AI just tackled a major software security challenge

Google's AI proved Signal's core encryption code works

Google DeepMind just used AI to mathematically verify the code protecting billions of private conversations.

That's a serious achievement.

Its new AlphaProof Code system checked Signal's core cryptographic implementation in four weeks, a process that has historically taken expert teams years. AI agents generated over six million lines of mathematical proofs, while human reviewers examined the 1% flagged for closer inspection.

The interesting part is the depth.

Most security verification checks whether a protocol's design is sound, while this system went further and checked whether the actual compiled code delivers the protections it promises.

I think this is one of the more convincing uses of AI in software engineering because the results can be independently checked by a mathematical proof system. The researchers also reintroduced two previously fixed bugs, and the system correctly identified both as insecure.

There's still a limitation: these proofs cover specified security properties, so they don't guarantee that every possible vulnerability has been eliminated.

But if this approach scales, verifying critical software could become dramatically faster.

How much software would pass the same test?

Adobe wants its creative tools to work inside AI chatbots

Adobe is making it possible for AI assistants like ChatGPT, Claude, and Gemini to access more of its creative tools directly.

That's worth paying attention to.

At its MAX event, Adobe introduced a new Model Context Protocol (MCP) integration alongside AI features across Photoshop, Premiere, Illustrator, and Firefly.

Users will be able to access more Adobe functions through external AI assistants, while new tools inside its apps handle tasks like removing objects from video, adjusting lighting, and editing specific parts of images.

I think the MCP announcement matters most because it could change how people use Adobe's software, especially if they can request edits through a chatbot instead of navigating multiple menus.

Adobe is also testing Builder, an AI prototyping tool in Firefly that puts it in more direct competition with Figma.

The challenge will be execution. Creative professionals still need precise control over the final result, and Adobe hasn't decided how much functionality to expose through these integrations.

Will creators eventually spend more time giving instructions to AI assistants than working directly inside Adobe's apps?

OpenAI's latest acquisition points toward AI-powered travel

OpenAI just acquired Odessia, a travel startup whose AI agent was designed to help people plan trips.

The bigger story is the founder.

Francis Davidson, who previously built the short-term rental company Sonder, is joining OpenAI after raising $6 million for Odessia and launching its travel agent just five months ago.

Odessia has already shut down, and OpenAI hasn't announced any new travel products.

This looks like a talent acquisition.

What's interesting is Davidson's experience running an actual travel business, including handling the operational challenges that eventually contributed to Sonder's bankruptcy.

I think OpenAI sees an opportunity to make ChatGPT more useful beyond recommending hotels and building itineraries, especially when users need help comparing options and completing bookings.

With Booking.com, Expedia, and KAYAK already among OpenAI's partners or customers, some of the industry relationships are in place.

The next question is practical: will ChatGPT eventually let you plan and book an entire trip without switching between multiple websites?

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.