Samurai can autonomously conduct kind, normative, or empathic interventions while educating disruptive users and nudging them towards a path to redemption without sacrificing user experience.
Use the only stand-alone solution in your communities without unnecessary cost. Reduce the loss of revenue associated with toxic users while improving the mental health of moderators and community managers.
Enterprise-level integration, 24/7 support, service level agreement. Regulatory compliance with GDPR, COPPA, and DCMS. Protect your brand, product, and your user data from the associated legal ramifications.
Integrate with your product seamlessly using our well-documented and easily customizable API. Understand precisely why Samurai is making a decision through its nuanced categories and detailed output.
76% of cyberbullies voluntarily change their tone after being redirected by Samurai's Cyber Guardian
The First Truly Proactive Response
The Samurai Cyber Guardian
98% Precision Detection of Personal Attacks
Samurai provides the transparency and control of AI which is necessary to be trusted to act autonomously. It detects personal attacks with a rate of 2% false positives.
Why Neuro-Symbolic AI?
Neuro-Symbolic AI requires a fraction of the data compared to deep learning, it is unbiased, so you can track and understand precisely why it's making a decision. As it detects cyberviolence, it can take action, intervene autonomously, or even notify moderators before the damage is done.
Profanity and Personal Attacks are not Mutually Exclusive
Samurai understands nuanced forms of communication from context to intent. Our language models manage the distinction between profanity and personal attacks, it also understands the difference between language targeted at first party or third party. Standard models for precision detection include profanity, sexual remarks, rejection, personal attacks, threats, and sexual harassment.
Blackmail Detection Example:
Send me more nudes or I'll publish the ones I already have
|Samurai Cyber Guardian Detection|
|Blackmail threat to reveal information|
|Sexual harassment attempt to solicit intimate photographs|
Resistance to Adversarial Attacks
Symbols, numbers, boundaries, special characters, abbreviations, or combinations are often caught by Samurai. When methods for resistance to adversarial attacks change, our AI adapts accordingly.
Every Community has Different Needs
Every community has different needs, Samurai is completely customizable to suit those needs. Set your own parameters for handling profanity, violent language detection and interventions. We can even set up new language models specific to the community.
Educate and Redirect
Samurai's precision has enabled our AI to act autonomously in various ways. Samurai can block messages from being sent, issue kind, or normative messaging while educating the user on why their message might be considered offensive while prompting the user to revise. Samurai can also proactively & autonomously intervene to cool down a user.
Identify the Bad Actors
Catch negative behavior in the moment, intervene, de-escalate, encourage and/or reward positive redirection. Identify the bad actors and place them on a path to redemption. Recognize the users who intervene on their own, those who set an example for the community by exhibiting the behavior of a community ambassador like a true samurai.
Samurai can turn moderators into superheroes, with 98% precision, Samurai reduces moderator load by over 90%. Moderators can proactively prioritize imminent crisis before the damage is done.
Community, Brand, and Customer Insight
See exactly how Samurai is not only impacting but improving your communities and your business. See real-time reporting on changes in toxicity, churn, and retention.
Setting the Tone for Protecting Your Communities
Trust & Safety Starts with Username Moderation
The most effective solution for guarding your communities against users with disruptive identities. Built for automation. Scalable for any application.
See it in Action
Protect Your Communities
With Username Guardian
Free Developer API
For Enterprise-Level Solutions, Free Community Assessment, Custom Models, or SLA
Our Unique Approach
Samurai takes a Neuro-Symbolic approach to detect hidden meanings in usernames. Identifying individual words within the username and generating as many variants as possible. We use these techniques to find the original attempt at creating a disruptive username. This method allows us to categorize different types of subversive language in a user's online identity.
% Increase in Disruptive Behavior Based on Username
Decomposition of the username - DigBildo - Big D*ldo
Categories - Offensive, Sexual
Toxic Element - D*ldo
Inappropriate - P41nbowRub3s - Rainbow Pubes
Offensive - D34thToFurr1es - Death to furries
Profanity - FolyHuck - Holy F*ck
Sexual - Aneed Morehead - I need more head
Simple Ideas Require Complex Solutions
Filter or keyword-based methods are inefficient & inaccurate. Leetspeak, numeric, and character workarounds require a nuanced understanding of composition, Samurai protects your communities from these adversarial attacks and more "creative" approaches.
There was a time when most people believed the social internet was nothing but full of possibility. We have long taken a stand that a strong vision, ethics, and approach to technology is necessary to protect children and online communities in a world that has the potential to become very dangerous.
We are a committed collective intelligence of experts across domains creating a new type of research organization that builds AI products with the highest integrity.
We combine rigor of deep, peer-reviewed scientific and academic research with the speed and agility of a decentralized private technology company.
Applied Research and Development
Private and public sector organizations work with us in developing foundational research and applying it to create high performance violence detection and intervention intelligence and tools. Every community and environment is unique. And the precision, flexibility and transparency of our language engine enables us to solve for a diversity of contexts and use cases.
NEW Paper on how personal attacks decrease user activity in social networking platforms. Read more
NEW See evidence from our work on Reddit proving we can reduce violence in online communities. Read more
NEW Read our paper on improving classifier training efficiency for automatic cyberbullying detection. Read more
Learn about our work with Hate Lab and Cardiff University. Read more
See how we are helping INACH on the hidden corners of the dark web. Read more
Read our published research tackling the problem of sarcasm detection with the use of machine learning and knowledge engineering techniques. Read more