Google's new AI model tackles coding and cybersecurity AI news.

Noted

Your daily brief in one short note from us to you.

· October 29, 2026

Your Daily Note,

In today’s Noted:

–Google’s Gemini 4 Carbon targets tougher coding and cybersecurity problems

–Anthropic launches tools to show businesses what their AI spending is buying

–Google expands AI Search advertising internationally as its core ad growth slows

–OpenAI and ServiceNow bring AI deeper into business workflows

–Chrome’s AI Autofill can use information from Gmail to complete forms

GOOGLE DEEPMINDthis one’s big →

Google’s Gemini 4 Carbon targets tougher coding and cybersecurity problems

Image source: Google

We noted:

Google has announced Gemini 4 Carbon, a new frontier AI model designed for complex, long-running engineering work and cybersecurity defence. Google says Carbon can work through multi-step coding tasks for days, including finding and fixing subtle bugs across large codebases.

•Benchmarks. Google reports 88.2% on SEC-bench Pro for finding vulnerabilities, 54.8% on ExploitGym for exploitation tasks, and 71% on CWE-bench for patching vulnerabilities. These are Google-reported results, not guarantees of real-world performance.

•Limited access. Google is initially making Carbon available to a selected group of cyber defenders through its Fairwind programme, with broader access planned later.

•Introductory pricing. The API price is $2 per million input tokens and $10 per million output tokens, before increasing to $4 and $20 respectively.

Why it matters

Developers and security teams may eventually be able to use Carbon to investigate vulnerabilities, test fixes and handle engineering tasks that require sustained work. Most businesses cannot access it immediately, and its benchmark results should be independently assessed alongside real deployment outcomes. Frontier models are competing on more than chatbot answers, and the ability to complete complicated technical work and help secure software could become an important differentiator as AI-generated code becomes more common.

ANTHROPICshow the value →

Anthropic launches tools to show businesses what their AI spending is buying

Image source: Anthropic

We noted:

Anthropic has introduced new reporting tools that help companies understand how employees use its AI products, what they produce and how much those workflows cost. The reports use internal analytics and employee surveys to show managers team activity over a chosen period.

•Work and cost. Reports show work produced, token expenditure and points where workflows fail.

•Recommendations. Anthropic says the tools will also recommend ways to improve results.

•Pooled budgets. Teams can manage token spending more effectively through pooled budgets.

•Data use. Anthropic says this reporting data will not be used for model training by default.

Why it matters

The launch addresses a growing business problem: AI usage and costs can increase rapidly, but companies may struggle to demonstrate corresponding productivity gains. Activity and output metrics do not automatically prove that AI has saved money or improved business performance, so companies still need to measure quality, time saved and the cost of checking or correcting AI work.

GOOGLEmore ad slots →

Google expands AI Search advertising internationally as its core ad growth slows

Image source: Google

We noted:

Google is expanding advertising in its AI-powered Search experiences to international markets as it looks for new ways to monetise AI-driven searches. During Alphabet’s third-quarter earnings call, Google announced that ads in AI Mode would expand beyond the United States.

•Multi-turn formats. Google is introducing multi-turn advertising formats in the US, allowing ads to feature in more conversational search experiences.

•Slower growth. Search advertising revenue grew 12% year over year in the third quarter, down from 17% in the second quarter and 19% in the first.

•Total revenue. Alphabet reported total revenue of $127.6 billion, up 25% year over year.

•Early click-through. Google says early click-through rates for sponsored product listings in AI Mode are comparable to organic listings. This is a company-reported result.

Why it matters

Search advertising may increasingly appear within conversational answers and AI-generated recommendations, rather than only in traditional search-result layouts. The rollout does not mean every market or query will show the same ad formats immediately. For businesses, visibility may depend not only on ranking in traditional results but also on how products are presented within AI-powered search experiences.

OPENAI × SERVICENOWinside the workflow →

OpenAI and ServiceNow bring AI deeper into business workflows

We noted:

OpenAI and ServiceNow are expanding their partnership to bring AI models and agents into IT operations, application development, cybersecurity and other enterprise workflows.

•Work from ChatGPT. A new ServiceNow integration lets teams access business context and perform selected workflow actions from ChatGPT. An IT worker could find similar incidents, assess the risk of a change and update a ServiceNow record without leaving the conversation.

•More connections. The companies plan to connect Codex with ServiceNow’s AI Workflow Factory, and combine OpenAI security capabilities with ServiceNow’s asset and identity information.

•Voice and rollout. They plan voice-driven experiences for describing and assessing AI use cases, and ServiceNow plans to expand ChatGPT Work and Codex access across its workforce. Some capabilities are planned for later releases rather than being generally available today.

Why it matters

Instead of copying information between a chatbot and business software, employees will increasingly be able to investigate issues and initiate approved actions from one interface. What an AI assistant can do will still depend on the connected system, the organisation’s permissions and the availability of each feature. The business value of AI may increasingly come from completing a workflow, not simply answering a question.

GOOGLE CHROMEfewer forms →

Chrome’s AI Autofill can use information from Gmail to complete forms

We noted:

Google is adding AI-powered Autofill capabilities to Chrome that can retrieve relevant details from Gmail to help complete online forms. Instead of relying only on information saved in your browser, Chrome can suggest details found in your emails.

•What it can suggest. Flight confirmation numbers, membership IDs, tracking numbers, order references and addresses.

•Your choice. Google says sensitive data is encrypted and that users can choose whether to enable enhanced Autofill through Chrome settings.

•Rollout. The feature is rolling out to US users on desktop Chrome and in the Gemini-in-Chrome experience on iOS and Android.

Why it matters

Form filling could become easier when the required information is buried in an email confirmation or receipt. Availability depends on the platform and rollout, and users should still check the information before submitting a form. AI is becoming embedded in everyday software features, not just standalone assistants, and connecting information across apps can reduce small but frequent tasks while also raising questions about permissions and personal data access.

Products worth trying

Dedica

Check whether AI agents can actually use your website.

Your website might look perfect to people but still confuse an AI agent trying to compare plans, sign up or make a purchase. Dedica tests how AI agents navigate your site, checking for issues such as JavaScript-only pricing pages, blocked access and broken payment journeys.

It runs 125 checks across discovery, access, usability and payments, then provides a score and suggested fixes. Its free audit takes under a minute, while a one-time fix kit is offered at a launch price of $30 per website.

BotGauge

Find out where your AI agent breaks before users do.

BotGauge tests AI agents for security and reliability problems by launching targeted attacks and evaluating how they respond. It traces the prompts, tool calls and execution paths behind failures, then turns those failures into repeatable tests that can run again as the agent changes.

It supports agents built with OpenAI, Anthropic, LangChain, LangGraph, CrewAI, AutoGen and standard SDKs or APIs. The platform offers a free option.

Tulu

See which marketing spend actually brings in revenue.

Tulu connects marketing activity with revenue data to help small and medium-sized businesses understand which campaigns contribute to closed deals. Instead of making teams manually reconcile spreadsheets and tracking links, it monitors performance, flags changes such as rising customer acquisition costs and sends alerts through WhatsApp.

It supports online and offline marketing spend, integrates with HubSpot and is designed to provide insights without requiring teams to maintain complex dashboards. Tulu offers a 14-day free trial without a credit card.

Everything else in AI

Legora says AI is giving legal teams more capacity for complex work

OpenAI’s latest customer story highlights Legora, a legal AI platform that helps lawyers review documents, conduct due diligence and investigate legal questions. The company says one Fortune 100 customer reduced contract turnaround time by 90%, while Legora’s own legal team reduced the share of its capacity spent reviewing non-disclosure agreements from around 20% to under 1%.

These are company-reported results from specific workflows, not a guarantee that every legal team will see the same savings. Lawyers still need to assess the output, particularly where legal interpretation and risk are involved. AI can potentially help professionals review more material and spend more time on judgement-intensive work, rather than simply completing the same workload faster.

Wiz warns that AI could accelerate the discovery and exploitation of software vulnerabilities

In a TechCrunch interview, Wiz’s Tomer Schwartz discussed the recent PHP2Shell vulnerabilities and why the cybersecurity industry may face a growing volume of newly discovered flaws. The bugs affected PHP, an open-source language used by a large portion of websites, and required urgent investigation and patching.

Schwartz described how AI-assisted research and custom tooling helped researchers test many potential approaches quickly. The concern is that the same acceleration can benefit attackers as well as defenders, while older software may remain unpatched. For businesses, the practical takeaway is to keep software updated, monitor exposed systems and have a clear process for responding to security advisories.

Want a note in your inbox?

Join Noted for 5–7 things worth understanding in AI, delivered daily.

Free, daily, unsubscribe anytime.

See you tomorrow
same time, different note.

~ Stackari

Want to get in front of AI-curious founders?
Sponsor a note →

Was this note forwarded to you? Get your own →

Stackari · More news