Perplexity releases AI security tool as open source

Perplexity’s release of its open‑source AI sandbox, SPACE, which it says can prevent autonomous AI agents from escaping controlled environments and accessing unauthorized data, marks a notable advancement in AI security tooling. For website owners, this means a new, freely available mechanism to sandbox any third‑party AI agents that interact with their sites—whether through chat widgets, recommendation engines, or content‑generation APIs. By deploying SPACE, owners can isolate these agents, ensuring they cannot crawl hidden pages, scrape user data, or trigger unintended actions that could inflate bot traffic or violate privacy regulations. This directly impacts how sites monitor AI‑generated traffic in their analytics and enforce the llms.txt directives that govern permissible AI interactions. **Actionable tips:** 1. Integrate SPACE into your server environment and configure it to route all inbound AI requests (e.g., from OpenAI, Anthropic, or custom models) through the sandbox before they reach your production endpoints. 2. Update your llms.txt file to explicitly reference the sandboxed endpoints, specifying allowed model IDs and disallowing any agents that bypass SPACE. 3. Use llmscentral’s bot‑tracking dashboard to flag traffic originating from sandbox‑bypassed IPs, enabling rapid alerts if an autonomous agent attempts to escape the sandbox.