diff --git a/SECURITY.md b/SECURITY.md
new file mode 100644
index 00000000..f2c1708d
--- /dev/null
+++ b/SECURITY.md
@@ -0,0 +1,217 @@
+Security Policy
+
+Overview
+
+This file provides repository-level security reporting guidance for the X For You Feed Algorithm repository (xai-org/x-algorithm).
+
+The repository contains core source code used to determine which posts a viewer sees in the For You feed on X, including candidate retrieval, ranking, filtering, visibility-related systems, and related model and pipeline code. X's repository documentation also states that some deployment-related code and infrastructure are not included.
+
+This file does not replace, expand, or modify the official X / xAI vulnerability disclosure or bug bounty terms. The current X / xAI Bug Bounty Program on HackerOne is authoritative for scope, eligibility, testing requirements, safe-harbor terms, bounty decisions, disclosure rules, and other program requirements.
+
+Reporting a Security Vulnerability
+
+Do not report sensitive security vulnerabilities in a public GitHub issue, pull request, discussion, commit message, or other public channel.
+
+X's official security guidance directs security researchers to report possible vulnerabilities through the X / xAI Bug Bounty Program on HackerOne:
+
+X / xAI Bug Bounty Program: https://hackerone.com/x
+
+X Help — Reporting security vulnerabilities and bugs: https://help.x.com/en/rules-and-policies/reporting-security-vulnerabilities
+
+X asks reports to include information and detailed instructions for reproducing the issue.
+
+When submitting a report, include as much of the following as is safely available:
+
+A clear description of the vulnerability and its security impact.
+
+The affected X / xAI asset, service, repository component, file, module, or code path.
+
+The exact conditions required to reproduce the issue.
+
+Step-by-step reproduction instructions.
+
+A minimal proof of concept, where appropriate and permitted by the official program rules.
+
+Relevant request/response data, logs, stack traces, screenshots, or other evidence.
+
+The tested environment, revision, commit hash, branch, or build information.
+
+Whether the issue was reproduced against open-source code only, an X / xAI service, or both.
+
+Any known prerequisites, permissions, account states, or configuration requirements.
+
+A concise explanation of the confidentiality, integrity, authorization, or other security boundary that is affected.
+
+Any mitigations or fixes you believe may be useful.
+
+Do not include unnecessary personal data, authentication secrets, access tokens, credentials, private user content, or other sensitive data in a report.
+
+Research and Testing Requirements
+
+Follow the current X / xAI HackerOne program rules before performing security testing.
+
+X's official Help Center specifically instructs security researchers to:
+
+Test against accounts they control.
+
+Use an X account with protected posts when researching security issues, to reduce the risk of publicly disclosing vulnerability details.
+
+Do not use this repository as authorization to test systems, accounts, data, infrastructure, or services that are not permitted by the current official X / xAI bug bounty policy.
+
+Do not intentionally access, modify, retain, destroy, or disclose data belonging to other users.
+
+If testing could affect other users, production availability, data integrity, or service reliability, stop and follow the official HackerOne program instructions.
+
+Repository-Specific Security Issues
+
+The current xai-org/x-algorithm repository is intended to provide transparency into code that affects post distribution in the X For You feed. According to the repository documentation:
+
+In-network and out-of-network posts are retrieved and ranked together.
+
+Ranking and visibility filtering are separate parts of the system.
+
+Visibility-related components can determine whether content is shown, dropped, or shown with a treatment.
+
+The repository includes model, retrieval, ranking, filtering, labeling, and enforcement-related code.
+
+Some code is designed to run end-to-end, while other areas omit deployment or infrastructure details.
+
+Some files are intentionally not published where disclosure could increase the risk of gaming or abuse.
+
+A security issue discovered in this repository may therefore differ from a directly exploitable vulnerability in X production.
+
+Potentially security-relevant findings may include, subject to the current official X / xAI HackerOne scope and rules:
+
+Unauthorized disclosure of non-public user or system information.
+
+Authorization or access-control bypasses.
+
+Security-boundary bypasses that expose protected or private content.
+
+Vulnerabilities that could alter, inject, or tamper with protected data or trusted system behavior.
+
+Remote or local code-execution vulnerabilities in runnable components.
+
+Injection vulnerabilities.
+
+Memory-safety vulnerabilities with a meaningful security impact.
+
+Exposure of secrets, credentials, tokens, or sensitive configuration.
+
+Vulnerabilities in security-sensitive filtering, labeling, enforcement, or request-handling logic where the result creates a genuine security impact.
+
+Supply-chain or dependency vulnerabilities that are actually reachable and security-relevant in an X / xAI in-scope asset.
+
+The presence of a bug, unusual ranking result, model error, or undesirable recommendation does not by itself establish a security vulnerability.
+
+What Belongs on GitHub Instead
+
+Use the repository's normal GitHub issue or pull-request workflow for non-sensitive engineering and transparency work, such as:
+
+Reproducible non-security bugs in the open-source code.
+
+Build or documentation problems.
+
+Feature requests.
+
+Code-quality improvements.
+
+Performance improvements that do not reveal a security weakness.
+
+Questions about how the published algorithm works.
+
+Suggestions for improving recommendation quality.
+
+Transparency or explainability feedback.
+
+Model or ranking behavior that does not create a security boundary violation.
+
+Before posting publicly, make sure the report does not contain exploit details or information that could expose a security vulnerability.
+
+If you are unsure whether an issue is security-sensitive, prefer the private HackerOne route.
+
+Algorithmic Abuse, Spam, Safety, and Policy Reports
+
+Not every harmful behavior is a software security vulnerability.
+
+Reports about spam, platform manipulation, content-policy violations, abusive accounts, or ordinary account moderation should use the appropriate X reporting and support channels unless the report demonstrates a qualifying security vulnerability covered by the official bug bounty program.
+
+Security reports should focus on a concrete technical weakness and its reproducible security impact.
+
+Personal Account Issues
+
+The security vulnerability reporting process is not the support channel for a compromised or locked personal account.
+
+If your own X account has been hacked, compromised, locked, or otherwise requires account support, use X's official Help Center and account-support processes instead of opening a repository security report.
+
+See:
+
+https://help.x.com/en
+
+https://help.x.com/en/rules-and-policies/reporting-security-vulnerabilities
+
+Supported Code and Versions
+
+This repository is an open-source transparency repository and does not publish a repository-specific security-support lifecycle in its current documentation.
+
+Accordingly:
+
+Do not assume that every historical commit, branch, fork, example, synthetic dataset, or local build corresponds to production X.
+
+Do not assume that code present in the repository is deployed exactly as published.
+
+Do not assume that omitted deployment or infrastructure code is in scope for testing.
+
+When reporting, identify the exact commit, branch, module, and environment you tested.
+
+The official X / xAI HackerOne program determines whether an affected asset and vulnerability are eligible for handling or bounty consideration.
+
+Security issues that only affect a third-party fork or modified deployment should normally be reported to that fork's maintainer unless the same issue also affects an asset covered by X / xAI's current program.
+
+Bounties, Eligibility, and Severity
+
+This repository does not independently promise a bounty, severity rating, response time, remediation deadline, or disclosure date.
+
+All bounty eligibility, severity assessment, duplicate handling, reward decisions, scope determinations, testing restrictions, disclosure conditions, and related terms are governed by the current X / xAI HackerOne program policy.
+
+Always review the live HackerOne policy before beginning testing and again before submitting a report, because scope and rules can change.
+
+Responsible Disclosure
+
+Keep vulnerability details private while X / xAI evaluates the report and follow the disclosure requirements of the current HackerOne program.
+
+Do not publish exploit instructions, proof-of-concept code, screenshots, logs, reproduction steps, or other sensitive details before disclosure is permitted under the applicable program rules.
+
+Public GitHub issues and pull requests are not appropriate for undisclosed vulnerabilities.
+
+Security Fix Contributions
+
+If a vulnerability requires a source-code change, report the vulnerability privately first through the official program rather than opening a public pull request that reveals the issue.
+
+A public fix can unintentionally disclose the vulnerability before affected systems are evaluated or remediated.
+
+Coordinate any security-sensitive code contribution or disclosure with X / xAI through the official reporting channel.
+
+Official References
+
+The following sources are the basis for this repository-level guidance:
+
+X For You Feed Algorithm — official xai-org/x-algorithm repository
+https://github.com/xai-org/x-algorithm
+
+X Help — How to report security vulnerabilities
+https://help.x.com/en/rules-and-policies/reporting-security-vulnerabilities
+
+X / xAI Bug Bounty Program — HackerOne
+https://hackerone.com/x
+
+Legacy official X/Twitter Recommendation Algorithm security policy
+https://github.com/twitter/the-algorithm/security/policy
+
+The legacy twitter/the-algorithm repository explicitly directed sensitive security issues to X/Twitter's bug bounty program rather than GitHub. The current X Help Center likewise directs possible security vulnerabilities to X's security team through HackerOne.
+
+Policy Authority
+
+If any statement in this file conflicts with the current X / xAI HackerOne policy or an official X security instruction, the current official X / xAI policy controls.
+
+This file should be updated when X / xAI changes its official vulnerability-reporting process, repository architecture, or security program.