Stalker uses Sora 2 to harass technology journalist
A stalker used OpenAI's Sora 2 video generator to create and distribute deepfake nude videos of technology journalist Taylor Lorenz without consent, reportedly operating hundreds of accounts to amplify harassment. OpenAI's weak content safeguards and moderation failures enabled the videos to be generated, though Lorenz was able to remove them from the platform.
What happened
A stalker used OpenAI's Sora 2 video generator to create and distribute deepfake nude videos of technology journalist Taylor Lorenz without her consent. The perpetrator also operated numerous online accounts dedicated to Lorenz to amplify the harassment.
Lorenz reported that she was able to block and delete unapproved videos containing her image from the app, though the stalker may have already downloaded the videos. According to OpenAI's system card, Sora 2 failed to block prompts for generating videos with nudity or sexual content using a real person's likeness 1.6 percent of the time.
Reported impact
Lorenz stated it was "scary to think what AI is doing to feed my stalker's delusions." She described the stalker as having hired photographers to surveil her, shown up at events she attends, and impersonated her friends and family members online to gather information.
Index analysis
Sora 2's weak content moderation safeguards enabled the stalker to bypass restrictions and produce inappropriate deepfake videos. The paucity of safeguards was compounded by poor visibility of OpenAI's complaint and appeal mechanisms.
Governance controls
OpenAI claims that the app blocks users from uploading photos with faces, though it is unclear how the stalker was able to bypass this restriction to create deepfake videos of Lorenz.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.