AI and defense are merging fast. Here’s what happened in three days.

The past 72 hours had a clear theme. The race to build bigger models is fading into the background. What’s front and center now is deployment. Who controls the infrastructure. Who wins the contracts. Who owns the data pipeline.

OpenAI is moving deep into defense.

Multiple outlets confirmed that OpenAI is in advanced negotiations with the US Department of Defense. The contract reportedly covers logistics, threat assessment and strategic planning across non-weapons systems. But critics point out that the line between support functions and combat applications gets blurry fast in modern warfare.

Sam Altman reportedly called the Pentagon deal “opportunistic and sloppy” in a private briefing that leaked to the press. It signals real internal tension about the speed at which the company is pursuing military contracts without public consultation or clear ethical frameworks.

The contract has already gone through multiple revisions following pushback from civil liberties groups and AI ethics researchers. The core concern is that large language models connected to government databases could enable mass surveillance at a scale that was previously impossible. What once required thousands of human analysts can now be automated. Existing legal frameworks were not built for that reality.

Anthropic is also in the mix. Fortune reported that both OpenAI and Anthropic have been in discussions with Pentagon officials. The competitive pressure between the two labs appears to be accelerating the timeline. Both companies face high compute costs and uncertain revenue paths. Government contracts are starting to look like a lifeline. For Anthropic, a company founded on AI safety principles, the shift toward military applications is a notable contradiction.

OpenAI is also talking to NATO.

Reuters reported on March 4 that OpenAI is considering a contract to deploy AI on NATO’s unclassified networks across 32 member states. If it goes ahead, it would be OpenAI’s first formal engagement with a multinational military alliance. EU officials are divided. Some welcome the capability boost. Others worry about American tech companies gaining deeper access to European defense infrastructure at a time when digital sovereignty is already a sensitive topic.

Qualcomm just declared 2026 the year of the AI agent.

At MWC Barcelona on March 3, Qualcomm CEO Cristiano Amon made the call directly. The chatbot era is over. The agent era is here.

The difference matters. Current AI systems respond when you ask them something. Agents act on their own. They plan across multiple steps, execute tasks across apps, learn from feedback and keep going without someone guiding every move. Book travel, handle changes, coordinate calendars, adapt when things go wrong. All without being asked each time.

Qualcomm’s Snapdragon chips are being optimized specifically for on-device agent execution. The reasons are practical. Cloud AI is slow, expensive and raises privacy concerns. Running agents locally reduces latency, keeps data on the device and works without internet. For real-time applications like autonomous vehicles or personal assistants, that’s not a nice-to-have. It’s a requirement.

The rest of the industry is moving the same direction. Google is building agent capabilities into Android. Apple Intelligence is pushing on-device AI. Microsoft Copilot and Amazon Alexa are both adding multi-step reasoning. The pattern is consistent across every major hardware and software platform.

A ResearchAndMarkets.com report published March 3 puts Agentic AI at the top of the enterprise adoption list for the coming year. The use cases breaking through right now are workflow automation across procurement, customer service and compliance. Personal scheduling and coordination across both work and life. And research and knowledge work that used to require a full team.

The agent era is not theoretical anymore. It’s being scoped, budgeted and deployed.

What part of this shift are you watching most closely?