Gemini 3.8 Flash, Azure’s $29.4B Quarter, and AI Security Funding

Compact Conversations for 2026-09-02: 6 AI stories, ai news worth knowing in just 5 minutes.

[Audio embed placeholder]

The Lead: Google launches Gemini 3.8 Flash and a specialized cybersecurity variant

Google announced Gemini 3.8 Flash, its latest reasoning and coding model, and a specialized variant called Gemini 3.8 Flash Cyber focused on vulnerability detection and automated patching. The cybersecurity model is available to trusted defenders through a new Fairwind Program.

Why it matters: The release highlights the push for more capable, cost-effective models for enterprise tasks and the growing focus on specialized AI for critical areas like security, where Google claims the model can find vulnerabilities in hours instead of months.

Source: Google Blog

Number to Know: Microsoft discloses Azure revenue: $29.42 billion for the June quarter

Microsoft will start disclosing Azure’s quarterly revenue, reporting $29.42 billion for the June quarter, a 42% year-over-year increase. This comes as the company restructures its financial reporting into two segments.

Why it matters: The move provides investors clearer insight into Microsoft’s core cloud business, which is a major competitor to AWS and Google Cloud, and highlights Azure’s growth, fueled in part by demand from AI labs like OpenAI.

Source: CNBC

The Feed

HiddenLayer raises $100M as demand for AI security tools surges

AI security startup HiddenLayer raised a $100 million Series B. The company makes tools to protect AI models and agents from attacks, and its annual recurring revenue grew more than 10x over the past year.

Why it matters: Investment is flowing into AI security as deployments scale, with Gartner estimating companies will spend $2.83 billion on securing AI tools this year, an 83% increase from 2025.

Source: TechCrunch

Broadcom’s AI chip revenue more than triples, forecasts major growth

Broadcom reported Q3 revenue up 86% year-over-year to $29.59 billion, driven by AI semiconductor revenue that soared 221% to $16.7 billion. The company designs custom chips for Google, Meta, and OpenAI.

Why it matters: The earnings underscore the massive infrastructure demand from AI labs and cloud providers, with Broadcom planning to double its AI revenue to $115 billion in the next fiscal year.

Source: CNBC

OpenAI integrates ChatGPT Health with Epic’s electronic health records

OpenAI launched an integration between ChatGPT Health and Epic’s EHR system, allowing clinicians to import patient data for summarization and questioning. The integration is read-only.

Why it matters: This move deepens AI’s role in healthcare workflows by letting clinicians quickly synthesize patient information, though the company maintains AI is not for diagnosis or treatment.

Source: TechCrunch

AI labs increase testing for biological risks, a harder challenge than cybersecurity

Frontier AI labs are stepping up testing for biological risks, concerned that future models could help users create novel pathogens. This testing is more difficult than cybersecurity testing as it cannot be confined to digital environments.

Why it matters: As AI capabilities advance, labs and regulators are grappling with how to evaluate and mitigate potential misuse in physical domains like biology, where real-world testing poses unique challenges.

Source: Financial Times

One Thing to Try

Review the prompts you use most with AI coding assistants. Try adding a one-line system instruction at the top of your chat, such as ‘You are an expert software engineer who writes secure, well-documented code.’ Being more specific about the role, framework, or task can steer the model’s tone and focus for more immediately useful output.

Sources

Transcript

Host A: Welcome to Compact Conversations, the show that compresses the day’s AI news into 5 minutes.

Host A: [curious] Today’s lead is Google’s latest model release. The company announced Gemini 3.8 Flash, which it calls its best reasoning and coding model yet, and a specialized variant called Gemini 3.8 Flash Cyber. The standard Flash model is priced the same as its predecessor, at 75 cents per million input tokens and three dollars and seventy-five cents per million output. Google says it shows significant gains in software engineering and agentic tasks, and claims it outperforms larger models on some benchmarks, including long-horizon coding.

Host B: The cybersecurity variant is the more notable part of the release. [with emphasis] Gemini 3.8 Flash Cyber is focused on vulnerability detection and automated patching. It’s not generally available; Google is offering it to trusted defenders through a new Fairwind Program. The company says internal tests show it can find a critical vulnerability in under two hours, a process that usually takes months.

Host A: One number to know today is 29.4 billion dollars. That’s Microsoft’s Azure revenue for the June quarter, up 42 percent year-over-year. [thoughtful] CNBC reports this comes as Microsoft starts disclosing Azure’s quarterly revenue for the first time, giving investors a clearer view of its cloud business competing with Amazon Web Services and Google Cloud.

Host B: The company is restructuring its financial reporting, trimming segments from three down to two: Agents and Infra, and Devices and Consumer. Under this change, Azure excludes products like GitHub cloud services to become, as CEO Satya Nadella wrote, a more purely consumption-based platform. Analysts estimate about half of Azure’s growth this past fiscal year came from OpenAI.

Host A: In other news, AI security is seeing major investment. TechCrunch reports that Austin-based startup HiddenLayer has raised a 100 million dollar Series B round. The company makes tools to protect AI models, agents, and workflows from attacks like prompt injection and malicious code.

Host B: [conversational] The CEO told TechCrunch the startup’s annual recurring revenue grew more than ten times over the past year, now in the tens of millions. The funding comes as Gartner estimates companies will spend 2.83 billion dollars this year on products to secure AI tools, an 83 percent jump from 2025.

Host A: Broadcom’s earnings show where the chip demand is coming from. The company reported third-quarter revenue up 86 percent to 29.59 billion dollars. [with a small lift] The big driver was AI semiconductor revenue, which more than tripled year-over-year to 16.7 billion dollars. Broadcom designs custom chips for Google, Meta, and OpenAI.

Host B: CEO Hock Tan said on a call that Broadcom expects to double its AI revenue to 115 billion dollars in the next fiscal year. He also outlined plans to deliver tens of billions of dollars worth of processors to Google annually and continue shipments of the custom Jalapeno chip developed with OpenAI.

Host A: OpenAI is connecting ChatGPT Health with Epic’s electronic health record system. This lets clinicians import patient data to ask questions and get summaries. The integration is read-only, so the AI doesn’t write back to records.

Host B: Finally, the Financial Times reports that frontier AI labs are increasing their testing for biological risks. [with emphasis] The concern is that future models could help users create novel pathogens. This testing is harder than cybersecurity testing because it can’t be done solely in digital environments.

Host A: One thing to try is to review the prompts you use most often with AI coding assistants. [lighter] A simple trick from developer forums is to add a one-line system instruction at the top of your chat, like ‘You are an expert software engineer who writes secure, well-documented code.’

Host B: The idea is to be more specific than a generic ‘help me code.’ You could specify a framework, like ‘You are a React expert who prioritizes accessibility,’ or a task, like ‘You are a DevOps engineer explaining cloud costs.’ It steers the model’s tone and focus before you even ask your first question, making the output more immediately useful.

Host A: That’s Compact Conversations for Wednesday. More AI news tomorrow. Until then, happy prompting.