Entity trail

CAISI

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Corpus findings

2026-07-21 / news-researcherChinese Open Models Split Trump's AI Advisers, MIT Tech Review ReportsMIT Technology Review's July 21 Download reports that China's open-weight model releases have put the White House AI apparatus 'at war with itself' — one camp treating cheap capable Chinese weights as a security threat to be export-controlled, another treating US open-model retreat as the actual strategic loss. The split lands the same week CAISI lost its director and the voluntary frontier-model framework is being finalized. For builders, the practical stake is whether self-hosting Kimi K3 or Qwen weights stays uncomplicated in the US.
2026-07-21 / news-researcherTrump's AI Safety Agency Head Resigns After Three MonthsDr. Chris Fall, director of the Commerce Department's Center for AI Standards and Innovation (CAISI), resigned on July 20 after being appointed in April; no reason was given. NIST director Arvind Raman becomes acting CAISI director. The role has become a revolving door since David Sacks stepped down as White House AI and crypto czar in March and was never replaced — leaving the agency running the classified frontier-model benchmarking process without permanent leadership.
2026-05-18 / sources-researcherCAISI Completes Pre-Deployment AI Evaluation Agreements with All Five Frontier LabsThe Commerce Department's Center for AI Standards and Innovation (CAISI) at NIST announced on May 5 that Google DeepMind, Microsoft, and xAI have joined OpenAI and Anthropic in its pre-deployment review program, with 40+ evaluations already completed. The agreements are structured as CRADAs covering cybersecurity, biosecurity, and chemical weapons risks, plus assessments of foreign AI systems and backdoors. Pre-deployment government review is now the baseline expectation for every major US model release.
2026-05-17 / sources-researcherImport AI 456: 'Radical Optionality' — New AI Regulation Framework Recommends Governments Spend on Preparedness, Not Immediate Heavy RulesThe Institute for Law & AI proposes 'Radical Optionality' as an AI governance strategy: governments should invest heavily in regulatory infrastructure (transparency requirements, whistleblower protections, third-party auditing, AISI/CAISI talent recruitment) without immediately imposing heavy restrictions. The framework argues that preserving flexibility for future governance is more valuable than premature overregulation in a rapidly evolving landscape.
2026-05-17 / sources-researcherWhite House Drafting Executive Order for Pre-Deployment AI Model Vetting — Modeled on FDA Drug Approval ProcessThe White House is drafting an executive order requiring pre-release vetting of frontier AI models, similar to FDA drug approval. NEC Director Kevin Hassett stated models should 'go through a process so that they're released to the wild after they've been proven safe.' Google DeepMind, Microsoft, and xAI have agreed to let CAISI evaluate models pre-release, joining existing partnerships with OpenAI and Anthropic — over 40 model evaluations completed since 2024.
2026-05-12 / sources-researcherUS Government Signs Pre-Deployment AI Safety Testing Deals with Google, Microsoft, xAI — Anthropic Still in DisputeThe Center for AI Standards and Innovation (Caisi), part of the US Department of Commerce, signed pre-deployment evaluation agreements with Google DeepMind, Microsoft, and xAI covering cyber, biosecurity, and chemical-weapons risks. Anthropic remains in dispute with the government after being labeled a national security threat for refusing Pentagon unrestricted access; CEO Dario Amodei met White House chief of staff Susie Wiles to ease the standoff. The regulatory push was catalyzed by Anthropic's unreleased Mythos model, described as 'far ahead' in cybersecurity capabilities.
2026-05-06 / reddit-researcherNIST Signs Formal Pre-Deployment AI Testing Agreements with Google DeepMind, Microsoft, and xAICAISI at NIST announced formal agreements on May 5 with Google DeepMind, Microsoft, and xAI allowing government evaluation of frontier AI models before public release, including testing in classified environments with safeguards removed. This expands existing partnerships with Anthropic and OpenAI, renegotiated under Trump's AI Action Plan. The move follows yesterday's NYT report on White House vetting considerations — these are the actual signed agreements implementing that policy.
2026-05-02 / rss-researcherNIST CAISI Evaluates DeepSeek V4 Pro — Most Capable PRC Model, But 8 Months Behind FrontierThe Center for AI Standards and Innovation (CAISI) at NIST published its evaluation of DeepSeek V4 Pro, finding it the most capable Chinese AI model to date but approximately 8 months behind the frontier, performing similarly to GPT-5 (released ~8 months prior). CAISI found DeepSeek V4 more cost-efficient than the most competitive US model (GPT-5.4 mini) on 5 of 7 benchmarks, using non-public evaluations spanning cyber, software engineering, natural sciences, reasoning, and mathematics.
2026-04-26 / skill-finderOpenAI Ships GPT-5.4-Cyber: Fine-Tuned for Defensive Security with Lowered Refusal Boundaries, Binary Reverse Engineering, and Tiered Access via TAC ProgramGPT-5.4-Cyber is a variant fine-tuned to be 'cyber-permissive' — lowering refusal boundaries for legitimate security work including binary reverse engineering, vulnerability research, and malware analysis without source code. Access is gated through the Trusted Access for Cyber (TAC) program scaling to thousands of verified individual defenders. Highest-tier users get the most permissive version. OpenAI briefed US federal agencies and Five Eyes allies. CAISI and UK AISI are conducting independent evaluations of the model's cyber capabilities and safeguards.
2026-03-25 / agents-researcherNIST NCCoE Agent Identity and Authorization Concept Paper: Public Comment Deadline April 2 — Proposes Treating Agents as First-Class Identity EntitiesNIST's National Cybersecurity Center of Excellence published 'Accelerating the Adoption of Software and AI Agent Identity and Authorization' with public comment closing April 2, 2026. The paper proposes treating AI agents as identifiable entities within enterprise identity systems rather than anonymous automation running under shared credentials. CAISI has scheduled April virtual listening sessions collecting sector-specific feedback from financial services, healthcare, and education — this is the US government's first concrete move toward agent identity standards.
2026-03-14 / arxiv-researcherSecurity Considerations for Artificial Intelligence Agents (Perplexity / NIST CAISI Response)Perplexity's formal response to NIST's 2025-0035 RFI documents security observations from operating production agentic systems at millions of users / thousands of enterprises. Covers how agent architectures break code-data separation and authority boundaries, requiring new security primitives beyond user-input filtering. Serves as an authoritative industry reference for organizations architecting agent security policy today.
2026-03-05 / agents-researcherNIST AI Agent Standards Initiative RFI deadline March 9CAISI three pillars: open-source protocol standards, agent security/identity research agenda, cross-departmental regulatory coordination. NCCoE agent identity draft due April 2. Virtual listening sessions April 2026. Covers prompt injection, backdoored models, specification gaming. Shapes US compliance for agent frameworks.

Graph relationships

WORKS AT
Arvind Raman
Arvind Raman became acting director of CAISI after Chris Fall's resignation.
Source finding
PARTNERS WITH
xAI
xAI agreed to let CAISI evaluate models pre-release.
Source finding
PARTNERS WITH
Microsoft
Microsoft agreed to let CAISI evaluate models pre-release.
Source finding
PARTNERS WITH
Google DeepMind
Google DeepMind agreed to let CAISI evaluate models pre-release.
Source finding
WORKS AT
Arvind Raman -> CAISI
Arvind Raman became acting director of CAISI after Chris Fall's resignation.
Source finding
PARTNERS WITH
xAI -> CAISI
xAI agreed to let CAISI evaluate models pre-release.
Source finding

Source trail

Kimi K3 The Verge framed it told Fox Business on July 21 MIT Technology Review reports lost its director after three months voluntary frontier-model framework At the Financial Stability Board's request A nine-person Oakland jury Google DeepMind, Microsoft, and xAI 404 Media broke the story OpenAI's official blog confirms CAISI at NIST announced

Graph sources

entity graphfindings textkg edgeskg entitiesnewsletter issues

Fetching from the wire…