The guardrails inside OpenAI’s War contract

OpenAI draws firm lines in Pentagon deal

OpenAI just signed a classified AI deal with the Department of War.

The company will deploy its models in cloud-only environments, keep its safety stack on, and place cleared OpenAI engineers inside the workflow. The contract explicitly bars mass domestic surveillance, autonomous weapons control, and high-stakes automated decisions, and references DoD Directive 3000.09 from January 25, 2023.

This is OpenAI saying it wants to be the grown-up in the room while still taking the defense money. The cloud-only setup and keeping their own staff in the loop show they do not trust policy alone to hold under pressure.

It also puts quiet pressure on Anthropic, which walked away from similar talks and warned about weak safeguards in earlier drafts.

If this structure holds, it becomes a template for how frontier labs plug into classified work without stripping safety layers. If it slips, the backlash will be loud and bipartisan, and it will land fast.

The real test is simple: who notices first if those red lines start to blur?

NVIDIA just took AI-RAN out of the lab and onto live 5G networks

T-Mobile, SoftBank and Indosat ran over-the-air trials using NVIDIA’s AI-RAN platform, including a 3.7GHz deployment and a 16-layer massive MIMO test. SynaXG says it hit 36 Gbps throughput with under 10 milliseconds latency on a single GH200 server, running both RAN and AI workloads together.

This is NVIDIA trying to turn the telecom network into another GPU data center, only this one sits at the edge and handles real traffic. If operators can run AI and radio workloads on the same box, that changes the economics of 5G and sets up 6G around software from day one.

At Mobile World Congress next week in Barcelona, with demo screens glowing across the hall, they are clearly aiming to lock in that architecture early.

If this model works at scale, carriers buy more GPUs and fewer purpose-built boxes. If it struggles, operators retreat to safer, single-purpose gear.

The real question is whether telecom operators want to become cloud companies.

DeepSeek signals another fast AI push

DeepSeek is about to release another AI model.

PYMNTS reports the Chinese lab is on the verge of unveiling its latest system, though details on size and capabilities are still thin. The story sits alongside coverage of OpenAI’s new Pentagon deal and Anthropic’s legal fight, which shows how tightly AI and geopolitics are now linked.

DeepSeek has moved quickly over the past year, and each release tightens the race between U.S. and Chinese labs. Even without specs, the timing matters because every new model resets expectations around cost, performance, and who is leading.

You can feel the pace picking up, like late at night when the office lights are still on and engineers are pushing one more build.

If the model is competitive, it adds pressure on U.S. labs already navigating defense contracts and political scrutiny. If it underwhelms, the market shrugs and waits for the next release cycle.

The question is whether this launch changes the leaderboard or just keeps the race crowded

Prohuman team

Free Guides

Explore our free guides and products to get into AI and master it.

All of them are free to access and would stay free for you.

Feeling generous?

You know someone who loves breakthroughs as much as you do.

Share The Prohuman it’s how smart people stay one update ahead.