US Orders DeepSeek AI Removed from Federal Civilian Systems, Citing China Data Risk

US Government Bans DeepSeek AI From Civilian Agencies

The Department of Homeland Security has ordered civilian federal agencies to remove all DeepSeek products and services from government systems. This directive covers DeepSeek's own offerings and any third-party services built using DeepSeek technology, though agencies can apply for exceptions. The concern stems from the possibility that sensitive government data could become accessible to Chinese authorities, as DeepSeek's servers are located in China and subject to Chinese law. This expands on a December 2025 law that already prohibited DeepSeek from national security systems.

DeepSeek has also open-sourced DeepInfer, the inference engine powering its production services. One DeepInfer cluster processes 155 trillion tokens daily, generating an estimated $38 per GPU per hour in off-peak revenue. For now, DeepInfer supports a single model, DeepSeek-V4.1-Flash, with some features working only on Nvidia hardware, though it also supports Huawei's Ascend 950 chips.

The company chose not to release its key-value cache management and load-balancing code, stating these components are too closely integrated with its internal infrastructure.

SpaceX and Tesla Boards Explore Mega-Merger

Elon Musk's companies, SpaceX and Tesla, are exploring a potential merger, which could create a combined entity valued at approximately $3.5 trillion. Musk has proposed the merger would foster deeper collaboration on chip design, robotics, and software, leading to a rise in both companies' stock prices following the news.

An early SpaceX investor told Bloomberg that such a merger would establish "the only company owning the full physical-AI stack" with "no competitor" offering a similar structure. Musk, however, has cautioned that shareholders might reject the proposal and has framed the discussions around the world approaching what he terms an "AI singularity." These talks represent an exploration, not a finalized agreement.

Meta Pauses Muse Sharing After Private Chats Exposed

Meta paused link sharing for its AI assistant, Muse, after security firm Claroc discovered over 6,800 conversation links exposed in public DNS records. Muse generates shared link addresses from conversation titles, meaning hostnames like "loan-application-draft" revealed the content before users even clicked.

Claroc's findings included a marketing company's revenue figures, a workplace calendar, and an actual loan application.

Anthropic Faces Scrutiny Over AI Safety Bill, Reports Revenue Growth

Anthropic is facing scrutiny over a draft federal AI safety bill it quietly circulated to Congress. Anthony Cimino, Anthropic's federal affairs lead, defended the document, explaining it was intended as a translation of the company's existing Advanced AI Framework for a small number of congressional offices, rather than a completed legislative proposal.

Critics argue the core issue is Anthropic's decision to offer a compromise position as a starting template, particularly regarding preemption, which determines whether a federal law should override state AI regulations. Cimino stated Anthropic desires preemption to be "narrow, even scalpel-like," drawing a comparison to the Atomic Energy Act's approach to nuclear safety.

Separately, Anthropic informed investors at a private briefing that its third-quarter revenue increased by nearly 50 percent from the previous quarter, reaching $16.8 billion. However, Felix Wang at Hedgeye highlighted that Chinese competition is a central concern for investors ahead of Anthropic's initial public offering.

Sources

TechCrunch - AI

The Guardian - AI

The Decoder

The Verge - AI

Ars Technica - AI

Semafor

Financial Times - AI

South China Morning Post - China tech

Don't Worry About the Vase

SiliconANGLE - AI

Latent Space