Denial Of Service (DoS)

tensorflow is vulnerable to denial of service DoS attacks. When a MirrorPadGrad is given outsize input paddings, the library gives a heap out of bound error, which allows an attacker to cause tensorflow to crash...