- The Summary AI
- Posts
- 🚀 Claude for Chrome
🚀 Claude for Chrome
PLUS: “Nano Banana” Tops Image Editing

Welcome back!
Browsers are becoming the next wave for AI, and Anthropic just introduced Claude for Chrome, an AI agent that can click, type, and manage tasks directly inside your Chrome browser. But serious security risks still remain when giving AI complete access to the browser. Let’s unpack…
Today’s Summary:
⚡ Anthropic previews Claude for Chrome
🚀 Google’s Gemini “Nano Banana” tops image editing
đź’§ How much energy one AI prompt uses
🗣️ ElevenLabs launches v3 TTS with emotion control
📦 xAI open-sources older Grok 2 model
đź“· Meta releases DINOv3 image model
🛠️ 2 new tools

TOP STORY
Anthropic previews Claude for Chrome
The Summary: Anthropic has opened a preview of Claude for Chrome, an AI agent that can click, type, and act directly in the Chrome browser. The pilot begins with 1,000 Max plan users, who will stress-test its safety systems and test the ability to directly manage tasks like emails, scheduling and form-filling.
Key details:
Pilot limited to 1,000 Max subscribers, waitlist open for others
Brave security researchers recently found Comet, Perplexity’s AI browser, vulnerable to prompt injection, allowing attackers to extract emails and OTP codes across open tabs
The attacks work by hiding malicious instructions in social media, for example asking the AI to fetch OTP codes and passwords from Gmail in another open tab and send the data to the attackers
Current Claude for Chrome mitigations reduced attack success rates from 23.6% to 11.2%, but anything above zero is too high
Why it matters: Web browsers may become the operating system for AI agents, but not until someone will solve the security puzzle. Anthropic’s very cautious rollout shows that the main goal is to improve security of browser-integrated AI, so as users can be sure it won’t backfire. If Anthropic finds a solution to these risks, it will set a big precedent for AI automation that scales beyond APIs.

FROM OUR PARTNERS
Build Your Own AI Agent in 2 Minutes
The Simplest Way To Create and Launch AI Agents
Imagine if ChatGPT and Zapier had a baby. That's Lindy.
With Lindy, you can build AI agents in minutes to automate workflows, save time, and grow your business. From inbound lead qualification to outbound sales outreach and web scraping agents, Lindy has hundreds of AI agents that are ready to work for you 24/7/365.
Stop doing repetitive tasks manually. Let Lindy's agents handle customer support, data entry, lead enrichment, appointment scheduling, and more while you focus on what matters most - growing your business.
Join thousands of businesses already saving hours every week with intelligent automation that actually works.

Google’s Gemini “Nano Banana” jumps ahead in AI image editing
The Summary: Google has rolled out Gemini 2.5 Flash Image, also known by its test alias “Nano Banana”, which debuted at the top of the LMArena image editing with a score of 1362, leaping past OpenAI, Alibaba and Flux. The model has character consistency across edits, natural language precision, and the ability to blend elements from multiple images.
Key details:
Gemini 2.5 Flash Image scored 1362 ELO in Image Editing LMArena, a huge +171 jump above Flux-1 Kontext and OpenAI’s GPT-Image-1
Faster than ChatGPT image model
1024Ă—1024 output resolution
First model to merge more than two input images cleanly
Early testers described it as “the GPT-4 moment for image editing”
Available now to free and paid Gemini users
Why it matters: Gemini 2.5 Flash Image shows how quickly AI models are evolving from eye candy into practical editing tools. By collapsing what once took hours in Photoshop into a few prompts, it will change design workflows, product visualizations, up to clothing try-ons and family photo restoration at scale. Its accessibility inside Gemini’s chat interface positions it as a casual daily tool, instead of a costly enterprise product.

FROM OUR PARTNERS
Never Forget a Conversation Again
Your Memory: Now instantly searchable
Limitless is the wearable AI that instantly captures, transcribes, and remembers every meeting, conversation, and brilliant thought you have - securely, seamlessly, and privately. Transform how you work, live and never miss a beat again.

INSIGHT
How much energy does a single AI prompt use?
The Summary: Google has released a detailed accounting of how much energy, water, and carbon each Gemini AI text prompt consumes. A single median query uses 0.24 watt-hours, emits 0.03 grams of CO₂, and requires 0.26 milliliters of water, far below many outside estimates. The findings show that Google’s AI systems became extremely energy-efficient in just one year, and reveal the complexity of measuring AI’s environmental impact.
Key details:
One Gemini text prompt is like 9 seconds of TV in energy
Each prompt requires about five drops of water for cooling
Only 58% of this energy is for the AI; CPUs, memory, idle systems, and cooling account for the rest
Energy demand per prompt fell 33x in one year
Why it matters: AI’s environmental story is usually told in hand-waving abstractions; this report pins it down to exact drops of water and fractions of watt-hours. The numbers also expose a paradox: energy per query is decreasing but since overall AI demand strongly increases, net energy use still increases. By choosing a clear methodology for energy accounting, Google sets the frame of the debate, showing how the way metrics are defined can shape the narrative of AI’s environmental cost.

QUICK NEWS
Quick news
ElevenLabs launches v3 text-to-speech with emotion controls
xAI open-sources Grok 2 model weights, promises Grok 3 in 6 months
Meta releases DINOv3 image model for commercial use

TOOLS
🥇 New tools

That’s all for today!
If you liked the newsletter, share it with your friends and colleagues by sending them this link: https://thesummary.ai/