ROOST.tools

Yesterday, we shared one side of AI at ROOST and how we're thinking about AI-assisted and agentic tools in our own software development process. Today, we're sharing the other side: how safety teams can use AI agents to defend their platforms. And what it takes to do it right. The threat landscape has changed dramatically. Bad actors now generate unique spam, misinformation, and harmful content at scale with the press of a key. Accounts mimic real browsing behavior to evade detection. Meanwhile, safety teams are still constrained by the same structural limits: bespoke infrastructure, siloed platforms, and false-positive fears that limit automation. In our latest blog, ROOST's Head of Product Juliet Shen and CTO/Co-Chair of our Technical Design Committee Vinay Rao, alongside Rashmi Raghunandan from Block, Mingyi Zhao from Notion, and Hailey Elizabeth of Discord share a practical template for deploying AI agents in safety workflows, all built around three key ingredients: 🔹Entity understanding — giving agents a map of your data terrain, not just raw events 🔹Safety taxonomies — a shared language between humans and agents about what harm looks like; and 🔹Agent personas with playbooks — scoped, auditable roles that bound behavior and limit blast radius. The blog also includes real examples from Block and Notion already running these patterns in production. Take a look 👇 https://lnkd.in/eDwfbQFg

AI-assisted and agentic engineering is a frequent topic, and at ROOST, we've been thinking about what it means for open source safety tools. In our latest blog, our COO Anne Bertucio, along with ROOST Technical Design Committee members Hailey Elizabeth, Tim Pepper, and Tom Thorley share observations and the principles guiding how we're approaching these questions, including: 🔹Why strong SDLC practices remain foundational and how ROOST holds AI-assisted contributions to the same standard as any other code 🔹Considerations around using open source projects as references vs. imports, and the tradeoffs of going off main, and 🔹How the ROOST Playground gives our community a space to experiment and share what they're learning. As always, our work is anchored in our mission: making critical safety tools openly available, auditable, and relevant for all. Read our blog here: https://lnkd.in/eKGp5ryJ

Spotlight on one of our Technical Design Committee members! Tom Thorley, Senior Director of Safety & Integrity at GitHub, recently wrote about the importance of shared architectural frameworks for Trust & Safety teams and why getting the foundations right matters more than any single tool. Read his article here: https://lnkd.in/dK-Tt_tz ROOST's Technical Design Committee brings together Trust & Safety builders from across the industry to help guide the technical direction of our open source safety infrastructure. Meet our full committee: https://lnkd.in/ejJzcMXD

ROOST just got back from ATmosphereConf 2026 in Vancouver, and we're still thinking about it. Our OSS Community Manager Cassidy James Blaede and Head of Product Juliet Shen led a trust & safety crash course and introduced Coop (our open source safety dashboard) to a room full of builders working on the decentralized social web. What stood out wasn't just the crowd of people that tuned into our talk. It was the hallway conversations and the unconference session we hosted where indie developers asked blunt, practical questions about #CSAM detection and user reporting flows. The throughline across the whole conference: people, not platforms. Build in the open. Share what works. That's exactly what ROOST is here for. Full recap (including slides and a recording of our talk) on the blog: https://lnkd.in/eKSCsWHT

Trust and Safety teams are increasingly trying to figure out how AI can be a tool in their workflows. Customizable models like gpt-oss-safeguard, jointly released by ROOST.tools and OpenAI, need policies to unlock its capabilities. Remix or contribute your own flavor of these policies to the ROOST Model Community: https://lnkd.in/e5fjcGBj

ATmosphereConf 2026 is THIS WEEK, and we're proud to be a media sponsor. Our Head of Product Juliet Shen and Community Manager Cassidy James Blaede will be goosing off this week and into the weekend. Make sure to find them and say hi, and catch our talk on Coop (our OSS review tool) on Sunday: https://lnkd.in/d7wRZVeX Don't forget to use our link for $50 off a standard ticket. Limited discount, so grab it now: https://lnkd.in/ezVzt34S

We are thrilled to announce that Camille François will be speaking at our Responsible Tech Summit! Dr. Camille François is the founding president of ROOST.tools and a professor of practice at Columbia University’s School of International and Public Affairs. She has held senior trust and safety and research roles across social media, cybersecurity, and gaming — including Principal Researcher at Google, Senior Director at Niantic, and Chief Innovation Officer at Graphika. She is known for building high-performing teams to detect and mitigate online harms at scale, from child safety to violent extremism, and her research has helped shape safety practices across major Silicon Valley platforms. Her public interest work includes investigating information operations for the U.S. Senate Select Intelligence Committee, leading France’s governmental inquiry into immersive technologies, and being appointed by President Emmanuel Macron to co-chair the national consultative assembly on the future of the information society. Her current research focuses on public interest AI and new open source practices. Recognized by TIME’s “100 Next,” MIT Tech Review’s “35 Under 35,” Fast Company’s “Most Innovative Companies,” and the World Economic Forum, she is also a Fulbright Scholar and a Young Leader of the French-American Foundation. She is affiliated with the Harvard Berkman-Klein Center and the French Institute of Geopolitics, and her work has been featured in The New York Times, The Washington Post, WIRED, and Le Monde. Our Responsible Tech Summit happens on October 29th at The New York Times Center in Manhattan; curated for 375 leaders across civil society, government, industry, academia, and philanthropy who are coming together to collectively chart a better tech future. The theme for this year's summit is Embedding Accountability in the AI Era. Learn more: https://lnkd.in/eWy3yQtK #ResponsibleTech #ResponsibleAI #AllTechIsHuman #TrustandSafety

We are proud to be a media sponsor of ATmosphereConf 2026 — the global AT Protocol community conference! Kicking off next week, ATmosphereConf brings together the builders, researchers, and community members shaping the open social web on AT Protocol, better known as the open standard powering Bluesky Social and hundreds of other applications. This year’s conference takes place March 26–29, 2026 at UBC campus in Vancouver, BC, Canada, with two days of workshops and pre-conference events (March 26–27) followed by two full days of talks, panels, and community (March 28–29). Remote attendance is also available. At ROOST, our mission is to develop open source tooling for a safer online environment, and the AT Protocol ecosystem is exactly where that work matters. We’re excited to support a community that’s actively building the next generation of open, decentralized social infrastructure. 🎫 As a media sponsor, we have a limited number of discounted tickets to share! Use our link for $50 off a standard ticket: 🎫 https://lnkd.in/ezVzt34S (Discount code ROOSTRATE is limited, so make sure to grab your ticket while it lasts)

Online safety shouldn't be a competitive advantage. It should be a shared foundation. That's why ROOST exists: as an independent nonprofit building free, open source safety tools that any platform can adopt, audit, and improve together. Our board chair Clint Smith (CLO at Discord) shared a post celebrating what that looks like in practice: Discord donated Osprey to ROOST, the ROOST community refined it, and the improved version made its way back into Discord's tech stack, while simultaneously becoming freely available to any platform in the world. That's open source working as intended. We worked similarly with OpenAI for the release of gpt-oss-safeguard last year. Osprey is an open source rules engine built for real-world trust and safety operations, helping platforms investigate and address priority threats at scale, in real time, without sacrificing data privacy or performance. The numbers speak to the momentum. 360M+ users are now part of an ecosystem where open source safety tooling is actively working on their behalf. Platforms like Bluesky Social are already running Osprey. And every two weeks, contributors from across the industry gather in ROOST's public working group to shape what comes next. Read Clint's blog to learn more.

We're on the ground at GDC Festival of Gaming this year! Let us know if you're there, too, and come say hi to our COO Anne Bertucio. And if you get a hold of one of our special, limited postcards, we just might send you something cool. 🐓 #FestivalofGaming #GDC #opensource