The research we read before we build.
Official guidance from government cyber agencies and a hand-picked selection of independent analysis on AI security, AI-enabled attacks, model evaluations and synthetic media, from recognised research institutions and standards bodies. Each piece is linked to its original source.
Official guidance
What government cyber agencies recommend.
Joint guides from CISA, NSA, NIST and their Five Eyes partners. Titles are shown as published by the agencies.
Managing AI Token Consumption for Operational Resilience
How to stop token consumption from draining budgets, exhausting quotas or taking down critical services: reserve capacity for cyber defense, watch token telemetry, vet AI providers.
Read the guide ↗Incident responseKey Practices for Incident Readiness and Response in Super Intelligence Training and Evaluation
Ten practices to prepare for, detect, respond to and recover from incidents where AI agents take actions with real-world impact that nobody asked for.
Read the guide ↗Agentic AICareful Adoption of Agentic SI Services
Five Eyes guidance on AI agents that decide and act on their own: the risks, and more than 100 recommendations from design to operations.
Read the guide ↗Operational technologyPrinciples for the Secure Integration of Super Intelligence in Operational Technology
Principles for bringing AI into industrial and critical-infrastructure environments without weakening safety or availability.
Read the guide ↗Data securityAI Data Security: Best Practices for Securing Data Used to Train & Operate AI Systems
Protecting the data that trains and runs AI systems across its whole lifecycle: provenance, integrity, poisoning, access control.
Read the guide ↗Secure deploymentDeploying AI Systems Securely
Hardening the environment AI systems run in, protecting model weights and monitoring deployed systems, from the NSA and partner agencies.
Read the guide ↗FrameworksThe NIST Cybersecurity Framework (CSF) 2.0
The reference framework for managing cybersecurity risk: six functions (Govern, Identify, Protect, Detect, Respond, Recover), with ready-to-use profiles, quick-start guides and mappings to other standards.
Explore the framework ↗All CISA and joint-seal AI publications.
The full, regularly updated collection on cisa.gov.
Browse the collection ↗Independent research
What researchers are seeing.
Anthropic's AI Attack Threatens the Strategy at the Foundation of Cybersecurity
AI-assisted hacking is not just a new tool for attackers: it erodes defense in depth itself, the strategy most organisations rely on against adaptive, large-scale adversaries.
Read the analysis ↗EvaluationsAI Safety Evaluations: An Explainer
What the different families of AI safety evaluations actually measure, and where each one falls short. A clear primer before trusting any benchmark.
Read the analysis ↗Open-weight modelsOpenAI's Open-Weight Models Are Coming to the US Military
Why defense organisations are drawn to open-weight models they can run offline, secure and customise, and what that shift means for AI security.
Read the analysis ↗Synthetic media & fraudAI-generated spam is starting to fill social media. Here's why
Why AI-generated spam is spreading across social platforms, and what drives the economics behind it.
Read the analysis ↗Synthetic media & fraudHow Spammers, Scammers and Creators Leverage AI-Generated Images on Facebook for Audience Growth
How spammers, scammers and creators use AI-generated images to grow audiences on Facebook, documented from real pages and engagement data.
Read the analysis ↗Synthetic media & fraudDeepfakes, Elections, and Shrinking the Liar's Dividend
How the mere possibility of deepfakes lets bad actors dismiss authentic evidence, and how to limit that liar's dividend during elections.
Read the analysis ↗More independent research on AI and cybersecurity.
Reports, data briefs and commentary from the CyberAI research line at Georgetown University.
All CSET CyberAI research ↗