We hope we’ve caught you at that sweet spot of post Christmas gluttony and pre-year-end gloom. Work has ebbs and flows—the last quarter of 2025, however, was all flow. The work is inspiring us for our resolutions of 2026. Hope it brings some inspiration to you too!

Highlights

We released a report with the Rati Foundation on how AI is weaponized to target private individuals. It examines how AI-generated content is impacting and reshaping online harassment.
A month-long cycle of Viral Spiral playtests wrapped up—bringing gamers and stand-up comedians into its world.
We conducted digital safety workshops to denormalize online abuse in schools with the Uli project as the fulcrum.
Made feature updates to Anno.milli and continued work on AI safety guardrails with Tech4Dev.

Make It Real – A Report on Deepfakes with RATI Foundation

Drawing from cases reported to Rati's helpline Meri Trustline, the report reveals a concerning trend: while media and headlines often center on celebrities and politicians targeted through AI, a more personal crisis is also unfolding.

Ordinary survivors are being targeted through images that are artificially generated but possess the capacity for real harm. Survivors’ reputation is attacked and consent is erased through technology. These violations are muted by shame, fear and trauma—the incidents are rarely revealed to close family circles, let alone feature in larger discourse.

The full report can be read here.

The report was quoted by:

Tattle’s comments to the Intermediary Guidelines Amendment were based on this report.

Project Updates

AI Safety

Work with Project Tech4Dev and their AI Cohort Program

In continuation of our work with Tech4Dev in quarter three, we went deeper into building AI safety guardrails for the non-profits they work with.

As part of this work we conducted an in-depth analysis of chat conversations from AI-enabled bots of two non-profits. This helped us flag unsafe or undesirable interactions in the real world.

Building on the open-source project Guardrails AI and their repository of Safety Modules, we're identifying and building solutions to harms that are not adequately addressed by existing open-source solutions.

We’ve built solutions for detecting slurs and Personally Identifiable Information (PII) in Indian languages. We’ll work on more challenging problems like Toxicity Detection and Hallucination next.

This work will be open sourced as part of Tech4Dev's new project:

Kaapi Guardrails

Progress updates were shared at the official Pre-Summit Event of the AI Impact Summit 2026 in Bangalore.

Red Teaming IIT Madras’ Chatbot

We did a short sprint to work with IIT Madras to test their AI bot and its guardrails for their BSc program before its launch in January.

Uli

Student Workshops

The features we built for pedagogical purposes within Uli have transformed into in-person and digital workshops facilitated by Poorvi.

Through interactive exercises, discussions, and India-centric research on cybercrime and AI-enabled online abuse, the workshops help students learn practical ways to stay safe online.

Over this quarter we:

Created workshop material for late teens and young adults.
Conducted 3 workshops.
Signed up to conduct 14 more workshops in January 2026.

We started these workshops with the aim of co-creating knowledge with students on how to shift the blame and shame of online abuse away from victims.

We're continuously improving the workshop material after every session by learning from students’ experiences with social media.

Bahu Bhasa 2025 at IIITH

We were invited to a room full of language practitioners and language technology builders at Bahu Bhasa.

Over three days we interacted with educators, technologists, linguists, researchers, policymakers, and community practitioners. We were thrilled to introduce Uli and its participatory approach, and receive valuable feedback on our work.

The room largely shared a dislike for the term "low-resource language" and optimism about improving language representation in NLP technology.

While many technologists spoke about data labelling and annotations, we were struck by how language practitioners focused on preserving and sharing their languages through blogs, WordPress, Wikipedia, Instagram Reels, and other community-driven platforms.

We left wondering whether Tattle should translate its website into romanized Indian languages, something our community already uses, instead of waiting for perfect translations that have proven difficult to scale.

Viral Spiral

During November we conducted one to two online playtests every week. These continued the in-person playtests we held in Mumbai during the previous quarter.

This time we focused on gamers and potential players rather than academics.

Highlights from these sessions:

The game is now live at https://viralspiral.net/.

Take it out for a spin with your friends and communities. If you'd like us to host a playtest, feel free to email us.

Anno.Milli

Our work on community software, decentralized social media and interoperable open standards comes together in Anno.Milli, a project we are building with NCBS.

This quarter the software received several updates:

Import support for more data formats.
Support for additional vocabularies.
A new Emotion/Affect annotation type.

Some of the archival protocols and data formats used by this project are over 30 years old.

Interestingly, many of the same ideas—federation, annotations and community participation—are now reappearing in conversations around the future of social media and community moderation.

Projects like GLAM-Wiki also demonstrate how galleries, libraries, archives and museums can collaborate with Wikipedia to improve public knowledge.

We're optimistic that Anno.Milli can similarly strengthen participation from traditional institutions in addressing online harms.

Tattle in Public

We participated in the UNDP Trust and Safety Reimagination Lab alongside teams from around the world.
Tarunima delivered a spotlight talk at The Reuters Foundation Trust Conference on building people-centered technology ecosystems.
Tarunima moderated "AI Content and Reality: Safeguarding Citizens from Harmful and Misleading AI-Generated Content" at TIAcon 2025.
Kaustubha participated in IT4Change and Antara Advisory's roundtable on advancing gender equality in India.
Baarish joined the ARISE Panel at Digital Citizen Summit 2025.
Tarunima spoke on the "Design ≠ Diversity" panel at Access, Agency, Azaadi: The Internet We Deserve in Mumbai.
Tarunima joined the Agency Fund and Center for Global Development working group on AI evaluations for the development sector.
Kaustubha attended Equality Now's strategy lab in London on Algorithmic Harms and Misogyny.
Denny participated in a roundtable discussion on AI Safety Commons during the IndiaAI Impact Summit 2026.

From Tattle's Blog

Conclusion

We welcomed four new people to Tattle in November.

After completing a quarter together, Baarish, our Research Manager, brought the team together to map Tattle's work in AI safety and better understand the motivations behind it.

The more we explored our projects, the more we saw recurring themes around centralized platforms, incentives and online harms—issues we've been working on for the past six years.

As we move into the new year, we're doubling down on our mission and our values.

2025 Q4 Updates

Tags: