This week, OpenAI unveiled Sora 2, the awaited successor to its generative AI video model. Sora 2 debuted as a distinct iOS app, available for free on an invite-only basis. I managed to secure a Sora 2 invite code and have been diving into the app and crafting videos ever since.

My first impressions are a mix. The technology is remarkable, and I appreciated exploring the app, but some of the content made me feel uncomfortable.

What is Sora 2?

Sora 2 is a novel video generation model and app from OpenAI, capable of producing videos with dialogue and audio driven by natural language prompts. It’s the first real contender to Google’s Veo 3 AI video builder. Tools like Meta’s Llama and Grok Imagine don’t measure up, although Meta might enhance its offerings by licensing Midjourney technology.

Using Veo 3 for the first time was a pivotal experience for me. The level of realism was both captivating and disconcerting. Sora 2 gives off a similar vibe. Like Veo 3, it’s primarily utilized for viral meme content and short videos. Clips of a Golden Retriever being arrested for shoplifting steaks or an emotional support kangaroo at the airport seem light-hearted rather than sinister.

However, the potential for Sora 2 to cause harm is directly related to its fidelity. The more realistic the videos, the greater my concern about deepfakes and misinformation.

Sora 2 has superior safeguards compared to Grok

When xAI and Elon Musk launched Grok Imagine, I was appalled by the insufficient safeguards. Musk promotes xAI and Grok as alternatives to AI applications with liberal bias. Grok’s method of content moderation and safety led to sexual deepfakes.

In contrast, OpenAI has instituted improved safeguards for Sora 2. If you upload an image to inspire a video, the app rejects it if it identifies a face. To produce a video featuring a real individual, you must utilize the Cameos feature, which necessitates the person’s consent.

When I attempted to generate videos of public figures such as Taylor Swift using jailbreaking techniques, the app denied the request. This is not the blonde showgirl I envisioned.

Cameos tool is intriguing and unsettling

Cameos is the standout feature of the new Sora app. When using the app, you must choose to opt in or out of the Cameos tool, which permits your likeness to be used in videos. You can create videos of yourself and allow access to contacts, specific users, or the public.

Cameos has resulted in numerous videos featuring OpenAI CEO Sam Altman’s likeness. It’s an inventive way to tackle the deepfake dilemma by allowing users to consent to their likeness being used.

I crafted a video of myself, and it felt strange. Sora didn’t accurately capture my voice, but my likeness was spot on. Witnessing yourself say and do things you’ve never actually done is odd.

Sora 2 and Veo 3 are leaders of the pack

Few generative AI tools can produce realistic videos with dialogue and sound effects. Sora 2 accomplishes this effortlessly, much like Veo 3. We’ll conduct an in-depth comparison of Sora 2 and Veo 3 soon. For now, Sora 2 meets the expectations in ways GPT-5 did not.

The Sora app can generate videos in a variety of styles—fake police body-cam footage, ’90s TV commercials, music videos, sports broadcasts—that don’t immediately present as AI-generated content.

Sora 2 appears to tread a fine line regarding intellectual property

Mashable has reported on the battle between artists, IP holders, and the AI sector. AI companies like Meta have secured some early wins, and President Trump’s executive orders favor the AI sector.

“You can’t have a successful AI program if you have to pay for every article, book, or anything else you’ve studied,” Trump stated while announcing The White House’s AI Action Plan.

Numerous artists and rights holders disagree. Disney’s legal team referred to Midjourney as a “bottomless pit of plagiarism” in its lawsuit against the AI firm. However, for the time being, the federal government seems to allow companies like OpenAI to utilize IP without restrictions.

Deepfakes and memes stretch as far as the eye can see

Recurring motifs populate the Sora app. Users are turning the speeches of Martin Luther King Jr. and JFK into video memes. I also encountered many SpongeBob memes and variations on Rick & Morty and Pokémon, likely because millennials cherish these shows.

Public figures and iconic TV characters have always been targets for memes. What’s new is the ability to create videos featuring these figures saying anything you desire. Google Veo 3 is more cautious with prompts involving IP.

This could pose risks. With Elon Musk spearheading a Netflix boycott over transgender characters in children’s TV shows, anyone can produce a realistic video with characters saying whatever they want. It could serve as a potent tool for outrage cultivation.

It’s time to enhance media literacy and learn how to recognize viral AI videos.

I reached out to OpenAI for comments.