PyTorch - #VU163099
Published: September 29, 2025 / Updated: November 9, 2026
Vulnerability details
The vulnerability allows an unauthenticated remote attacker to crash a PyTorch application and cause a denial of service.
The vulnerability exists due to missing bounds checks on attacker-controlled offsets in PyTorch's FlatBuffer parser when torch::load processes input bearing the FlatBuffer format signature. An unauthenticated remote attacker can supply crafted data to an application that loads untrusted input, triggering an out-of-bounds memory read and terminating the process.