Meet Lavanya - HeyGen’s Head of Trust and Safety

HeyGen’s mission is to help businesses grow by making visual storytelling accessible to all. In a video-first world, businesses struggle to keep up with the consumer demand for more video content, whether an HR training video, marketing content, sales videos, and more. With HeyGen, businesses can create videos from text in seconds, localize videos into 175+ languages and dialects, and personalize videos at scale.

Working in AI video generation means Trust and Safety is crucial to the business. You can read our ethics statement, moderation policy, and community guidelines for more information.

I recently joined the HeyGen team to lead Trust and Safety, and I’ve compiled some of the questions you may have about Trust and Safety below. 

Trust and Safety at HeyGen

As Head of Trust and Safety, I know that trust is not built in a day, it takes several years of consistency and several years of persistence. Having consistent policies and enforcing them effectively in the product is always top of my mind.

HeyGen is one of the few companies that has made a proactive and conscious effort to put Trust and Safety at the forefront of product and business development. Since the company’s inception, Trust and Safety has been a top priority. Safety is not just a buzzword at HeyGen, but built into the product and a crucial facet of everything we do. 

How we prevent misuse?

HeyGen demonstrates an automation-first with a human-in-the-loop approach when it comes to moderation. Our key concerns are centered around identity authentication - how do we know these people are who they say they are - and how do we make sure content is appropriate and in line with our guidelines, and how we handle issues should they pop up.

After the initial content review by machine, HeyGen’s human moderators conduct final approvals for the most critical points in the process, and manage user exceptions. The human moderators also help our automation improve via human feedback. Here’s what the process looks like in practice: 

1. Text moderation: HeyGen automation checks for banned words and prevents the user from saving the video if banned words are detected. If the machine rejects text submitted by a customer, the rejection is queued for human moderators to review manually. 
2. Audio moderation: All audio files are converted to text and moderated as described above.
3. Image and video moderation: Image and video moderation is done via models that are looking for content like nudity, pornography, hate symbols, violence, and more, and removes images that violate HeyGen’s policies.
4. Consent video: All users on HeyGen will share a consent video for HeyGen to use their biometrics to create an avatar and future videos. This is perhaps the most sensitive and critical part of HeyGen’s workflow. This consent video goes through two levels of validations: Automatic and Manual. The machine will check for authenticity as well as “Is_celebrity” flag to identify videos of public figures, who are at higher risk of being targeted for impersonation. We are committed to ensuring all content is made with approval and scanning for celebrities, public and political figures' likenesses and block them by default. This consent video is also reviewed by humans to ensure that the user creating the avatar is the same person as giving the consent without question.

How does HeyGen moderate content that does not hold up to standards, or may be inauthentic - such as deepfakes or bots?

HeyGen is a platform for businesses to generate video content. We do not allow political or election content on our platform. This includes a ban on political actors from using HeyGen to create avatars for any purpose. We believe this is a critical step in protecting election integrity. HeyGen's policies and products strictly prohibit the creation of unauthorized content, and we take abuse of our platform seriously. Our security safeguards include advanced user verification, including live video consent, dynamic verbal passcodes, and rapid human review of all avatar verifications, as well as active scanning and removal of content that violates our policies. 

Trust & Safety is critical to our business, and we are actively partnering across the industry to continue developing the tools and best practices necessary to combat misinformation and AI misuse.

HeyGen built tools and technology to monitor traffic and users right from sign up through transactions (risk and payment) to help prevent bots from joining the platform. This approach has helped us to allow, block or flag these users and proceed accordingly. We continue to work on expanding our digital fingerprinting further to improve accuracy.   

‍Where we’re heading?

HeyGen is doubling down on their investment in terms of commitment to content that complies with community guidelines and code of conduct. 

How we do it:

1. Safety as part of the design: The products go through a regular review for policy and safety design check. Every sprint planning goes through T&S evaluation.
2. Continual updates to banned words: HeyGen is adding thousands of more banned words to the keyword repository to make text moderation even more scalable and fight hate speech. 
3. Expand image and video moderation: HeyGen is integrating with third party tools to add more extensive matches on celebrities and political figures. This will enhance HeyGen’s deepfake efforts and exponentially reduce malicious content that could be generated on the platform. With expanded keyword and image moderation, video creations and translations will continue to be reduced. 
4. Remove bad actors: HeyGen continues to invest in identifying and removing bad actors. The unique fingerprinting technology is used to block users that are flagged as high risk and prevent them from doing sensitive activity including and not limited to creating avatars, videos and payments.

Trust and Safety are the core of our platform and business. We serve some of the world’s largest companies and nonprofits, so leading in trust and safety is imperative, and hence, safety is built as part of the design. 

If you have additional questions about trust and safety at HeyGen, please contact support at [email protected].