MIT Technology Review · Feb 15, 2024
MIT Technology Review reports on OpenAI’s Sora video model and its risks
Markdown · JSON · Original source
MIT Technology Review published this account on February 15, 2024. It reports that ’s Sora turns short text descriptions into detailed video clips and says the model could generate clips up to a minute long. The article describes detailed sample scenes but also visible flaws, including cars that appear too small and objects that pop in and out of view. OpenAI shared four samples, but had not released a technical report or demonstrated Sora working; the article says their representativeness was hard to judge. OpenAI adapted the used in DALL-E 3 for video and combined it with a transformer that processes video chunks across space and . Tim Brooks said this let the team train on more varied video. The article presents both the creative promise and the risk: Sam Gregory of Witness said the model could let more people tell stories, while warning that synthetic video could misinform people about conflict zones or protests. OpenAI said it would share Sora with safety testers and selected video makers and artists, but had no current plans for public release. The article describes filters, a detector and C2PA tags as safeguards, while noting that detectors can be unreliable and metadata can be removed. How well these safeguards address the risks remains unknown.
Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims.
Original source
MIT Technology Review · Article
Source details
Original source date: Feb 15, 2024 (2024-02-15T18:15:00Z).
Date basis: Date from the publisher’s own records, through its API.
Added to Escape Meridian: 2026-11-21T15:33:14.015Z. First observed: 2026-11-03T08:21:55.619Z. Retrieved: 2026-11-03T08:23:24.103Z.
Referenced by
Recorded citations in public readings. A reference alone does not establish agreement or influence.
No public references recorded here yet.
Limits
Text as retrieved on November 3, 2026; originally published February 15, 2024. The article’s technical and capability descriptions rely in part on samples shared by ; it says their representativeness was uncertain and that OpenAI had not demonstrated the model working. The article describes intended safeguards and risks, not results from public deployment.
Open question
How representative are the shared samples of Sora’s typical output, and how well can its safeguards address the risks of realistic generated video?
Source-stated
MIT Technology Review reports that OpenAI’s Sora creates detailed video clips from short text descriptions and says the model can generate clips up to a minute long.
OpenAI has built a striking new generative video model called Sora that can take a short text description and turn it into a detailed, high-definition film clip up to a minute long.
Source-stated
The article describes visual flaws in a Tokyo sample, including cars that look too small and pop in and out of view.
In the Tokyo video, cars to the left look smaller than the people walking beside them. They also pop in and out between the tree branches.
Source-stated
The article says the samples may not represent Sora’s typical output and that OpenAI had not released a technical report or demonstrated the model working.
Without more information, it is hard to know how representative they are of the model’s typical output.
Source-stated
OpenAI adapted the diffusion model used in DALL-E 3 for video and combined it with a transformer that processes video data divided across space and time.
Sora takes this approach and applies it to videos rather than still images. But the researchers also added another technique to the mix.
Source-stated
Sam Gregory of Witness says the model’s expressive capabilities could enable more people to tell stories with video, while also creating avenues for misuse.
“The expressive capabilities offer the potential for many more people to be storytellers using video. And there are also real potential avenues for misuse.”
Source-stated
The article reports Gregory’s concern that synthetic video could misinform people about conflict zones or protests.
Gregory notes that you could use technology like this to misinform people about conflict zones or protests.
Source-stated
OpenAI said it would share Sora with safety testers and selected video makers and artists, and had no current plans to release it publicly.
OpenAI’s announcement of Sora today is a tech tease, and the company says it has no current plans to release it to the public.
Source-stated
The article says Sora’s proposed safeguards include filters, a detector adapted from DALL-E 3, and C2PA tags, while noting limits to detection and metadata.
Fake-image detectors are hit-or-miss. Metadata is easy to remove, and most social media sites strip it from uploaded images by default.
Source and contributor context
MIT Technology Review · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Diffusion model
A kind of generative AI that makes images, audio or video by starting from random noise and refining it step by step.
OpenAI
OpenAI develops AI models and products and publishes research, release announcements and evaluations.
TIME
TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Editorial provenance
Version 1. Publication ID: e89163f5-3b6a-0b6a-ff01-10ab1de40839.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: passed by anthropic/claude-opus-5.5 on 2026-11-21T15:28:35.020Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.
Observation: e8605a78ed880b49aef3128923a6e128f0078d888bc860939f88d6c2af009d54. Source text hash: dbfb8edc18f6011e80f8843a2743548cd9ce204789d60942f8ef1af08c5358a1.
Revision and source-check history
Version 1 · 2026-11-21T15:33:14.015Z
MIT Technology Review reports on OpenAI’s Sora video model and its risks
MIT Technology Review published this account on February 15, 2024. It reports that ’s Sora turns short text descriptions into detailed video clips and says the model could generate clips up to a minute long. The article describes detailed sample scenes but also visible flaws, including cars that appear too small and objects that pop in and out of view. OpenAI shared four samples, but had not released a technical report or demonstrated Sora working; the article says their representativeness was hard to judge. OpenAI adapted the used in DALL-E 3 for video and combined it with a transformer that processes video chunks across space and . Tim Brooks said this let the team train on more varied video. The article presents both the creative promise and the risk: Sam Gregory of Witness said the model could let more people tell stories, while warning that synthetic video could misinform people about conflict zones or protests. OpenAI said it would share Sora with safety testers and selected video makers and artists, but had no current plans for public release. The article describes filters, a detector and C2PA tags as safeguards, while noting that detectors can be unreliable and metadata can be removed. How well these safeguards address the risks remains unknown.
Original source
MIT Technology Review · Article
Source details
Original source date: Feb 15, 2024 (2024-02-15T18:15:00Z).
Date basis: Date from the publisher’s own records, through its API.
Added to Escape Meridian: 2026-11-21T15:33:14.015Z. First observed: 2026-11-03T08:21:55.619Z. Retrieved: 2026-11-03T08:23:24.103Z.
Limits
Text as retrieved on November 3, 2026; originally published February 15, 2024. The article’s technical and capability descriptions rely in part on samples shared by ; it says their representativeness was uncertain and that OpenAI had not demonstrated the model working. The article describes intended safeguards and risks, not results from public deployment.
Open question
How representative are the shared samples of Sora’s typical output, and how well can its safeguards address the risks of realistic generated video?
Source-stated
MIT Technology Review reports that OpenAI’s Sora creates detailed video clips from short text descriptions and says the model can generate clips up to a minute long.
OpenAI has built a striking new generative video model called Sora that can take a short text description and turn it into a detailed, high-definition film clip up to a minute long.
Source-stated
The article describes visual flaws in a Tokyo sample, including cars that look too small and pop in and out of view.
In the Tokyo video, cars to the left look smaller than the people walking beside them. They also pop in and out between the tree branches.
Source-stated
The article says the samples may not represent Sora’s typical output and that OpenAI had not released a technical report or demonstrated the model working.
Without more information, it is hard to know how representative they are of the model’s typical output.
Source-stated
OpenAI adapted the diffusion model used in DALL-E 3 for video and combined it with a transformer that processes video data divided across space and time.
Sora takes this approach and applies it to videos rather than still images. But the researchers also added another technique to the mix.
Source-stated
Sam Gregory of Witness says the model’s expressive capabilities could enable more people to tell stories with video, while also creating avenues for misuse.
“The expressive capabilities offer the potential for many more people to be storytellers using video. And there are also real potential avenues for misuse.”
Source-stated
The article reports Gregory’s concern that synthetic video could misinform people about conflict zones or protests.
Gregory notes that you could use technology like this to misinform people about conflict zones or protests.
Source-stated
OpenAI said it would share Sora with safety testers and selected video makers and artists, and had no current plans to release it publicly.
OpenAI’s announcement of Sora today is a tech tease, and the company says it has no current plans to release it to the public.
Source-stated
The article says Sora’s proposed safeguards include filters, a detector adapted from DALL-E 3, and C2PA tags, while noting limits to detection and metadata.
Fake-image detectors are hit-or-miss. Metadata is easy to remove, and most social media sites strip it from uploaded images by default.
Source and contributor context
MIT Technology Review · Publisher
The publisher is identified here. No individual author has been inferred.
Source media
Media retrieval status: unchecked.
Terms and references
Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement.
Diffusion model
A kind of generative AI that makes images, audio or video by starting from random noise and refining it step by step.
OpenAI
OpenAI develops AI models and products and publishes research, release announcements and evaluations.
TIME
TIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Editorial provenance
Version 1. Publication ID: e89163f5-3b6a-0b6a-ff01-10ab1de40839.
Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5.
Source check: passed by anthropic/claude-opus-5.5 on 2026-11-21T15:28:35.020Z. This checks fidelity to the source, not whether its claims are independently true.
Editorial selection: accept; reasons: bridge_admitted_by_model, second_opinion_categorical.
Observation: e8605a78ed880b49aef3128923a6e128f0078d888bc860939f88d6c2af009d54. Source text hash: dbfb8edc18f6011e80f8843a2743548cd9ce204789d60942f8ef1af08c5358a1.
passed · 2026-11-21T15:28:35.020Z · anthropic/claude-opus-5.5
Publication: e89163f5-3b6a-0b6a-ff01-10ab1de40839
attribution — pass: This is a single MIT Technology Review news article, not a listing. The reading attributes reporting to the publication and quotes to Tim Brooks, Aditya Ramesh's colleagues and Sam Gregory of Witness. The 'Deep Dive' sidebar titles are not used. One small nuance: the training-variety claim is attributed in the source to 'the researchers', and Brooks adds 'It really helps the model'. Crediting Brooks is close enough not to mislead.
The researchers say that this let them train Sora on many more types of video than other text-to-video models
summary — pass: The reading leads with the actor (OpenAI), the development (the Sora text-to-video model) and the article's skepticism about cherry-picked samples. It keeps the visible flaws, OpenAI's secrecy, the creative-versus-misuse tension and the limits of the safeguards. Diffusion model is defined beside the reading. The transformer is described by its function, processing video chunks across space and time, and C2PA is framed as metadata. Both are understandable enough. The title and question follow from the evidence.
Impressive as they are, the sample videos shown here were no doubt cherry-picked to show Sora at its best.
date — pass: February 15, 2024 matches the originalPublished UTC day. Gregory's comments were added in an update the same day, so they belong to the reading's dated account. The sidebar items describing later pieces are neither used nor relied on. Nothing hindsight-framed appears, and the lack of public-release plans is reported as OpenAI's stance at the time. The evidence kind is publisher-page, so no archive disclosure is needed.
Update 2/15: Comments from Sam Gregory were added.
support — pass: Each claim is grounded in the text: the minute-long clips, the Tokyo flaws, uncertain representativeness, no technical report, the diffusion-plus-transformer design, Gregory's promise and misuse remarks, sharing with testers and artists, and the safeguards with their limits. The closing line that safeguard effectiveness remains unknown is consistent with 'far from foolproof'. The later update note says OpenAI subsequently shared technical details. The reading's past-tense 'had not released' is accurate as of the original piece.
But these steps are far from foolproof. Fake-image detectors are hit-or-miss.
media — pass: No media was selected and no verified candidates were supplied, so there is no media identity or relevance to check.
Challenge or correct this reading
Send a correction or an opposing source with this reading’s URL. Disagreement and uncertainty can remain unresolved.
OpenAI develops AI models and products and publishes research, release announcements and evaluations.
Explore OpenAIA kind of generative AI that makes images, audio or video by starting from random noise and refining it step by step.
Definition referenceTIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMEOpenAI develops AI models and products and publishes research, release announcements and evaluations.
Explore OpenAIOpenAI develops AI models and products and publishes research, release announcements and evaluations.
Explore OpenAIA kind of generative AI that makes images, audio or video by starting from random noise and refining it step by step.
Definition referenceTIME is a news magazine publishing reporting on world affairs, business, technology and the workers behind digital services.
Explore TIMEOpenAI develops AI models and products and publishes research, release announcements and evaluations.
Explore OpenAI