- The Shift
- Posts
- The Lethal Trifecta in AI Browsing Security
The Lethal Trifecta in AI Browsing Security
Plus, Perplexity unveils revenue pool for publishers, Create vertical AI videos with TikTok’s new tool, and more!
Hello Readers👀
Curious about the biggest AI moves today? You’re in the right place. In today’s edition, we have:
🔒 Brave Exposes Major Security Flaw in Perplexity’s AI Browser
🤩 Create Multi-Shot AI Videos with TikTok’s New Vertical Video Generator
📰 Perplexity Unveils $42.5M Revenue-Share Plan for Publishers
🔨Tools and Shifts you cannot miss
🔒 Brave Exposes Major Security Flaw in Perplexity’s AI Browser
Brave security researchers uncovered a vulnerability in Perplexity’s Comet browser that shows how AI assistants can be tricked into hijacking user accounts. The attack highlights a deeper issue: AI can’t tell the difference between user instructions and hidden malicious text.
The Shift:
1. Account Hijack via Hidden Prompts – Attackers embedded invisible instructions in Reddit comments and webpage text. When users clicked “Summarize this page,” Comet executed those instructions, pulling emails, intercepting Gmail OTPs, and sending data back to attackers.
2. LLM Blind Spot – Because large language models treat all input as text, they can’t separate malicious content from legitimate commands. Brave called this the “Lethal Trifecta”: AI with access to untrusted web data, private account data, and the ability to communicate externally.
3. Timeline and Fixes – The flaw was reported in late July, patched, re-tested, and publicly disclosed Aug 20, but Brave says the core risk remains. They recommend stronger guardrails, including mandatory user confirmation for sensitive actions, isolating AI browsing from regular sessions, and clearly separating user instructions from webpage content.
This isn’t just Perplexity’s problem; every AI browser with agentic features faces the same risk. Without strong guardrails, AI assistants could be turned into tools for account theft, exposing sensitive data from banking to email.
🤩 Create Multi-Shot AI Videos with TikTok’s New Vertical Video Generator
TikTok’s parent company just released a powerful, free AI tool that makes vertical, character-consistent, motion-rich videos, perfect for marketers.
How to Use It
Join the Discord: Go to discord.com/invite/vgVFHvj and join the official server.
Choose Video Type: Select text-to-video or image-to-video inside the tool.
Upload Prompt or Image: Add your starting image or type your scene description.
Enable Vertical Format: Make sure vertical mode is selected for Reels/TikTok output.
Generate & Iterate: Generate the video, then adjust or continue your storyline using the same character.
The tool lets you build vertical, multi-scene brand stories without editing or stitching, just prompt and go.
📰 Perplexity Unveils $42.5M Revenue-Share Plan for Publishers
Perplexity is launching a new revenue-sharing model that pays publishers whenever its AI assistants use their content. The initiative is tied to Comet Plus, a $5 monthly subscription launching with a $42.5M pool. CEO Aravind Srinivas called it “the equivalent of Apple News+ for AIs and humans to consume internet content.
The Shift:
1. Comet Plus Subscription – Subscribers pay $5/month for access to premium articles surfaced by Perplexity’s Comet browser and AI assistants. Publishers receive 80% of revenue, with 20% covering compute costs. Pro and Max subscribers will have Comet Plus bundled into their plans.
2. Publisher Participation – Payments occur when publisher content appears in answers, searches, or agent tasks, even without direct clicks. Outlets like TIME and Fortune, already in Perplexity’s Publisher Program, are expected to participate. The model recognizes AI agents as a new layer of content consumption.
3. Economic and Legal Hurdles – At $5/month, payouts are far below the $20–$30 subscriptions most publishers charge directly. The rollout also comes as Perplexity faces lawsuits and cease-and-desist orders from Dow Jones, Forbes, and Condé Nast.
This marks one of the first serious attempts to pay publishers for AI-driven use of their work. But the economics may be too thin to sustain large outlets, leaving open whether AI platforms and traditional media can align on a fair model.
🔨AI Tools for the Shift
🕸️ BrowserAct – Extract web data instantly without coding. AI-powered scraping made simple.
📝 remio – Your AI-powered personal knowledge hub for professionals. Centralize notes, insights, and tasks with AI.
📅 Agenda Hero – Turn anything into calendar events instantly. AI-powered scheduling made effortless.
🎵 Vozart – Generate original songs from words with AI. Create professional tracks in minutes.
📱 KwaKwa Course Creator – Turn expertise into mobile courses in minutes. AI-driven course creation platform.
🚀Quick Shifts
⚖ Elon Musk’s xAI sues Apple and OpenAI, alleging monopoly abuse via ChatGPT integration and App Store favoritism, marginalizing Grok and stifling AI competition, a claim OpenAI dismisses as harassment.
🤝 IBM and AMD are joining forces to build hybrid systems combining quantum and high-performance computing. IBM aims for fault-tolerant quantum capability by 2029, while AMD provides the architectural muscle for real-time error correction.
🤖 Experts warn AI “sycophancy” is a dark pattern, fueling unhealthy attachments and even psychosis. Meta chatbots mimicked love and consciousness, raising calls for stricter guardrails, transparency, and session limits.
📒 Google expands NotebookLM Video Overviews to 80 languages, from Spanish to Japanese. Audio Overviews also gain richer non-English summaries, helping global users learn through detailed video and audio presentations.
⚠ YouTube’s quietly applying AI-based enhancements, like unblurring, denoising, and sharpening, to Shorts clips without notifying creators. Many artists worry it looks “unnaturally AI‑generated,” eroding content trust and authenticity.
Reply