# CAISI The Center for AI Standards and Innovation (CAISI) is the US government's primary body for AI evaluation, standards development, and frontier model testing — housed within the National Institute of Standards and Technology (NIST) under the [[Department of Commerce]]. It is the American counterpart to the UK's [[AI Security Institute]] and the institutional bridge between the AI industry and government oversight of frontier capabilities. ## Name and mission evolution CAISI was originally established in November 2023 as the US AI Safety Institute (USAISI), part of NIST's response to the Biden administration's October 2023 Executive Order on AI. Its founding mission was to advance the science of AI safety: pre-deployment evaluation of [[Frontier Model]] capabilities, dangerous-capability research, safety methodology development, and international cooperation. The Institute secured voluntary testing agreements with [[Anthropic]], [[OpenAI]], [[Google DeepMind]], and other frontier labs, gaining access to models before public release. In June 2025, the Trump administration renamed it the Center for AI Standards and Innovation — a deliberate linguistic shift from "safety" to "standards and innovation" that mirrored the UK's parallel rebranding of its AI Safety Institute to the AI Security Institute. The name change signaled the administration's framing of AI governance as a competitiveness and national security issue rather than a safety and ethics issue — consistent with the broader pattern of the Trump-era AI policy documented across the vault. ## Current operations CAISI's work in 2026 includes evaluating frontier models for national security risks (cybersecurity, biosecurity, chemical weapons capabilities), developing standards for agentic AI systems (the AI Agent Standards Initiative launched February 2026), and managing the NIST AI Consortium — a group of 280+ organizations developing voluntary guidelines for AI measurement. In April 2026, CAISI evaluated the open-weight DeepSeek V4 Pro model, reflecting its role in assessing both domestic and foreign AI systems. The center also houses a classified government taskforce established in late 2024 that collaborates with the intelligence community on evaluating AI models' national security capabilities and risks — connecting CAISI directly to the [[ODNI]], [[NSA]], and the defense AI ecosystem. ## KB relevance CAISI matters to the vault as the institutional point where AI safety evaluation meets government power. The [[Anthropic Model Export Directive]] — the June 2026 suspension of [[Claude Fable 5]] and [[Claude Mythos 5]] — was informed by the kind of dangerous-capability evaluations that CAISI conducts. The center's evolution from "AI Safety Institute" to "Center for AI Standards and Innovation" encapsulates the vault's thesis that the Trump administration reframes safety as weakness and standards as competitive advantage: same evaluators, same tests, different framing of what the results mean. ## See Also - [[AI Security Institute]] — UK counterpart - [[Department of Commerce]] — parent department - [[Dual-Use AI]] — the framework CAISI evaluates against - [[Anthropic Model Export Directive]] — policy informed by frontier evaluations - [[Guardrails]] — the safety mechanisms CAISI tests - [[Frontier Model]] — the class of models CAISI evaluates - [[Global Intelligence Hub]] — CAISI's connection to intelligence community evaluation