Exploring the complexities of surveillance and identity in our digitally mediated world through speculative reconstruction of identities from unreadable, obscured, and fragmented images. Utilizing a speculative forensic technique with generative AI infills and extrapolations, Perp traverses the spaces between seen and unseen, known and unknown, reconstructing identities from obscured images—ranging from crime watch CCTV footage to the intimate perspectives of Ring cameras at the thresholds of personal spaces, and state-censored imagery, including the highly redacted 2016 Department of Defense release documenting detainee abuse in Iraq and Afghanistan—to reveal hybrid forms of us all. This project exposes the biases embedded in both AI training datasets and surveillance technologies, reflecting on the limitations and implications of both. The term "perpetrator" (perp) highlights the blurred lines between guilt and innocence in a surveillance-laden society. By suggesting that anyone can be labeled a "perp" in this surveilled landscape, Perp inverts the legal maxim, underscoring a shift towards an environment of suspicion where we are all guilty until proven innocent, and identities are fabricated when they cannot be found.