The Download: AI Agent Protocols and OpenAI’s Vision for Safe AGI

By Charlotte Jee — August 5, 2025 This is today's edition of The Download, our weekday newsletter providing a daily dose of what's going on in the world of technology.

These protocols will help AI agents navigate our messy lives

A growing number of companies are launching AI agents that can perform tasks on your behalf—actions like sending emails, creating documents, or editing databases. However, initial reviews for these agents have been mixed because they struggle to interact seamlessly with the many components of our digital lives. Anthropic and Google are among the companies working to address this challenge. Over the past year, both have introduced protocols aiming to define how AI agents should interact with each other and the world around them. If successful, these protocols could form a crucial part of the infrastructure needed for AI agents to become genuinely useful. Read more about these protocols here. — Peter Hall

A glimpse into OpenAI’s largest ambitions

OpenAI has set itself a dual mandate: on one hand, it is a tech giant rooted in products, including ChatGPT, which reportedly handles 2.5 billion messages daily worldwide. On the other hand, its original mission is to function as a research lab dedicated to creating “artificial general intelligence” (AGI) and ensuring it benefits all of humanity. My colleague Will Douglas Heaven recently conducted an exclusive interview with the two key figures at OpenAI responsible for these research ambitions. The full story reveals insights into how OpenAI approaches product safety, what AGI truly means, and more. Read the full interview here. Here’s one standout insight from the conversation:

This story is from The Algorithm, our weekly newsletter all about the latest goings-on in AI. Sign up to receive it every Monday.

The must-reads

I’ve curated today’s most fun, important, and fascinating technology stories:

OpenAI is adding mental health guardrails to ChatGPT

OpenAI plans to reduce direct advice from ChatGPT and encourage users to take breaks during lengthy conversations.

*   NBC News

*   The Verge on AI medical errors

*   OpenAI’s research on ChatGPT’s emotional impact

The US wants to build a nuclear reactor on the moon

Aiming to beat Russia and China, the US plans lunar nuclear power, despite recent mission setbacks.

*   Politico

*   Engadget on NASA’s failed mission

*   Nokia’s lunar cellular network

How to live forever (or at least get rich trying)

Longevity research is booming, but many clinics sell unproven treatments.

*   New Yorker

*   MIT Technology Review on longevity clinics

Welcome to Silicon Valley’s ‘hard tech’ era

The focus shifts from consumer software to massive military contracts.

*   NYT

*   MIT Technology Review on military AI

There’s a big problem with the Gulf’s trillion-dollar AI dream

Building data centers in water-scarce regions raises environmental concerns.

*   Rest of World

*   MIT Technology Review on US data center boom

*   Quartz on Google’s energy scaling

Tesla’s board awarded about $30 billion of shares to Elon Musk

The board emphasizes retaining Musk as crucial.

*   Financial Times

*   WSJ on CEO pay

What happens if you respond to scam job texts?

The exploitation can get surprisingly strange.

*   Slate

Why there’s uproar over Vogue’s AI-generated ad

The controversy highlights ongoing debates about AI use in creative industries.

*   TechCrunch

Earth’s core seems to be leaking out of the surface

This discovery challenges long-held geoscientific assumptions.

*   Quanta

*   MIT Technology Review on volcanic brain glass

Could lasers help us see inside people’s heads?

Promising new brain imaging methods face clinical adoption hurdles.

*   IEEE Spectrum

Quote of the day

"Hate it! Don't want anything to do with it."

Business Insider

One more thing

What happened to the microfinance organization Kiva?

Read the full story

For more technology insights, subscribe to MIT Technology Review.

FAQs from the Article

AI Agents and Protocols

Q: What are AI agents and what challenges are they currently facing?

A: AI agents are software programs designed to perform tasks on behalf of users, such as sending emails or creating documents. They currently struggle to seamlessly interact with the complex components of our digital lives.

Q: Which companies are working on protocols for AI agents?

A: Anthropic and Google are among the companies developing protocols to define how AI agents should interact with each other and their environment.

Q: Why are these protocols important for AI agents?

A: If successful, these protocols could form a crucial part of the infrastructure needed for AI agents to become genuinely useful and integrated into our daily lives.

OpenAI's Vision for AGI

Q: What is OpenAI's dual mandate?

A: OpenAI operates with a dual mandate: to function as a tech giant with products like ChatGPT, and as a research lab dedicated to creating Artificial General Intelligence (AGI) that benefits all of humanity.

Q: What does OpenAI mean by AGI?

A: AGI, or Artificial General Intelligence, refers to AI that possesses human-like cognitive abilities and can understand, learn, and apply its intelligence to a wide range of tasks.

Q: What are some of the key concerns OpenAI addresses regarding AGI?

A: OpenAI focuses on ensuring that AGI benefits all of humanity and addresses product safety, the meaning of AGI itself, and its potential impact.

The Download: AI agent infrastructure, and OpenAI’s ambitions