Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchShort answer: the celebrity deepfake controversy concerned OpenAI’s Sora video generator, not ordinary ChatGPT. OpenAI described a system that blocked unauthorized depictions of public figures and allowed a person to opt in to an approved “character” representing their appearance and voice. That was a product safeguard—not a universal internet-wide opt-out—and Sora’s web and app experiences ended on April 26, 2026.
ChatGPT and Sora were different products
ChatGPT is OpenAI’s conversational and multimodal assistant. Sora was the company’s short-form video-generation service. The likeness controversy involved Sora’s realistic video and audio capabilities, including its handling of recognizable faces and voices. OpenAI’s published Sora material, rather than a general ChatGPT setting, described the relevant controls (Launching Sora responsibly).
That distinction matters because “ChatGPT deepfakes” suggests a single switch that protects a celebrity everywhere. OpenAI’s controls applied to media generated, uploaded or shared within Sora and to the company’s own moderation and reporting systems.
What “opt out” actually meant
Headlines compressed several different mechanisms into the phrase “celebrities can opt out.” They were not interchangeable.
Recommended Free Tools
#1 Best Overall
| Issue | Relevant mechanism |
|---|---|
| Unauthorized AI video of a public figure in Sora | Restrictions on real-person and public-figure generations, plus moderation and reporting |
| Authorized use of a person’s likeness | An opt-in Sora “character” controlled by the person represented |
| Removing a Sora video, profile, comment or character | In-product abuse reporting and takedown processes |
| False personal information in ChatGPT responses | A separate personal-information correction or removal process (OpenAI Help Center) |
| Using a user’s conversations to improve models | Separate data controls; not a likeness-protection switch (OpenAI data-use policy) |
OpenAI’s Sora 2 safety documentation said the system did not support text-to-video generation of public figures at launch and blocked generations involving real people unless they had consented through the likeness-control feature (Sora 2 risk mitigations; Sora 2 System Card). In practical terms, the more accurate description is unauthorized use was meant to be blocked, while authorized likeness use required opt-in consent.
How Sora’s character system worked
OpenAI described “characters” as verified representations that could include both a person’s appearance and voice (Creating with Sora safely). Its stated workflow was:
- Claim or create a character: the person completed identity checks and supplied a short video-and-audio recording to capture the likeness.
- Control access: the character owner decided who could use that representation.
- Authorize uses: other users could create videos only under the permission granted by the owner.
- Revoke permission: OpenAI said the owner could withdraw access.
- Review activity: videos containing the character—including drafts made by other users—were visible to the character owner, who could review, delete or report them (Sora 2 is here).
This was OpenAI’s stated product design, not an independently verified guarantee that every attempted impersonation would be stopped. It also described platform permission, not a talent contract, compensation arrangement or transfer of publicity rights.
What the safeguards did—and did not—cover
Living public figures
The published Sora 2 policy focused on preventing unauthorized depictions of real people and public figures. A celebrity’s consent to an authorized character did not imply consent to every use of their name, image, voice, mannerisms or signature phrases.
The Tool Desk
Outbyte Driver Updater FREEFix the driver behind crashes, sound loss and screen glitchesFind Drivers →Outbyte PC Repair FREEClear out junk files and repair common Windows errorsFree Scan →Uploaded photographs and footage
OpenAI said users had to attest that they had consent and rights before uploading images containing people. Consent to upload an old photograph is not automatically consent to animate a person, publish the result or monetize it. OpenAI described stricter safeguards for real-person imagery and minors (Creating with Sora safely).
Voice and face are related, not identical rights
The character feature covered appearance and voice, but legal rights can differ. Publicity, privacy, copyright, trademark and unfair-competition rules may apply differently to a face, a voice, a name, a costume or a fictional character.
When the policy failed in practice
SAG-AFTRA reported that outputs resembling actor Bryan Cranston’s voice and likeness appeared without consent or compensation shortly after Sora 2 launched. The union said it worked with Cranston, OpenAI and talent agencies toward an opt-in policy for voice and likeness (SAG-AFTRA statement). This is the union’s account, not a comprehensive independent technical audit.
The episode illustrates the difference between a written rule and enforcement. OpenAI could state that unauthorized generations were prohibited while users still encountered outputs that appeared to violate that rule. A consent-first design reduces risk; it does not make prevention infallible.
Deceased celebrities and estates were a separate fault line
Reports described objections from relatives or estates connected with Robin Williams and George Carlin over AI-generated videos using their likenesses (Axios). A living performer’s direct consent is not the same as an estate’s authority over a deceased person’s identity.
OpenAI’s feed-policy language addressed likenesses of living individuals without consent and also restricted some depictions of deceased public figures, while leaving context-dependent treatment for historical figures (Sora feed philosophy). Rights vary by jurisdiction: an estate’s position may depend on state publicity law, contracts, trademarks, copyright and unfair-competition doctrines. Platform policy does not settle those underlying legal questions, and there was no universal estate-wide opt-out established by the materials cited here.
Watermarks, provenance and detection were supporting layers
OpenAI described several additional safeguards: visible watermarks, C2PA provenance metadata, automated scanning of video frames, scene descriptions and audio transcripts, user reporting, enforcement and an internal reverse-video search intended to identify Sora-generated material (Launching Sora responsibly).
- Prevention tries to stop a prompt or output before publication.
- Detection looks for violations after generation.
- Provenance can help establish that a file originated in Sora.
None is equivalent to perfect protection. Watermarks can be cropped or removed, metadata can disappear during re-encoding or reposting, and detection can miss novel or adversarial outputs. A provenance signal may show where a file came from without stopping copies on other services.
Rank #4
What to do about an unauthorized Sora video
OpenAI said videos, profiles, comments and characters could be reported for abuse, but it did not publish a celebrity-specific escalation form or guaranteed response time. A rights holder’s practical evidence-preservation and reporting sequence was:
- Save the URL, downloaded file, account name, timestamps and screenshots before content changes.
- Use the product’s report or abuse function and identify the exact video, character, account or post.
- Request removal through OpenAI’s applicable support, privacy or legal channel if the in-product route fails.
- Notify an agent, union, publicist or lawyer, particularly where the content is fraudulent, threatening, sexual or defamatory.
- Report every copy separately to the service hosting it.
Removing a Sora item would not guarantee deletion of files that users had already downloaded or reposted.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.Why an OpenAI takedown could not clean up the whole internet
Sora’s controls did not govern TikTok, YouTube, Instagram, X, Telegram, private messaging groups, independent websites or rival generators. A rights holder may need each host’s impersonation or abuse process and, depending on the jurisdiction and facts, copyright, trademark, publicity-rights, privacy, fraud, harassment or nonconsensual-intimate-image remedies. Those options differ substantially by location and are not guaranteed.
The same limitation applies when a character owner revokes access: revocation can end the authorized permission and support removal within the product, but it cannot ensure that every downloaded or externally reposted copy disappears.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Best Value
Opt-in versus opt-out is a policy choice
| Model | Strengths | Trade-offs |
|---|---|---|
| Opt-in consent | Clearer permission trail; less burden on performers; lower risk of unauthorized impersonation | Requires verification and rights administration; can restrict satire, historical uses and harmless fan creativity; does not control external copies |
| Opt-out removal | Easier to operate at scale; permits more content by default; offers a formal removal route | Places the burden on the person impersonated; harmful material may appear first; difficult for estates or people without platform access; does not reach rival services |
The dispute is therefore not just a technical setting. It concerns who must obtain permission, who bears the cost of enforcement and how creative or historical uses should be balanced against personal control.
Sora’s current status
OpenAI says the Sora web and app experiences were discontinued on April 26, 2026. The Sora API was scheduled for discontinuation on September 24, 2026; as of October 1, 2026, that date has passed. Users were directed to export Sora content through sora.chatgpt.com/sunset (OpenAI Help Center).
Consequently, the likeness rules should be described as the former Sora 2 policy, not as a currently available ChatGPT control. The shutdown also means that any surviving dispute may concern archived material, downloaded files, reposts or other services rather than an active Sora feed.
The Bottom Line
OpenAI did not give celebrities a universal “ChatGPT deepfake opt-out.” Sora’s former policy moved toward consent: unauthorized public-figure generations were supposed to be blocked, while verified people could opt in to characters using their appearance and voice, control access and revoke permission. Reports of unauthorized outputs showed the safeguards were not flawless, and neither Sora’s controls nor its takedowns could govern deepfakes made or reposted elsewhere.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
