Content moderation remains a heated topic in the online media landscape. With emerging regulations and growing public concern, it is poised to be a key focus for years ahead. However, the rise of weaponized AI and other technological advancements complicates the situation. A startup from Cambridge, England, named Unitary AI, claims to have found a more effective solution by employing a "multimodal" approach to address the complexities of content moderation, particularly in video.
Today, Unitary is excited to announce $15 million in Series A funding to enhance its market momentum. This round, led by prominent European venture capital firm Creandum, with contributions from Paladin Capital Group and Plural, coincides with significant growth for Unitary. The company has increased its daily video classification from 2 million to an impressive 6 million, covering billions of images, and is expanding its language offerings beyond English. While specific customer names were not disclosed, Unitary’s annual recurring revenue (ARR) is now in the millions.
Unitary plans to utilize this funding to extend its reach and recruit top talent. Although the startup hasn’t shared its current valuation, it has previously secured just under $2 million in earlier funding rounds, including $8 million in seed capital. Notable investors include Carolyn Everson, the former Meta executive.
In recent years, many startups have emerged, leveraging various artificial intelligence techniques to develop content moderation tools. The vast scale of video content presents a unique challenge that is well-suited for such technology. An army of humans alone could never effectively analyze the exponential amount of data generated and shared on platforms like YouTube, Facebook, Reddit, and TikTok, accounting for over 80% of all online traffic.
This reality has also caught the attention of investors. “In the online landscape, there’s a critical need for innovative technology to identify harmful content,” stated Christopher Steed, chief investment officer at Paladin Capital Group.
However, the content moderation arena is highly competitive. Companies like OpenAI, Microsoft's proprietary AI, Hive, ActiveFence/Spectrum Labs, Oterlu (now under Reddit), Sentropy (now with Discord), and Amazon’s Rekognition are among those currently operating in this space.
From Unitary AI’s perspective, many existing tools fall short of their potential, especially regarding video. Traditional moderation tools have generally focused on analyzing singular data types—such as text, audio, or images—but not across multiple modalities simultaneously, resulting in numerous false positives and negatives.
“What sets Unitary apart is our genuine multimodal models,” explained CEO Sasha Haco, who co-founded the company alongside CTO James Thewlis. “Rather than merely processing a series of frames, we analyze text, sound, and visuals simultaneously, emulating how a human moderator views a video to discern nuances, such as whether a clip is artistic or violent.”
Customers configure their own moderation parameters and typically use Unitary in conjunction with human teams, reducing the workload and stress for moderators.
With the clear advantages of multimodal moderation, one might wonder why it hasn’t been implemented sooner. Haco noted that many have relied on conventional, visual-only models, highlighting a significant market gap for innovation.
Content moderation challenges persist for social media platforms, gaming companies, and other digital outlets that enable user-generated media sharing. Recently, some social media giants have indicated a shift away from stringent moderation policies, while fact-checking organizations are losing traction. Questions continue to arise regarding the ethics of moderating harmful content, leading to diminished enthusiasm for combatting it.
Haco brings a unique perspective from her background. Before founding Unitary AI, she worked on black hole research with Stephen Hawking and holds a PhD in quantum physics. This experience informs her understanding of the complexities in moderation—a domain that, while terrestrial, can be just as intricate as the mysteries of spacetime.
Haco realized there was an overwhelming amount of noise in the content moderation market, but few products aligned with the actual needs of customers. Thewlis contributes his expertise with a PhD in computer vision from Oxford, focusing on enhancing visual comprehension with minimal manual labeling.
The name "Unitary" reflects dual meanings: it symbolizes the unification of various parameters for better video comprehension and harkens to Haco’s previous studies, as unitary operators describe complex quantum states, paralleling the unpredictable nature of online content and human behavior.
While multimodal research in AI has been underway for years, we are now witnessing a surge in real-world applications. Recently, Meta highlighted multimodal AI in its keynote, showcasing new assistant tools. Unitary stands at an exciting intersection of advanced research and practical implementation.
“We first met Sasha and James two years ago, and we have been thoroughly impressed,” remarked Gemma Bloemen, principal at Creandum and board member. “Unitary has emerged as a clear leader in the vital AI sector of content safety, and we are thrilled to support this exceptional team as they innovate in content classification technology.”
“From the outset, Unitary has demonstrated some of the most advanced AI for identifying harmful content. This year, the company has impressively achieved seven figures in ARR—an extraordinary milestone for such an early-stage company,” said Ian Hogarth, partner at Plural and board member.
Ranking
