Welcome back! The walls around AI are starting to crack. Brave just exposed hidden attacks inside AI browsers like OpenAI’s Atlas, showing how invisible text in images can quietly take control of an agent. Meanwhile, Claude is learning finance, Google is dominating creative AI tools, and OpenAI wants to make music with Juilliard. The race to integrate AI into everything is moving faster than its guardrails can catch up.
In today’s Generative AI Newsletter:
• Hidden vulnerabilities surface inside AI browsers
• Claude expands into real-time financial analysis
• Google tightens its grip on creative AI tools
• OpenAI experiments with machine-made music
Latest Developments
Brave Exposes Hidden Attacks in AI Browsers, Prompt Injections Found in OpenAI’s Atlas and Others
Image source: Paloaltonetworks
A new Brave Security report has exposed serious vulnerabilities in AI browsers like OpenAI’s Atlas, showing how agents can be silently hijacked through hidden prompts and screenshots. Researchers discovered that invisible text embedded in images can issue commands to AI assistants, while ordinary web pages can trick the model into leaking data or taking unintended actions.
Here’s what Brave uncovered:
Screenshot Prompt Injection hides text within images that the AI interprets as direct instructions. Attackers can use this method to exfiltrate data or perform actions on behalf of users.
Navigation-Based Injection occurs when an AI agent opens a website that quietly feeds its text back into the model, altering the next step in the workflow.
Systemic Weakness Across AI Browsers shows how few boundaries exist between what a user says and what the web says, leaving assistants open to silent manipulation.
Microsoft’s Gaming Copilot raised similar alarms after automatically capturing gameplay screenshots for “context,” a feature that’s difficult to turn off.
These flaws make it possible for attackers to manipulate agents that have access to sensitive user accounts, emails, or even financial information. One malicious prompt could cause real-world damage. The same tension is surfacing in business. Without clear disclosures, markets cannot assess the true exposure behind this AI gold rush.
Special highlight from our network
From idea to deployment: AI tools finally work together
Sinapsis is the only all-in-one AI platform built for real-world use. From data ingestion to model training and deployment, everything runs in one drag-and-drop interface.
- Use pre-built templates or start with full agents.
- Connect your models and logic with ease.
- Deploy to edge or cloud without custom builds.
Large-scale AI rollouts now take days instead of months.
Claude Learns Finance, From Excel to Market Data
Image Source: Anthropic
Anthropic is extending Claude into financial work with new tools that make analysis faster and more reliable. The update introduces an Excel add-in, live market data connections, and built-in Agent Skills for modeling, research, and valuation. Inside Excel, Claude can read and edit spreadsheets, trace its reasoning, and build new models while keeping formulas intact.
Here’s what’s new:
Claude for Excel lets users analyze and update spreadsheets inside Microsoft Excel.
New Connectors link Claude to data from Aiera, LSEG, Moody’s, Chronograph, Egnyte, and other providers.
Agent Skills include tools for valuation, earnings summaries, and due diligence reports.
Enterprise Beta gives early access to 1,000 users from Max, Enterprise, and Teams accounts.
The company says it wants Claude to handle the routine work that slows analysts down. Claude’s expansion shows where Anthropic thinks the real value of AI lies. It is embedding intelligence inside the slow, precise work that keeps financial systems running. In a field built on accuracy and speed, it is learning the value of doing less and doing it well.
Google Dominates AI Media Tools, Survey Finds Gemini and Veo Leading in 2025
Image Credit: Artificial Analysis
Benchmarking firm Artificial Analysis has published its 2025 “State of Generative Media” report, revealing a major lead for Google in both image and video AI creation. The survey, which polled 300 developers and creators, found that Google’s Gemini is the top model for AI image generation, used by 74% of respondents, while Veo leads video tools with 69% adoption.
Here’s what the report shows:
Adoption Trends: 89% of personal creators now use AI image tools in their workflows, while 58% have adopted AI video tools.
Enterprise ROI: 65% of organizations reported a return on investment within 12 months, and 34% already see profits from AI media projects.
Selection Priorities: 76% of personal users ranked model quality as their top factor, while 57% of enterprise teams cited cost reduction.
Market Share: Gemini dominates image generation, and Veo leads video creation, outpacing OpenAI, Midjourney, and Chinese models such as Kling.
These results suggest that Google has built a stronghold in generative media, a space long associated with OpenAI and Midjourney. Even with a modest sample size, the high ROI figures hint that creative and commercial use of AI media is maturing faster than the broader enterprise AI market.
OpenAI Is Building a Music Generator, Partnering With Juilliard Students
Image Credit: The Information
OpenAI is developing new AI models for music creation, marking its third attempt to enter the audio space after earlier experiments with MuseNet and Jukebox in 2019 and 2020. The company has enlisted students from the Juilliard School to annotate musical scores. The goal is to build models that can generate original music from text prompts, layer instrumentals over vocals, or compose full soundtracks for video content.
Here’s what’s in the works:
Juilliard Collaboration is producing detailed training data for a wide range of instruments and arrangements.
Text-to-Music Generation will allow users to create songs or soundtracks from written prompts.
Commercial Uses include advertising jingles and soundtrack creation for media companies.
Strategic Push positions OpenAI against AI music startups Suno and Udio, which currently lead the space.
The timing, however, may be off. Many artists and listeners see AI-generated music as hollow, stripped of the human choices that make art feel alive. Musicians worry about models trained on their work without credit or consent, while audiences call the results derivative and soulless. Earning trust from a culture that values emotion over automation could be its hardest challenge yet.
AI companies are soaring. You can still get a stake before this wave crests.
Bonus-share deadline extended — your extra 22% is still within reach.
Join GenAI.works today, before the offer closes on Nov 21.
GenAI.works is:
• Building the infrastructure for agentic AI
• Serving 200+ companies and 13M+ users
• Bootstrapped to $3M ARR
👉 Take your stake in the boom.
Reg CF offering via DealMaker Securities. Investing is risky.
🚀 Boost your business with us. Advertise where 13M+ AI leaders engage!
🌟 Sign up for the first (and largest) AI Hub in the world.
📲 Follow us on our Social Media.
