DX Today AI Daily Brief - Monday, May 4, 2026

DX Today AI Daily Brief - Monday, May 4, 2026

Today: the Pentagon strikes classified AI deals with eight tech giants while leaving Anthropic on the bench

Transcript

SPEAKER_05 0:00

It's Monday, May 4th, 2026. You're listening to the DX Today AI Daily Brief. Today, the Pentagon Science Classified AI deals with eight tech giants while leaving Anthropic on the bench. Meta snaps up a humanoid robotics startup to fast track its physical AI ambitions. And Greg Abel uses a deep fake demonstration to put cyber risk on Berkshire Hathaway's agenda. Let's get into it.

SPEAKER_00 0:29

The Department of Defense announced Friday it has reached agreements with seven major AI companies and added an eighth hours later to deploy frontier models on its most secure classified networks. OpenAI, Google, Microsoft, Amazon Web Services, Nvidia, Elon Musk's SpaceX, the NVIDIA back startup Reflection, and later Oracle will all see their tools running in what the Pentagon calls impact level six and seven environments, the same systems that handle mission planning, intelligence analysis, and weapons targeting. Notably absent from the list, Anthropic, which the Trump administration blacklisted earlier this year over the company's insistence on safety guardrails for warfare applications. The Pentagon directed agencies to migrate off anthropic tools within six months.

SPEAKER_02 1:22

From battlefield to factory floor, Meta announced Friday it has acquired humanoid robotics startup, Assured Robot Intelligence, known as ARRI, for an undisclosed sum. The 20-person team, led by co-founders Xiaolong Wang, a former Nvidia researcher, and Leryl Pinto, founder of Fauna Robotics, will join Meta's Superintelligence Labs Effective Immediately. ARI specializes in whole-body humanoid control. AI, models that let robots understand and adapt to human behavior in messy real-world environments. Meta's pitch is not to compete in robot sales directly, but to be the platform layer for the entire industry, a kind of android for humanoids. The deal lands as forecasts for the humanoid market range from Goldman's$38 billion by 2035 to Morgan Stanley's 5 trillion by 2050.

SPEAKER_04 2:29

Now to Omaha. Berkshire Hathaway held its first annual meeting Saturday under new chief executive Greg Abel, and artificial intelligence was a central theme. Abel told shareholders Berkshire will deploy AI only in narrow value-creating use cases, saying the firm is, quote, not going to do AI for the sake of AI. He pointed to operational improvements at BNENSF Railway as a starting point and said booming data center demand is a major tailwind for the conglomerate's utility businesses. The most striking moment came when Abel showed shareholders a video clip he later revealed was an AI-generated deep fake, using the demonstration to highlight cyber risk in the agentic AI era. Analysts said Abel earned a solid scorecard in his debut. Across the Atlantic, a new flagship.

SPEAKER_01 3:21

French AI lab Mistral last week launched Mistral Medium 3.5, a 128 billion parameter dense model with a 256,000 token context window that folds chat, reasoning, and code into a single set of weights. The model scored 77.6% on CWE bench verified, putting Mistral firmly back in the frontier coding conversation. Alongside the model, Mistral launched work mode for its LaChat Assistant, an agentic mode that runs multi-step tasks in parallel across email, calendar, documents, JIRA, and Slack, with explicit human approval required before any sensitive action. The combination signals Mistral's pivot from pure model provider to enterprise agent platform.

SPEAKER_05 4:09

From Mistral, a quieter agent push.

SPEAKER_03 4:12

Enterprise AI company Writer announced Thursday a major upgrade to its writer agent platform, event-based triggers that let AI agents act on business signals without anyone prompting them. The platform now listens for events across Gmail, Gong, Google Calendar, Google Drive, Microsoft SharePoint, and Slack, kicking off multi-step workflows the moment a relevant signal arrives. Writer is positioning the release as the next stage of agentic AI, where systems move from chatbots that wait for input to autonomous coworkers that detect work to be done. The launch puts Writer in direct competition with Microsoft, Salesforce, and Amazon for the enterprise agent market. A category Salesforce alone now puts at$540 million in annual recurring revenue.

SPEAKER_05 5:02

Inference faster and cheaper.

SPEAKER_00 5:04

Cloud AI infrastructure provider RunPod also announced general availability Thursday of RunPod Flash, an open source Python SDK released under the MIT license that lets developers turn a local Python function into a live auto-scaling inference endpoint in minutes, with no containers, no images, and no infrastructure to configure. Flash supports both Q-based batch processing and load-balanced real-time inference and ships with skill packages for coding agents like Claude Code, Cursor, and Klein, so the agents can write functional deployment code on their own. RunPods has more than 750,000 developers now use its platform, with 37,000 serverless endpoints created in March alone.

SPEAKER_05 5:50

Up next, the desktop assistant push.

SPEAKER_02 5:53

Amazon Web Services last week launched Amazon QIC, a desktop AI assistant that connects to a user's local files, calendar, communications, and now Google Workspace, Zoom, Airtable, Dropbox, and Microsoft Teams without requiring a browser. QUIC can also generate documents, presentations, infographics, and images directly inside the chat. Alongside the desktop launch, AWS expanded Amazon Connect into four agencai solutions covering supply chain decisions, hiring, customer experience, and healthcare, moving the contact center product into broader enterprise workflows. AWS also brought OpenAI's frontier models, including GPT 5.5, to Amazon Bedrock and launched Bedrock Managed Agents powered by OpenAI in limited preview.

SPEAKER_04 6:57

Now to a Wall Street wobble. Shares of OpenAI infrastructure partners tumbled last week after the Wall Street Journal reported that ChatGPT maker has missed its own internal targets for both revenue and user growth. OpenAI had aimed for 1 billion weekly active users for ChatGPT by the end of last year, a figure it has yet to reach. Oracle, which signed a$300 billion five-year compute deal with OpenAI, fell roughly 4% on the news. AMD and CoreWeave also slid. Reports suggest OpenAI's chief financial officer is at odds with Chief Executive Sam Altman over how to handle the revenue gap, even as combined hyperscaler capital expenditure on AI is on track to exceed$650 billion this year. Microsoft, by contrast, is humming.

SPEAKER_01 7:49

Microsoft reported fiscal third quarter results that put numbers on the boom. Revenue reached$82.89 billion, up 18% year over year, with the company's AI business now running at a$37 billion annualized revenue rate. Azure cloud growth and copilot adoption drove the gains, and the company's commercial backlog continued to swell. Investors are watching whether Microsoft's massive capital expenditure, including the recently announced$10 billion investment in Japan and ongoing buildouts with SoftBank and Secure Internet, can keep pace with that demand without compressing margins. So far, the market is rewarding the spend.

SPEAKER_05 8:34

Turning to state policy.

SPEAKER_03 8:36

Minnesota became the first U.S. state to ban so-called nudification apps, AI tools that strip clothing from photos to generate fake nude images. The state Senate passed the bill unanimously, 65 to 0, and the governor signed it into law in the past week. The statute makes it illegal to develop, distribute, or knowingly use such tools, and authorizes the state attorney general to seek fines of up to$500,000 per flagged image. Courts can also order companies to stop offering the apps to Minnesota residents and award damages to victims for mental anguish. Advocates expect other states to follow with similar legislation in the coming months.

SPEAKER_05 9:22

Now to a major ed tech breach.

SPEAKER_00 9:51

The exposed records include names, email addresses, student ID numbers, and direct messages between users. Shiny Hunters also says it compromised in structure's Salesforce instance. The incident is the second cyber attack the company has disclosed in less than a year, raising fresh questions about the security posture of AI era ed tech platforms.

SPEAKER_02 10:14

Closing with an open weights play, Alibaba's Quen team announced Saturday a strategic partnership with inference platform Fireworks AI, debuting Quen 3.6 Plus, exclusively on fireworks outside of Alibaba's own infrastructure. Quen 3.6 Plus is Alibaba's latest flagship closed weights model and routes automatically across the Quen 3.6 family to deliver the best response for a given query. The partnership gives Western developers a low latency, low-cost path to Alibaba's frontier model and gives fireworks a meaningful exclusive amid intensifying competition from open weights rivals like DeepSeek, Moonshot's Kimi, and Z.ai's GLM. The deal is the latest sign that Chinese labs are working hard to lock in international distribution.

SPEAKER_05 11:16

That's your briefing for Monday, May 4th, 2026. For DX today, stay curious.