Anthropic and the U.S. government go to war: OpenAI partners with the U.S., claims additional safeguards

Analytics DeepLearning.AI -

In today’s edition of Data Points, you’ll learn more about:

  • Nano Banana 2, Google’s fast, powerful image generator
  • GitHub Copilot’s multi-model and security scanning updates
  • Perplexity Computer, a premium multipurpose agent
  • OpenAI’s multibillion-dollar partnership with Amazon

But first:

U.S. Defense Department ends Anthropic government contracts

Department of War secretary Pete Hegseth designated Anthropic a supply chain risk following months of failed negotiations over two exceptions the company requested to government use of Claude: mass domestic surveillance of Americans and fully autonomous weapons systems. Anthropic says it refuses to permit capabilities the company contends today's frontier AI models cannot reliably and safely execute. Anthropic argues the supply chain risk designation is unprecedented — historically reserved for U.S. geopolitical adversaries, not American companies — and legally unsound. The company states it has supported US warfighters since June 2024 and will challenge the designation in court, refusing to abandon its position regardless of pressure from the Department of War. (Anthropic)

OpenAI reaches agreement with U.S. to use AI in war and intel

OpenAI announced a deal with the U.S. Department of Defense (also called the Department of War) to deploy advanced AI systems in classified environments. OpenAI said the deal incorporates stronger safeguards than previous military AI agreements. The company established three red lines: no use for mass domestic surveillance, no autonomous weapons direction, and no high-stakes automated decisions like social credit systems. OpenAI will enforce these these through cloud-only deployment, control over safety systems, cleared personnel involvement, and contractual protections anchored to existing law — an approach the company argues is more responsible than competitors (e.g., xAI) who have reduced safety guardrails in defense deployments. OpenAI requested the Pentagon make equivalent terms available to all AI companies and specifically asked the government to resolve its dispute with Anthropic, which had been designated a supply chain risk by the DoD. (OpenAI)

Google updates its Nano Banana image generator

Google launched Nano Banana 2 (Gemini 3.1 Flash Image), merging the advanced capabilities of Nano Banana Pro with the speed of Gemini Flash. Google says the model includes advanced world knowledge grounded in real-time web search, precision text rendering with translation support, and improved subject consistency for up to five characters and fourteen objects in a single workflow. It also offers enhanced instruction following, production-ready output from 512 pixels to 4K resolution, and richer visual fidelity compared to the original Nano Banana. Nano Banana 2 replaced the Pro model in the Gemini app, with Pro now available only via regeneration for Ultra subscribers, The model also expanded to Search and Lens in 141 new countries and eight additional languages, and launched in preview for AI Studio, the Gemini API, Vertex AI, and as the default model in Google Flow at zero credits. (Google)

GitHub Copilot’s coding agent gets smarter and safer

GitHub has shipped five major updates to its Copilot coding agent, making autonomous background development work faster and more reliable. The agent now includes a model picker for choosing between quick models for routine tasks and powerful ones for complex work. The agent also self-reviews its own code before opening pull requests and provides free and automatic security scanning for vulnerabilities, secrets, and dependencies. Teams can create custom agents by defining workflows in local agents files to enforce specific processes, and developers can move work between cloud and local environments without losing context. Upcoming features include private mode, planning-before-coding, and non-PR tasks like issue summarization. (GitHub)

Perplexity mixes multiple models in generalist computer use agent

Perplexity released Perplexity Computer, a browser-based interface that orchestrates multiple AI models from different providers into coordinated workflows. Users describe desired outcomes, and the system autonomously deploys sub-agents for research, document creation, data processing, and API interactions. The core reasoning engine runs Claude Opus 4.6, with supplementary access to Gemini, Grok, ChatGPT 5.2, Nano Banana for image generation, and Veo 3.1 for video. Each task executes in isolated environments with its own browser, file system, and tool connections. Perplexity positions this approach around the thesis that modern workflows require specialized models rather than single general-purpose agents. The service is available as part of Perplexity’s Max plan at $200 per month, competing with similar agentic interfaces like OpenClaw and Claude Cowork. (Perplexity)

OpenAI cuts deal with Amazon to deploy custom models and agents

OpenAI and Amazon announced a multi-year strategic partnership including a $50 billion investment from Amazon ($15 billion upfront, $35 billion conditional) and an expansion of their existing cloud infrastructure agreement by $100 billion over 8 years. The companies will jointly develop a Stateful Runtime Environment powered by OpenAI models for Amazon Bedrock, enabling developers to build AI agents with persistent context, memory, and access to compute—launching within months. AWS becomes the exclusive third-party cloud provider for OpenAI Frontier, OpenAI’s enterprise platform for deploying teams of AI agents across business systems. OpenAI will consume approximately 2 gigawatts of AWS Trainium capacity to support these services and other advanced workloads. OpenAI and Amazon will also develop customized models for Amazon’s consumer-facing applications, complementing Amazon’s existing Nova family. This deal values OpenAI at $730 billion and consolidates AWS as OpenAI’s primary infrastructure provider, while also giving Amazon developers more access to frontier AI capabilities. (OpenAI)

Want to know more about what matters in AI right now?

Read the latest issue of The Batch for in-depth analysis of news and research.

Last week, Andrew Ng talked about the release of the Skill Builder tool designed to help individuals assess and enhance their AI skills, and the importance of having guidance in navigating the rapidly evolving AI job landscape.

“There are many job opportunities in AI! Employers are eager to hire people with AI skills. But the landscape of AI technology is large, growing, and rapidly changing. To navigate this landscape, many people find that occasional conversations with a knowledgeable, trusted mentor are helpful for deciding where to go next.”

Read Andrew’s letter here.

Other top AI news and research stories covered in depth:

A special offer for our community 

DeepLearning.AI recently launched the first-ever subscription plan for our entire course catalog! As a Pro Member, you’ll immediately enjoy access to:

  • Over 150 AI courses and specializations from Andrew Ng and industry experts
  • Labs and quizzes to test your knowledge
  • Projects to share with employers
  • Certificates to testify to your new skills
  • A community to help you advance at the speed of AI

Enroll now to lock in a year of full access for $25 per month paid upfront, or opt for month-to-month payments at just $30 per month. Both payment options begin with a one week free trial. Explore Pro’s benefits and start building today!

Try Pro Membership

Data Points is produced by human editors with AI assistance.