AI Security
Security guides for the AI tools your organization uses
What each tool can access, where the real risks are, the questions to answer, and the controls to put in place. Written for security, IT, legal, and compliance teams.
AI assistants
- OpenAIChatGPTWhat ChatGPT can access in a business, the difference between consumer and business accounts, connector and agent risks, HIPAA considerations, and the controls security teams should put in place.Read the guide
- AnthropicClaudeHow organizations use Anthropic's Claude, what it can access through projects, uploads, and connectors, the risks for law firms and regulated industries, and the controls to put in place.Read the guide
- MicrosoftMicrosoft CopilotWhy Microsoft 365 Copilot's biggest risk is existing permission sprawl, what Copilot can reach through Microsoft Graph, Copilot Studio agent risks, and the controls to deploy before and after rollout.Read the guide
- GoogleGeminiHow Gemini reaches organizational data through Google Workspace, the difference between consumer, Workspace, and API use, prompt-injection exposure in email and documents, and recommended controls.Read the guide
- PerplexityPerplexitySecurity considerations for Perplexity's AI answer engine: what employees send to it, file uploads and connectors, agentic browsing and prompt injection, account types, and controls.Read the guide
AI coding tools and agents
- GitHubGitHub CopilotSecurity considerations for GitHub Copilot completions, chat, IDE agent mode, and the Copilot coding agent: repository access, secrets exposure, prompt injection through issues and code, and governance controls.Read the guide
- AnthropicClaude CodeHow Claude Code reads files and runs commands, what that means for .env files, cloud credentials, and production access, permission modes, and the controls engineering teams should set.Read the guide
- AnysphereCursorSecurity considerations for the Cursor AI code editor: code sent to model providers, privacy settings, agent command execution and auto-run, rules files and prompt injection, and team controls.Read the guide