In partnership with
November 4, 2026 · READ ONLINE
Welcome back, AI enthusiasts.
Imagine paying for a premium AI model, only for the app to secretly use a cheaper one. Anthropic has introduced a way to change that, and it could have bigger implications for AI security, the tools we pay for, and the systems businesses rely on. Let’s Dive In!
IN TODAY'S ISSUE
01 Anthropic Wants to Detect When AI Gets Hijacked
02 Google Gives Schools an AI Ultimatum
03 Google’s Gemini Could Face New EU Copyright Restrictions
01 / AI IMPOSTER
Evolving AI: Anthropic unveiled FlashProve, which could expose secret model swaps by proving which AI generated a response.
KEY POINTS
An AI app could charge for a premium model while secretly sending requests to a cheaper one.
Auditors could confirm deployed models match the versions they tested.
Proofs take 20 seconds to check. Generating them uses 70–95× normal inference GPU time.
DETAILS
Anthropic’s FlashProve uses provable inference, attaching mathematical evidence to AI responses. Its zero-knowledge proofs check calculations while keeping model weights, the numbers learned during training, private. A cryptographic fingerprint acts as a model's digital ID, allowing independent checks without running the AI. Claude helped improve the system, which remains experimental and limited to short contexts and individual output tokens.
WHY IT MATTERS
Anthropic's FlashProve could help users catch hidden downgrades when paid writing or coding tools quietly switch to cheaper AI models. It also addresses a deeper security risk. Hackers who compromise AI servers could alter replies, insert malicious instructions, or impersonate trusted models. Detecting tampering could strengthen audits and vendor oversight. As AI agents access company systems, knowing which model handled a task becomes increasingly important. OpenAI, Google, and third-party platforms could offer stronger guarantees, though FlashProve's high computing costs limit its use.
Want to get the most out of ChatGPT?
ChatGPT is a superpower if you know how to use it correctly.
Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.
Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.
02 / EDU AI
Source: Bloomberg
Evolving AI: Google is giving schools more AI control, but rejecting AI-powered Search means losing Google Search access.
KEY POINTS
Gemini, Gemini Notebook, Gemini in Classroom, and Search will be off by default for K-12 student accounts.
Administrators can enable each separately, but cannot offer AI-free Google Search.
Guided Learning is available globally in Classroom, with a U.S. Search pilot this month.
DETAILS
Google revised its policy after criticism of its August Gemini rollout to younger students. Starting in early 2027, schools can set permissions by grade and product. Google says Search and AI are deeply integrated, while Guided Learning helps students work through problems instead of supplying answers.
WHY IT MATTERS
Google Search is a common starting point for research. When AI answers appear first, students may be less likely to examine original sources, potentially weakening how they evaluate evidence. These habits may carry into university and work. The wider issue is control: as AI becomes built into essential software, institutions may have to accept unwanted features or switch providers. Google's decision shows how a company's product choices can shape the way people find information.
03 / EU COPYRIGHT
Source: CNBC
Evolving AI: An EU court adviser says Google’s Gemini could violate copyright by reproducing protected news content in its answers.
KEY POINTS
Gemini may infringe publishers’ rights when its answers reproduce enough protected content to replace the original article.
Bypassing paywalls or other access restrictions to retrieve articles could also violate publishers’ rights.
AI training may qualify for the EU’s data-mining exception, subject to lawful access and copyright opt-outs.
DETAILS
Europe’s highest court is reviewing Hungarian publisher Like Company’s lawsuit over Gemini’s use of its articles. The dispute involves “grounding,” which retrieves live web content for answers. Court adviser Maciej Szpunar suggested the publisher’s example may not establish infringement, though his nonbinding opinion could shape the eventual ruling.
WHY IT MATTERS
Gemini can turn publishers’ reporting into instant answers, reducing visits to the sites that fund it. If judges agree, publishers could sue AI providers for damages or seek court orders stopping infringing outputs. Google, ChatGPT, and Perplexity could face pressure to license content and strengthen safeguards. AI companies could still summarize news facts, but reproducing protected wording would carry greater legal risk. Readers might see fewer article-like answers and more links to original sources.
04 / HUMAN OR MACHINE?
One is real. Trust your eyes?
Two images. One photograph. Click the one you think is real.
A
B
05 / QUICK HITS
🧬 Google DeepMind demonstrated AI-assisted drug discovery with Antigravity, using agents to search thousands of compounds and propose potential designs in about an hour.
⚡ AMD reported 97% growth in data-center revenue to a record $8.5 billion, driven by demand for its server processors and AI chips.
🍎 Apple opened Siri AI integrations to all developers in its iOS 27.3 beta, allowing apps to handle custom actions like bookings and orders.
🎬 Netflix launched AI Discover, a new TV tab suggesting shows by mood and theme, initially for a small group of U.S. members.
06 / TRENDING AI TOOLS
🚀 Fyxer - email assistant that organizes your inbox, writes draft replies in your tone, and takes meeting notes*.
🔐 Koreshield - Checks AI support messages and actions before they execute.
🎬 Rushilo - Edit videos by telling Claude Code or Codex what to change.
📬 Agent Inbox - Gives AI agents email, Slack and Telegram accounts for approvals.
*PARTNER LINKS
BEFORE YOU GO
Worth your four minutes?
Help us make the next edition better.
See you tomorrow,
The Evolving AI team
