US watchdog group urges OpenAI to pull Sora over deepfake risks

US watchdog Public Citizen has called on OpenAI to withdraw its AI video app Sora 2, citing dangers including deepfakes, nonconsensual images, and reckless disregard for product safety.

Public Citizen, a nonprofit advocacy group, said in a letter to OpenAI and CEO Sam Altman that Sora 2’s hasty release ahead of competitors lacked safety measures, and could threaten people’s rights to their likeness and the stability of democracy.

The group also sent its letter to the US Congress.

Concerns have been raised about Sora 2’s potential to generate nonconsensual images and deepfakes, even as OpenAI blocks nudity, and moderates public figure depictions.

OpenAI has not responded to requests for comment as of November 11, 2025.

🔗 Source: Associated Press

🧠 Food for thought

Implications, context, and why it matters.

Sora 2 fakes videos 80% in tests on known false claims

NewsGuard tested 20 false claims with Sora 2, and it produced misleading videos in 16 cases, including five tied to past Russian disinformation 1.
Examples included a Moldovan election official destroying pro-Russian ballots and a toddler detained by U.S. immigration officers, which never happened, yet viewers could be fooled 1.
OpenAI’s safety filters act inconsistently, so the tool blocks some false claims but lets others through, creating trial and error for bad actors 1.
The Sora watermark for AI content can be stripped in minutes with free tools, which weakens this safeguard 1.

US deepfake and AI laws grow, raising pressure on video platforms

Half of U.S. states filed 2025 bills on AI election deepfakes, and 21 would see first restrictions 2.
California requires disclaimers on AI political ads and platform takedowns of deceptive content 3. New York’s digital replica law (restricting unauthorized use of a person’s voice or likeness in synthetic media) requires written consent, clear contracts, and pay. The state mandates disclosures for AI-generated political content 3.
AI video platforms face compliance work. They must add incident response plans, training on deepfake risks, and vendor due diligence that includes screening plus contractual controls for third-party tools or data suppliers 3.
New York and Tennessee allow civil claims for unauthorized digital replicas 3. Colorado’s AI Act, a comprehensive state AI governance law, imposes risk assessment duties starting Feb. 1, 2026, and Minnesota sets criminal penalties for non-consensual deepfakes 3.