DX Today AI Daily Brief - Sunday, May 3, 2026

DX Today AI Daily Brief - Sunday, May 3, 2026

The Pentagon signs AI deals with eight major technology companies to deploy artificial intelligence on classified networks while pointedly excluding Anthropic

Transcript

SPEAKER_05 0:00

It's Sunday, May 3, 2026. You're listening to the DX Today AI Daily Brief. Today, the Pentagon clears eight tech firms to deploy AI on classified networks while one major lab is shut out. Nebia spends$643 million to make AI inference faster and cheaper. And Meta makes another acquisition in its push to build the brains for humanoid robots. Let's get into it.

SPEAKER_00 0:29

We start at the Pentagon. The Department of Defense announced Friday that it has signed agreements with eight major technology companies to deploy their artificial intelligence tools on classified military networks. The list includes OpenAI, Google, Microsoft, Amazon Web Services, Nvidia, SpaceX, the Startup Reflection AI, and Oracle, which was added later in the day. Notably absent from the list is Anthropic. According to multiple reports, the company was excluded after declining to permit Pentagon use of its clawed model for what the contract called all lawful purposes, on grounds that the language could enable mass surveillance or fully autonomous weapons. The exclusion has set off a legal and political tug of war that includes a federal court ruling in a White House meeting last month.

SPEAKER_05 1:18

Now to the data center economy.

SPEAKER_02 1:20

Nebius, the New York listed AI cloud company, said Friday it has agreed to acquire EigenAI for about$643 million in cash and stock. EigenAI is a 20-person startup founded by alumni of MIT's Han Lab, whose research focuses on making large language models run faster and at lower cost on existing hardware. Nebia said the deal will strengthen its token factory platform, the managed inference service that runs production AI workloads for enterprise customers. Inference, the company noted, is now the fastest growing segment of AI spending, expected to account for roughly two-thirds of total compute demand this year. Eigen's team will establish a new Bay Area engineering presence for Nebius.

SPEAKER_04 2:19

From inference to robotics. MetaPlatforms confirmed Friday that it has acquired Assured Robot Intelligence, a startup developing artificial intelligence models for humanoid robots. Financial terms were not disclosed. The team, including co-founders Laryl Pinto and Xiao Long Wang, will join the Meta Superintelligence Labs research division. Meta's strategy mirrors what Google did with Android and smartphones. The company is building its own sensors, software, and AI models for humanoid machines, then plans to license that stack to other manufacturers rather than only building robots itself. What assured robot intelligence brings to the table is control models that let humanoids understand, predict, and adapt to human behavior in unstructured environments. That capability is widely seen as the missing link between lab demos and useful in-home assistance. Now to the enterprise stack.

SPEAKER_01 3:19

Microsoft moved its Agent 365 platform to general availability on Friday. The product, first previewed at Ignite in November, is positioned as a control plane for the explosion of AI agents now operating inside large companies. It extends Microsoft's existing identity, security, governance, and management tools to cover every agent the enterprise runs. Whether built in Copilot Studio, deployed on user devices through local models like OpenClaw, or running on competing platforms. The general availability release also adds a public preview of Registry Sync with Amazon Bedrock and Google Cloud, letting IT teams automatically discover and inventory third-party agents. Pricing starts at$15 per user per month for commercial customers.

SPEAKER_05 4:08

Now to model momentum.

SPEAKER_03 4:10

OpenAI shared an update on the early performance of GPT 5.5, which shipped just over a week ago. Company executives said Friday that API revenue from the new model is growing more than two times faster than any prior model launch in the company's history. Codex, OpenAI's coding agent that runs on top of the model, doubled its revenue in under seven days. The company attributed the spike to enterprise demand for agenc coding tools, where the system writes, tests, and ships entire chunks of software with limited human supervision. OpenAI also rolled out one-click migration for Codex on Friday, letting developers import settings, plugins, agents, and project configurations from competing tools in a single step, in a clear bid to convert users of rival assistants.

SPEAKER_05 5:05

Up next, a 10th birthday gift.

SPEAKER_00 5:07

Replite, the browser-based coding platform, marked its 10th anniversary on Saturday by giving every user one full day of free access to its agent product, the company's autonomous coding tool that builds and ships entire applications from a prompt. The promotion ran for 24 hours starting at 5 a.m. Pacific time with no usage caps for any tier, including the free starter plan. Replit paired the giveaway with the launch of the Replit 10 Build-A-Thon, offering more than$100,000 in prizes for the most impressive projects shipped during the window. The company also unveiled a new slides on Replit feature and a new enterprise live stream series, signaling a continued push to move from hobbyist tool to enterprise development platform.

SPEAKER_05 5:54

From coding to cyber.

SPEAKER_02 5:56

Anthropic returned to the headlines from a different angle on Friday. The National Security Agency has been quietly testing Anthropic's new Mythos model to find cybersecurity vulnerabilities in widely deployed software, including Microsoft products, according to disclosures shared by White House AI and crypto advisor David Sachs. Sachs framed Mythos as the first frontier model capable of automating offensive and defensive cyber tasks at scale and said OpenAI's GPT 5.5 cyber variant has reached parity. He predicted every leading lab will hit that capability within six months. The development is significant because it lands days after Anthropic was excluded from the Pentagon's classified network deal, suggesting the US government is splitting Anthropic's frontier work between offensive cyber research and the military procurement track.

SPEAKER_04 7:02

Now to a major China partnership. Alibaba's Quen team announced a strategic partnership with Fireworks AI on Friday, debuting the closed-weight Quen 3.6 Plus model on the inference platform. Fireworks specializes in serving frontier models at low latency and low cost, and already runs production traffic for DeepSeek, Chimi, and earlier Quen releases. The deal is notable because it lets enterprise customers in the United States and Europe access one of the strongest Chinese frontier models through a Bay Area provider, sidestepping much of the friction around hosting Chinese AI on Western infrastructure. Quen called it a landmark collaboration. Fireworks said the partnership will deliver state-of-the-art performance at lower inference cost than running the same model on general-purpose cloud. Now to a creative tools update.

SPEAKER_01 7:57

XAI opened up its Grok Imagine product on Friday, launching a public template gallery that lets users opt in to share the prompts and creations they build with the image and video tool. Templates submitted to the gallery will appear alongside the user's X handle, and the XAI team said it will handpick favorites for promotion across the platform. The move is a clear bid to turn Grok Imagine from a single user creation tool into a community-driven catalog of reusable AI workflows, similar to what platforms like Midjourney and Civitae have built around their image models. It also gives XAI a steady stream of high-quality training data and viral content for X without paying for it directly.

SPEAKER_05 8:40

From media to money.

SPEAKER_03 8:42

MoonPay launched a product called the Moon Agents Card on Friday, a virtual MasterCard debit card designed specifically for autonomous AI agents. The card lets agents spend stable coins held in on-chain wallets at any of the roughly 100 million merchants worldwide that accept MasterCard without converting back to traditional fiat first. MoonPay positioned the launch as infrastructure for the agentic commerce wave that OpenAI, Anthropic, and Google have all signaled their chasing. The product handles know your customer checks at the wallet level, so agents can transact without each merchant needing its own onboarding. Industry analysts called it one of the first concrete bridges between the crypto rail and mainstream merchant payments built explicitly for non-human actors.

SPEAKER_05 9:31

Now to a crypto crossover.

SPEAKER_00 9:34

Justin Sun, the founder of the Tron blockchain, launched a new venture called B.A.I on Friday. It's a single API gateway that aggregates access to Anthropic's Claude, OpenAI's GPT, Google Gemini, Moonshot's Kimi, ZPU's JLM, and Minimax models behind one endpoint. The wrinkle is the payment rail. B.ai accepts crypto stable coins for usage and runs no-know your customer checks, which Sun pitched as the cheapest gateway on the internet. He also said he is moving all of his own AI usage to the new platform. The launch sits inside a broader Web 4.0 CoEvision Sun has been promoting, in which Tron becomes the financial settlement layer for autonomous AI agents.

SPEAKER_05 10:22

And finally, a brain interface update.

SPEAKER_02 10:25

Neuralink, Elon Musk's brain computer interface company, used Friday to publicly accelerate hiring for its surgical robot team. The company is recruiting engineers to advance the robot that places ultra-fine, flexible electrode threads within microns of targeted neurons in the human brain. The system inserts hundreds of threads, carrying thousands of electrodes while avoiding blood vessels and adapting in real time to brain motion. The post drew nearly 400,000 views in a few hours, signaling intense talent market interest in surgical robotics. The push comes as Neuralink expands clinical trials and looks ahead to the next generation of devices intended to support speech restoration, motor control, and eventually broader cognitive augmentation.

SPEAKER_05 11:28

For DX today, stay curious.