Guy Rosen

Sharing Meta’s Q2 2025 transparency reports and other integrity updates. transparency.meta.com/integrity-repo… (1/4)

AI Agents are already transforming the industry, but they also introduce novel security risks. To make AI agents more secure, our team developed a practical framework, the “Agents Rule of Two” - a practical way to reduce AI security risks: ai.meta.com/blog/practical…

We’re testing new Community Notes features at Meta: Anyone can now request a note or rate if a note is helpful - Users get notified when posts they’ve interacted with receive a Community Note - 70,000+ contributors have written 15,000+ notes (6% published). Learn more or join: meta.com/technologies/c…

Heads up, we're also changing the format of our Adversarial Threat Report and will publish the updated report next quarter. Stay tuned for more insights into our ongoing efforts to protect our platforms. (4/4)

We’re also providing an update on our efforts to combatting illicit drugs and supporting our users in prevention, treatment and recovery. We've invested in a dedicated investigative team, deployed new detection methods, and launched prevention campaigns to protect at-risk youth from recruitment. (3/4)

Sharing Meta’s Q2 2025 transparency reports and other integrity updates. transparency.meta.com/integrity-repo… (1/4)

Looking forward to more research in this area to help our industry build stronger/safer AI systems. More broadly, malicious groups keep evolving tactics - that’s why we keep examining our threat disruption strategies and how we can improve them.

👉AI: we joined peers for a first-ever public Red Team Challenge at @defcon this month. Over 2K security researchers and students stress-tested 8 LLMs to help probe for bugs and unintended behaviors – from bad math to misinformation to providing bad user security practices.

Today we published our quarterly integrity & security reports, and shared an under-the-hood view into our defense strategy and how we build it into products. about.fb.com/news/2023/08/i…

Today, academic journals Science and Nature have published four landmark research papers to better understand the impact of Facebook and Instagram on key political attitudes and behaviors during the US 2020 election cycle. about.fb.com/news/2023/07/r…

3️⃣ These reports show how persistent these threats can be. Malicious groups count on the industry working in silos while they target people across services. That’s why it’s great to see teams working together to protect people, share threat research and scale this security work.

2️⃣ We took action against malware targeting businesses across the internet, including browser extensions posing as ChatGPT and other GenAI tools. We shared findings with peers and rolled out new security features to help protect against malicious tactics. about.fb.com/news/2023/05/h…

🧵Sharing Meta's Q1 security and integrity reports. As part of our quarterly reporting, we just shared updates on our work to combat a range of threats globally, including covert influence operations, cyber espionage and malware campaigns. about.fb.com/news/2023/05/m…

Sharing a personal intimate image online can be scary and overwhelming, especially for young people. It can feel even worse when someone tries to use those images as a threat for more images, sexual contact or money — a crime known as sextortion. [a 🧵] youtube.com/watch?v=pAaXbB…