U.S. expands pre-release AI safety testing program to Google, Microsoft and xAI models
- Both agree
- Early access lets federal scientists evaluate unreleased frontier AI models for capabilities and security risks before public launch, extending an existing arrangement to more major companies and shifting oversight earlier than post-release scrutiny alone would allow.
- They split on
- Whether the story is about building public safeguards around advanced AI before deployment, or about keeping security testing voluntary rather than letting it become a more formal government review process.
Analytical frames for this storyTap to explore
The Facts
- Google DeepMind, Microsoft and xAI agreed to provide the U.S. government early access to unreleased AI models for review before public release.
- The reviews will be conducted through the Center for AI Standards and Innovation (CAISI) within the U.S. Department of Commerce.
- Under the agreements, CAISI will conduct pre-deployment evaluations and targeted research to assess model capabilities and security risks.
- The new agreements expand an existing voluntary testing arrangement that already included OpenAI and Anthropic.
- A central purpose of the testing is to examine national security-related risks from advanced AI systems, including cybersecurity threats.
- The program matters because it gives federal scientists a way to assess frontier AI systems before launch, rather than only after the tools are already public.
- The arrangements are voluntary agreements with the companies, not a mandatory government approval system for AI models.
- What remains unresolved is whether the Trump administration will move beyond these voluntary agreements to create a more formal government review process for AI models.
Context
What will the government actually do with these AI models?
CAISI says it will perform pre-deployment evaluations and targeted research on unreleased models to better understand their capabilities and security risks before launch POLITICO,Reuters,CNN International.
Which companies were already participating before this expansion?
OpenAI and Anthropic were already voluntarily working with CAISI to test unreleased models for vulnerabilities, and the new deals extend that program to Google DeepMind, Microsoft and xAI Reuters,BBC,U.S. News & World R….
Why is the U.S. focusing on these reviews now?
Sources say officials are focused on demonstrable risks from advanced AI, especially cybersecurity and other national security dangers, and some reports tie the urgency to concerns raised by Anthropic's newly unveiled Mythos model Reuters,Al Jazeera Online,CNN International.
View all 100 sources
Wire services (6)
Independent coverage (50)
HürriyetYapay zekada devlet denetimi dönemi başlıyor
驱动之家美国政府将干预新AI模型发布：全球五大前沿实验室均已加入
Anadolu AjansıGoogle, Microsoft, xAI to give US early access to AI models ...
ai.zhiding.cnOpenAI 网络安全专用模型 GPT-5.5-Cyber 即将上线，仅面向可信防御者开放
ElNacional.catAcuerdo en los Estados Unidos: Google, Microsoft y xAI somet...
news.cn新闻分析丨美政府为何酝酿收紧AI监管？
20 minutosGoogle, Microsoft y xAI, bajo lupa: EEUU revisará sus modelo...
NewsBytesOpenAI gives US government access to GPT-5.5 for security te...
Notícias ao MinutoEUA assinam acordo para rever IAs da Google, da Microsoft e ...
驱动之家比DeepSeek V4体量大5倍 中国AI距离最强AI模型Mythos还落后6-12个月
www.diariolibre.comEE. UU. acuerda con Google, Microsoft y xAI revisar sus mode...
Insurance JournalMicrosoft, Google and xAI to Give Government Early Access to...
The New York TimesLa Casa Blanca plantea revisar los modelos de IA antes de su...
The Star US to assess new AI models before their release
Hoy DigitalEU revisará modelos IA antes publicarse
BloombergHTGoogle, Microsoft ve xAI, ABD'ye yapay zeka modelleri için t...
MyJoyOnline.comUS to safety test new AI models from Google, Microsoft, xAI ...
FirstpostWhat we know about US stress tests of Google, xAI and Micros...
东方财富网白宫与三家科技巨头签订协议：将获得谷歌、微软和xAI未发布模型
SiliconANGLEGoogle, Microsoft and xAI agree to allow government safety c...
The Times of IndiaAnthropic Mythos fears make Google, Microsoft and Elon Musk'...
Android HeadlinesGoogle, Microsoft & xAI to Give US Government Early Access t...
Brisbane TimesTrump goes 'woke' with a sudden change of mind
Nikkei ChineseAnthropicCEO:中国AI追上Mythos需6-12个月
infobaeFuncionarios federales probarán los modelos de IA de Google ...
ExpansiónTrump impone el "filtro" del gobierno: Evaluará los nuevos m...
RTTNewsU.S. Government Steps Up AI Oversight With New Testing Agree...
NextgovCommerce AI center will evaluate Google Deepmind, Microsoft ...
CNN InternationalMicrosoft, Google and xAI will let the government test their...
Son DakikaMicrosoft ve Google, ABD ile Yapay Zeka Anlaşması İmzaladı
A HaberABD'den yapay zeka hamlesi: Dev şirketlere erken test izni
Anadolu AjansıGoogle, Microsoft ve xAI, ABD'ye yapay zeka modelleri için t...
HaberlerGoogle, Microsoft ve xAI, ABD'ye yapay zeka modelleri için t...
Milli GazeteGoogle, Microsoft ve xAI, ABD'ye yapay zeka modelleri için t...
DiarioBitcoinGoogle, Microsoft y xAI acuerdan dar acceso anticipado de su...
BloombergHTTeknoloji devleri, ABD hükumetiyle güvenlik anlaşması yaptı
Urgente 24Trump gira 180 grados: el gobierno que desreguló la IA evalú...
Semana.com Últimas Noticias de Colombia y el MundoEl gobierno de EE. UU. intervendrá en los nuevos modelos de ...
GEO TVMicrosoft, Google and xAI to give US government early access...
La TribuneLa Maison-Blanche veut voir les nouveaux modèles d'IA avant ...
TecMundoEUA terão acesso antecipado às novas IAs de big techs para t...
The GuardianUS announces deals with tech firms for national security rev...
UPIU.S. government to test AI models, expand oversight - UPI.co...
FirstpostMicrosoft, Google, and xAI grant US early access to AI model...
LExpress.frDans une volte-face, l'administration Trump examine la possi...
Fox BusinessTrump admin to review AI models from Google, Microsoft, xAI ...
BBCUS to safety test new AI models from Google, Microsoft, xAI
Senego.com - Actualité au Sénégal, toute actualité du jourÉtats-Unis : Google DeepMind, Microsoft et xAI ouvrent leurs...
Portafolio.coGobierno de EE. UU. tendría entre sus planes revisar los nue...
Al Jazeera OnlineMicrosoft, Google, xAI give US access to AI models for secur...
About these frames
The Architect: Stability, law, enforcement, institutional design, separation of powers, regulatory process, rule of law. How are order and governance maintained?
The Guardian: Sanctity, degradation, bodily autonomy, moral boundaries, human dignity, bioethics, environmental purity. Where are the lines that should not be crossed?
The Watchdog: Wrongdoing, responsibility, corruption, transparency. Who knew what, when, and what they did about it.
See this differently than someone you know would? Two ways to keep it going.
The dial works on any URL — paste an article you read elsewhere this week.