OpenAI is developing tools to combat AI-generated video: Here’s how

2024, just as 2023 was, is the year of AI. Now, artificial intelligence, it’s no mystery, has been making steady inroads into all walks of life everyday – whether academics or industry, whether through its text, image, or video generation capabilities. At the same time, however, there have also been concerns raised by many over potential security breaches by way of identity theft, or even fraud using what’s called deepfakes.

Accordingly, several AI industry stalwarts are also developing tools to keep the same in check. Recently, one of the biggest names in the business, OpenAI, announced that it is developing several new tools to detect whether any given piece of content is AI-generated. Read on.

What are the new AI content-detection tools OpenAI has developed?

AI-generated image detection tool 

OpenAI has developed a new image detection classifier that uses AI to detect if any given image is AI-generated. The tool, for now, is mostly capable of detecting images developed using DALL-E 3, with about 98 per cent accuracy.

ALSO READ: What is a deepfake video and how do you protect yourself from one?

However, AI content-detection with images from other tools such as Midjourney is still a work in progress, with the tool showing only about 10 per cent accuracy in their case. Additionally, it has also developed a tamper-proof watermark, which can tag content with invisible signals, to ensure they cannot be recreated or manipulated in any form.

C2PA metadata

The company also started to recently integrate C2PA metadata to all text generated on ChatGPT, images generated using DALL-E 3, and even videos generated on Sora, its text-to-video generation platform. A while ago, the company also began to add watermarks to clips generated from Voice Engine, a new text-to-speech platform it has been working on.

Interestingly, OpenAI has been working on AI-content detection tools for a while now However, it shelved an AI-text classifier it had developed in 2023, owing to low accuracy levels.

ALSO READ: Deepfake audio are near impossible for humans to spot, study suggests

Bear in mind though, there’s been no word on whether these tools will be integrated into OpenAI’s existing platforms, or will be available as standalone apps. Secondly, there’s also been no work on which platforms they wil be available on – whether solely desktops and laptops, or smartphones too.

OpenAI and the C2PA: What you need to know

OpenAI, in a recent blog post, announced that it has joined the steering committee of the Coalition for Content Provenance and Authenticity, or the C2PA. For the uninitiated, C2PA is a standard used for the certification of authenticity of all forms of digital content, whether text, images, audio or video. It has been adopted by several major tech giants, including the likes of Microsoft, Google, and even Adobe and Sony.