Michael O'Herlihy

In August, X open-sourced the code that affects a post’s visibility in the For You feed – including ranking, model configuration, filters, and core systems. The repository is publicly available, updated regularly, and accepts pull requests from outside contributors. Alongside the open-sourced code, Under the Hood gives eligible users aggregate data on any visibility-limiting labels applied to their account or posts, including cases driven by local legal demands. Together they make the system inspectable. That is the standard.

Excited to have Michael O’Herlihy as the new Head of Safety at X. He’ll lead our work to keep people and brands safe across the platform and our AI systems, while protecting free expression and making Safety more transparent. @michaelo will be at the Brand & AI Safety Summit in New York next Tuesday November 10 making the case for why open systems and controls people can see matter for both safety and advertiser confidence.

We take seriously our responsibility to protect minors. X has zero tolerance for CSAM, including AI-generated CSAM. As an industry leader, we actively report and block users who violate our Terms of Service, which absolutely prohibit all forms of child sexual exploitation, including posting or engaging with child sexual abuse material. So far this year, we have reported 1.3 million cases to the National Center for Missing and Exploited Children. Additionally, we employ industry-leading tools and processes to protect minors, including: - We scan media against known-CSAM hash libraries (including hashes from NCMEC and partners such as IWF) using industry tools such as PhotoDNA alongside proprietary detection. We run models at post-creation time (LLMs and custom classifiers, legal/policy-reviewed) and conduct manual Child Safety operations reviews to catch CSE/CSAM that isn’t already in a hash set. We also run proactive keyword and media sweeps, plus heuristics that are partially or fully automated, when signals indicate possible CSAM. - Apparent CSAM leads to removal, and we have permanently suspended over 780,000 users in the past month alone, with ban-evasion controls. Whenever we become aware of apparent CSAM (including AI-generated CSAM), we report it to NCMEC. AI-generated CSAM is enforced under the same Child Sexual Exploitation policy as non-generative CSAM. We also continuously improve hash coverage, classifiers, and review workflows because this threat is ever evolving and highly adversarial. - We have also brought and will, if necessary, continue to bring lawsuits against users who have intentionally circumvented safeguards to generate this despicable content. In the first half of 2026 alone, X Corp made more than 1.3 million reports to the National Center for Missing and Exploited Children, resulting at least 543 arrests - Finally, we are an active member of the Tech Coalition, an alliance of global tech companies who are working together to combat child sexual exploitation and abuse online. Tech Coalition is an invaluable resource to understand the best practices and new approaches used across the industry to detect and prevent the sharing of CSAM. It allows member companies to share best practices, challenges and new insights that ensure that we are up-to-date when it comes to trends related to these high severity violations.