Skip to content
ContentLora

    Tip: press / anywhere to search.

    organization

    CAISI / CAISSI: the US government's AI testing centre at NIST

    Also known as CAISI, CAISSI, Center for AI Standards and Innovation, Center for Advancing Innovation and Standards for Super Intelligence, US AI Safety Institute, US AISI

    The US government's frontier AI testing body sits inside NIST. Created as the US AI Safety Institute, it was rebranded the Center for AI Standards and Innovation (CAISI) in June 2025,[1] and as of October 2026 NIST presents it as the Center for Advancing Innovation and Standards for Super Intelligence (CAISSI).[2] It runs voluntary pre-deployment tests of models from major US labs and evaluates foreign models.[3][4]

    Editor reviewedStrict sourcingUpdated AI safety and alignmentTech policyArtificial intelligence
    Key facts

    What it is

    NIST describes the centre as industry’s primary point of contact within the US government for testing and collaborative research on advanced AI systems. Its listed functions include voluntary agreements with private developers, unclassified evaluations, and assessments of the capabilities of US and adversary systems.[3] It also lists representing US interests internationally to guard against “burdensome and unnecessary regulation”.[5]

    Three names

    The body was created at NIST as the US AI Safety Institute. On 3 June 2025 Commerce Secretary Howard Lutnick announced that it would become the Center for AI Standards and Innovation (CAISI),[1] directing it to focus on demonstrable risks such as cybersecurity, biosecurity and chemical weapons, and on malign foreign influence from adversaries’ AI systems.[6] On 29 September 2026 Executive Order 14434 directed executive agencies to use “Super Intelligence” and “SI” in place of “Artificial Intelligence” and “AI” in non-statutory communications.[7] As of October 2026 NIST’s page presents the centre as the Center for Advancing Innovation and Standards for Super Intelligence (CAISSI).[2]

    Testing work

    Pre-deployment testing began under the old name. In December 2024 the US and UK institutes jointly evaluated OpenAI‘s o1 across cyber, biological, and software and AI development capabilities, and shared their findings with OpenAI before release.[8] On 5 May 2026 the centre announced pre-deployment national-security testing agreements with Google DeepMind, Microsoft and xAI, adding to agreements with OpenAI and Anthropic.[4] By then it had completed more than 40 evaluations, including of models never released to the public.[9]

    Assessing foreign models

    Much of the centre’s published work compares Chinese models with the US frontier. A September 2025 study compared three DeepSeek models with four US models on 19 benchmarks and found DeepSeek’s R1-0528 12 times more likely to follow malicious instructions.[10] In May 2026 it estimated that DeepSeek‘s V4 Pro lagged the US frontier by about eight months, scoring on CAISI’s own tests like GPT-5 even though DeepSeek’s self-reported results compared it with newer models.[11] It also found V4 more cost-efficient than models of similar capability.[12]

    In July 2026 it assessed Moonshot AI’s Kimi K3 jointly with the UK AISI: on a simulated 32-step corporate network attack, Kimi K3 reached step 17 on average, against 28.5 for the most cyber-capable US models.[13] In September it called Z.ai’s GLM-5.3 the most cyber-capable open-weight model released to date, while estimating that it trailed the US frontier by about four months on cyber benchmarks.[14] These comparisons feed a wider policy debate over US technology controls; see US AI chip export controls.

    International role

    According to NIST, the centre established the international network of government AI institutes in November 2024, now the International Network for Advanced AI Measurement, Evaluation and Science.[15] In February 2026 the network published shared practices and open questions for automated evaluations, alongside CAISI’s own draft best practices for automated benchmark evaluations.[16] A fellow network member is the UK AI Security Institute, which was itself renamed from “Safety” to “Security” in February 2025.[17]

    Questions readers ask

    Is CAISI the same as the US AI Safety Institute?

    Yes. Commerce Secretary Howard Lutnick announced on 3 June 2025 that the US AI Safety Institute at NIST would become CAISI, and NIST's site now presents it as CAISSI.[1][2]

    Why is it now called CAISSI?

    A September 2026 executive order directs agencies to use "Super Intelligence" in place of "Artificial Intelligence" in non-statutory communications, and NIST's page now uses the CAISSI name.[7][2]

    Which companies' models does it test?

    In May 2026 it announced agreements with Google DeepMind, Microsoft and xAI, adding to existing agreements with OpenAI and Anthropic.[4]

    What did it find about DeepSeek?

    Its September 2025 evaluation found DeepSeek's R1-0528 model 12 times more likely than the US frontier models it tested to follow malicious instructions.[10]

    How do Chinese models compare in its tests?

    CAISI estimated in May 2026 that DeepSeek V4 Pro lagged the US frontier by about eight months overall, and in September 2026 that GLM-5.3, the most cyber-capable open-weight model to date, lagged by about four months on cyber benchmarks.[11][14]

    Sources

    Each numbered claim is a statement we checked against the sources listed with it. Status shows how well established it is.

    1. [1]

      Commerce Secretary Howard Lutnick announced on Tuesday 3 June 2025 that the US AI Safety Institute would be reformed into the Center for AI Standards and Innovation; FedScoop reported it on 4 June and Broadband Breakfast on 6 June. confirmedas of 2025-06-03

    2. [2]

      As of October 2026, NIST's web page presents the centre as the Center for Advancing Innovation and Standards for Super Intelligence (CAISSI). confirmedas of 2026-10-10

    3. [3]

      NIST describes the centre as industry's primary point of contact within the US government for testing and collaborative research on advanced AI systems, including voluntary agreements with developers and evaluations of US and adversary systems. confirmedas of 2026-10-10

    4. [4]

      On 5 May 2026 CAISI announced pre-deployment national-security testing agreements with Google DeepMind, Microsoft and xAI, adding to existing agreements with OpenAI and Anthropic. confirmedas of 2026-05-05

    5. [5]

      NIST lists among the centre's functions representing US interests internationally to guard against burdensome and unnecessary regulation. confirmedas of 2026-10-10

    6. [6]

      The rebranded centre was directed to focus on demonstrable risks such as cybersecurity, biosecurity and chemical weapons, and to assess malign foreign influence from adversaries' AI systems. confirmedas of 2025-06-06

    7. [7]

      Executive Order 14434, "Inaugurating the Era of Super Intelligence", signed on 29 September 2026, directs US executive agencies to use "Super Intelligence" and "SI" in place of "Artificial Intelligence" and "AI" in non-statutory communications. confirmedas of 2026-09-29

    8. [8]

      The UK and US AI Safety Institutes jointly evaluated OpenAI's o1 before its December 2024 release, testing cyber, biological, and software and AI development capabilities and sharing findings with OpenAI before launch. confirmedas of 2024-12-18

    9. [9]

      As of May 2026 CAISI said it had completed more than 40 evaluations, including of state-of-the-art models never released to the public. confirmedas of 2026-05-05

    10. [10]

      In September 2025 CAISI published an evaluation of three DeepSeek models against four US models across 19 benchmarks, finding the DeepSeek R1-0528 model 12 times more likely to follow malicious instructions than the US frontier models evaluated. confirmedas of 2025-09-30

    11. [11]

      The US Center for AI Standards and Innovation (CAISI) evaluated DeepSeek V4 Pro in April 2026 and estimated that its capabilities lagged the US frontier by about eight months, performing on CAISI's tests similarly to GPT-5 even though DeepSeek's self-reported results compared it with newer models. confirmedas of 2026-05-01

    12. [12]

      CAISI found DeepSeek V4 more cost-efficient than other models of similar capability. confirmedas of 2026-05-01

    13. [13]

      In July 2026 UK AISI and CAISI jointly assessed Moonshot AI's Kimi K3 and found it performed significantly below the most recent frontier cyber-capable models, reaching step 17 of a 32-step simulated network attack against 28.5 for the best US models. confirmedas of 2026-07-23

    14. [14]

      CAISI assessed Z.ai's GLM-5.3 in September 2026 as the most cyber-capable open-weight model released to date, while lagging the US frontier by about four months on its cyber benchmarks. confirmedas of 2026-09-17

    15. [15]

      The International Network of AI Safety Institutes, which NIST says the US centre established in November 2024, now operates as the International Network for Advanced AI Measurement, Evaluation and Science, with members Australia, Canada, the EU, France, Japan, Kenya, South Korea, Singapore, the UK and the US. confirmedas of 2026-02-13

    16. [16]

      In February 2026 the network published key practices and open questions for automated evaluations of AI capabilities, and CAISI released draft best practices for automated benchmark evaluations for public comment. confirmedas of 2026-02-13

    17. [17]

      On 14 February 2025 the UK AI Safety Institute was renamed the AI Security Institute, with a sharper focus on chemical and biological weapons, cyberattacks and criminal misuse, and no longer focusing on bias or freedom of speech. confirmedas of 2025-02-14

    Revision history (2)
    1. Page created.
    2. Corrected the CAISI rename date to 3 June 2025; confirmed the May 2026 testing agreements with a second outlet; added joint testing with the UK AISI and 2026 assessments of DeepSeek V4, Kimi K3 and GLM-5.3.

    Created Oct 10, 2026. Last reviewed by an editor on Oct 10, 2026. Next scheduled review: Jan 10, 2027.

    Cite this page

    "CAISI / CAISSI: the US government's AI testing centre at NIST." ContentLora, updated Oct 10, 2026. https://contentlora.com/wiki/caisi

    Spotted an error? Suggest a correction or emailcorrections@contentlora.com.

    Keep exploring

    Connections
    UK AI Security Inst… Counterpart agencyOpenAI Testing agreementAnthropic Testing agreementGoogle DeepMind Testing agreementDeepSeek-R1 Evaluated
    ☰ Contents