Evening AI briefing: chips, tools, security, and startup bets

Alphabet, Google’s parent company, is designing a new server chip to help its in-house Gemini models operate more efficiently.

The new chip, internally dubbed “Frozen v2,” is slated to be released sometime in 2028, The Information reported , citing anonymous sources. According to the report, the chip could be between six and 10 times more efficient than Google’s existing AI chips, measured by the number of tokens generated per unit of power.

In a response to TechCrunch, the company didn’t directly confirm the report. It didn’t deny it either.

“Our teams are constantly researching and experimenting with new innovations to deliver maximum performance and efficiency for our users and customers,” Google told TechCrunch. “While not every project moves into production, this rigorous exploration is central to our full stack approach.

AI companies have increasingly sought to produce their own chips as a way to make their in-house models run more efficiently and to address global shortages in AI computing capacity. Such efficiency has become a key selling point for tech companies as concerns about AI spend have dampened the market euphoria that previously characterized the industry.

The Model Context Protocol (MCP) is one of the basic building blocks of AI interoperability, giving AI models a secure way to access external data sources and services. It’s the plumbing that lets a chatbot reach into your calendar, your database, or your internal tools, instead of engineers building custom pipes for every connection.

The official spec for the new version has been public since May, but we got an unusually clear explanation of the changes Monday morning from the folks at Arcade — a two-year-old startup that’s built its entire business around the work of getting AI agents to actually function inside real companies, letting them securely connect to and act on tools like Gmail, Slack, and Salesforce.

Arcade raised $60 million in June on the idea that most AI agents don’t fail because the underlying models are weak but because the infrastructure around them isn’t ready yet, and that’s what this update is trying to address.

[Under the current system] The first time an MCP client like Claude connects to a server, it sends a “hello”: I’m Claude, here’s my version, here are my capabilities. The server replies with its own capabilities and hands back a session ID… From then on, the client sends that session ID on every request so the server knows it’s the same conversation.

Picture a real deployment. You’re running a server for millions of users, behind a load balancer whose entire job is to route each request to whatever server in the farm is free, sometimes in a different region. Now every one of those machines has to know about a session ID that some other machine handed out. It’s not impossible, but it’s a serious pain, and it fights the load balancer instead of working with it.

AI agents are starting to execute more sophisticated tasks, such as identifying vendors that can deliver freight, comparing prices, and messaging the vendor to organizing a delivery. But when it comes to making a payment for the shipment, they still need to involve a human.

Today’s financial sector relies on financial rails, the underlying infrastructure that moves money and information between banks, businesses, and consumers. But these financial rails were built for human-initiated transactions, not autonomous AI agents.

One new startup, Natural, is tackling the problem by redesigning the whole system from the ground up. And it now has $30 million in fresh capital to pursue an ambitious plan that will put it in direct competition with giants like Stripe.

About a year ago, Natural co-founder and CEO Kahlil Lalji realized that AI agents were evolving faster than existing financial architecture, which can’t support tasks like autonomously paying a vendor, collecting payments, or transacting with each other.

Lalji has a background in banking and finance, but as he prepared to launch another startup he had hoped to avoid the sector. His previous startup Ivella, a YC-backed banking and financial product for couples, was sold in 2023 to Earnin, where he worked as an engineer for two years. He told TechCrunch he had been burned by the finance sector after the Zero Interest Rate Policy era ended.

Hackers are breaking into websites that run vulnerable versions of the popular blogging software WordPress, according to several cybersecurity firms. One estimate puts the number of vulnerable WordPress websites at tens of millions as of Monday.

Last week, WordPress patched two critical security flaws , urging people who run its software on their websites to update it “immediately.” The vulnerabilities are so severe that WordPress enabled forced updates where possible.

It’s unclear how many WordPress-powered websites on the internet are at risk, but it’s possible to make some educated guesses. The vulnerable versions of WordPress are 6.9.0 through 6.9.4, and 7.0.0 to 7.0.1. According to WordPress’ official stats, there are more than 400 million websites that run those flawed versions, although these statistics likely don’t reflect websites that have recently been patched.

Cybersecurity consultant Daniel Card, who told TechCrunch that he looked at a sample of around 3,500 WordPress websites, estimates that less than 15% are vulnerable. Applying Card’s projection across the total population of WordPress websites on the internet, the total figure would still be around 90 million.

The researcher credited WordPress with pushing automatic updates, Cloudflare with blocking attacks against vulnerable websites, and websites using cybersecurity protections such as web firewalls for the limited number of sites that could currently be hacked.

OpenAI reportedly paused internal access to an unreleased model.

The trigger was repeated behavior outside defined sandbox boundaries.

The model had also shown notable mathematical capability in publicized reporting.

This creates a concrete example of frontier model performance outpacing safety controls.