OpenAI Under Fire as Sora 2 Sparks Deepfake, Harassment Alarms

OpenAI’s Sora 2 is under scrutiny for enabling lifelike fakes and digital harassment, which watchdogs claim threaten online privacy and trust. Critics called it another rushed release without the guardrails to match its power.

In a letter, Public Citizen called on OpenAI to pull Sora 2 from public access, describing the release as “reckless and dangerous.” The group warned that the model’s unchecked realism endangers privacy and democratic stability by making it nearly impossible to distinguish fake, unauthorized videos from the real thing.

Unsafe by design

Public Citizen stated that OpenAI released Sora 2 in a rushed and unsafe manner — part of a consistent, dangerous pattern. The group pressed CEO Sam Altman to adopt a “measured, ethical, and transparent pre-deployment process” with clear guardrails before any public launch.

The watchdog also sent its letter to Congress, escalating the dispute beyond a company policy fight into a public accountability push ahead of high-stakes elections. 

A democracy risk wrapped in viral content

Public Citizen’s warning extended far beyond copyright or celebrity likeness disputes. The group pointed to a wave of nonconsensual and fetishized videos created with Sora 2, saying the AI tool has already become a vehicle for digital harassment, especially against women. 

Researchers and users have flagged clips depicting women in degrading or violent contexts, with watchdogs noting that Sora’s filters appear easily bypassed.

Speaking to The Associated Press, Public Citizen’s J.B. Branch said the “biggest concern is the potential threat to democracy,” warning that the technology could erode public trust in what people see online. “We’re entering a world in which people can’t really trust what they see,” Branch told the outlet.

When watermarks don’t leave a mark

Researchers and watchdogs said Sora 2’s safety systems proved weak soon after launch. 

Anti-impersonation filters were reportedly bypassed within 24 hours, and the model’s “mandatory” watermarks can be easily removed using basic editing tools. Public Citizen also noted inconsistencies in the copyright “opt-out” process, which allowed protected images and celebrity likenesses to appear without authorization.

Estates of several deceased public figures have lodged complaints over such uses, adding to broader fears that the model could amplify AI-driven misinformation. Advocates warned that without stronger guardrails, Sora 2 risks becoming a ready tool for political deepfakes and other deceptive content as the 2026 elections near.

Listening isn’t enough

OpenAI has yet to directly respond to Public Citizen’s demand for the withdrawal of Sora 2. However, the company has stated that it is “listening to feedback” and working with studios and rights holders to improve safeguards. 

In recent weeks, it has announced talks with entertainment and creative industry groups, including those in Japan, to prevent the misuse of well-known characters and copyrighted material.

Advocates said those steps fall short of what’s needed. They argued that OpenAI’s pattern of launching first and correcting later shows how fast innovation can outpace responsibility. Public Citizen and other watchdogs viewed Sora 2 as a reminder that once realism becomes frictionless, the damage to truth and trust may be just as hard to reverse.

Morgan Freeman spoke out against AI systems copying his voice for online videos without his approval.