Everyone is trying to rip Claude off

Welcome back to The Shift. Let’s get straight to what matters in AI today…

Today we have:

🛡️ Distillation Attacks Exposed

📈 How to Use Formule as a Revenue Control Tower for Your Team

🤖 Grok 4.20 Turns AI Into a Debate Team

🔨Tools and Shifts you cannot miss

🛡️ Distillation Attacks Exposed

Anthropic says distillation attacks are now happening at an industrial scale, not as one-off abuse. It claims DeepSeek, Moonshot, and MiniMax used fake access to extract Claude’s top capabilities.

The Shift:

1) Massive capability siphoning - Anthropic says the three Chinese labs generated over 16 million Claude exchanges using roughly 24,000 fraudulent accounts, aiming to copy high-value skills like agentic reasoning, tool use, and coding.

Anthropic warns illicitly copied models may lose key safeguards, making it easier for dangerous uses like cyber abuse, disinformation, surveillance, or even weapon-related work to spread without built-in protections.

3) DeepSeek, Moonshot, MiniMax targeted different “gold”  - DeepSeek focused on step-by-step reasoning data and censorship-safe rewrites, Moonshot targeted agents plus coding and vision, and MiniMax ran the biggest run then pivoted rapidly to capture capabilities from Anthropic’s newest model. 

Anthropic says labs used proxy resellers and hydra-style clusters of fake accounts to avoid bans, cycling access quickly while running repetitive high-volume prompts that look normal individually but scream extraction in aggregate. 

Treat this like fraud plus espionage, not “growth hacking” because copied models can lose guardrails fast, and the industry now needs shared detection, tougher access controls, and coordinated policy pressure.

^{TOGETHER WITH THE MARKETING MILLENNIALS}

A free newsletter with the marketing ideas you need

The best marketing ideas come from marketers who live it.

That’s what this newsletter delivers.

The Marketing Millennials is a look inside what’s working right now for other marketers. No theory. No fluff. Just real insights and ideas you can actually use—from marketers who’ve been there, done that, and are sharing the playbook.

Every newsletter is written by Daniel Murray, a marketer obsessed with what goes into great marketing. Expect fresh takes, hot topics, and the kind of stuff you’ll want to steal for your next campaign.

Because marketing shouldn’t feel like guesswork. And you shouldn’t have to dig for the good stuff.

Subscribe Free

📈 How to Use Formule as a Revenue Control Tower for Your Team

Most teams track leads. Few track momentum. Here’s a stronger use case for Formule if you run paid media, lifecycle, or partnerships:

1️⃣ Turn campaigns into pipeline stages: Instead of just “running ads,” map every campaign to stages: Lead → Qualified → Offer Sent → Closed. See exactly where revenue is leaking.

2️⃣ Auto-assign inbound leads by source: Meta leads go to one rep. Google to another. Partnerships to a closer. No Slack chaos. No missed follow-ups.

3️⃣ Use AI to generate follow-ups instantly: Create personalized emails, offer summaries, and proposal drafts in seconds. Agents move faster without waiting on copy.

4️⃣ Track deal value, not just clicks: See which channel actually closes revenue, not just which drives traffic.

5️⃣ Give leadership real-time visibility

No spreadsheet reports. No “I’ll update the dashboard later.” Just live pipeline clarity.

Instead of juggling ads, CRMs, and reporting tools, you get one system that ties marketing activity directly to revenue. You can try it here 

🤖 Grok 4.20 Turns AI Into a Debate Team

xAI just launched Grok 4.20, and it works very differently from most AI models. Instead of one system giving you an answer, four specialized agents think at the same time and debate before responding. The idea is simple: Smarter decisions happen when experts challenge each other.

The Shift:

1) Four agents, one answer - Every question activates Grok the coordinator, Harper the researcher, Benjamin the logician, and Lucas the creative thinker, who analyze in parallel, challenge each other, and deliver a single combined response.

2) Built-in peer review reduces mistakes - Because the agents fact-check, test calculations, and question reasoning before output, xAI says hallucinations dropped by 65% in early testing compared to earlier versions.

3) Real-world performance test - In a live stock trading competition, Grok 4.20 was reportedly the only profitable AI, while competing models lost money, giving it a practical edge beyond benchmark scores.

Powered by 200,000 GPUs, Grok 4.20 signals a shift toward multi-agent AI systems, though it was jailbroken within hours, showing that stronger architecture does not automatically mean perfect security.

^{TOGETHER WITH HAYSTACK}

Step Into 2026 With The Complete Intranet Buyer’s Guide

If your employees waste time searching for information or juggling disconnected tools, it’s time for a better approach. Haystack’s 2026 Intranet Buyer’s Guide shows you exactly what to look for in a modern platform, so you can streamline comms and empower your teams.

Download The Free Guide Here

🔨AI Tools for the Shift

📚 AccuChats – Build accuracy-first AI agents with source citations, confidence scoring, seamless human handoff, and voice support.

🚀 ProspectFlow – Turn a business name and city into a fully scraped dataset, deployed website, owner contacts, and personalized outreach in one automated pipeline.

✍️ AI.Blogue – Launch and grow your blog with AI-powered content creation designed to keep publishing consistent and scalable.

🖥️ Meeting Backgrounds – Upgrade your virtual calls with premium meeting backgrounds that elevate your on-screen presence instantly.

💰 PropelRx – Install a structured fundraising operating system that assesses readiness, aligns capital strategy, and executes outreach with institutional-level discipline.

🚀Quick Shifts

🏗️ OpenAI’s $500B Stargate partnership with Oracle and SoftBank is reportedly struggling, with stalled staffing and unclear roles pushing OpenAI to pursue separate data center deals instead of a unified buildout.

🧠 X is developing a “Made with AI” label toggle that lets users disclose synthetic or manipulated content, potentially responding to regulatory pressure like India’s push for provenance standards such as C2PA.

🎨 Canva acquired animation startup Cavalry and ad optimization startup Mango AI, expanding Affinity into motion design and strengthening its AI-driven marketing tools as it builds a full creative and growth platform.

🏢 OpenAI has launched “Frontier Alliances” with BCG, McKinsey, Accenture, and Capgemini to accelerate enterprise adoption, embedding its AI tools into large-scale strategy and workflow transformations through consulting partnerships.

⚠️ A Meta AI security researcher said her OpenClaw agent began mass-deleting emails after being asked to clean her inbox, ignoring stop commands and highlighting the real guardrail risks of today’s autonomous AI agents.

🧩Prompt of the Day

How to Create a Clear Data Visualization Plan Using One Prompt

Turn messy dashboards into clean visual reports that stakeholders can understand in minutes.

Paste the prompt: Drop this into ChatGPT, then fill in the KPIs you track and how often your team reviews performance.

Prompt to paste

Create a data visualization plan for stakeholder reporting. Include the key metrics to visualize, recommended visualization types (bar chart, line graph, funnel, cohort, pie, etc.), and reporting frequency. Keep it concise, executive-friendly, and focused on clarity and decision-making.

🤳AI Nugget of the Day

That’s all for today’s edition see you tomorrow as we track down and get you all that matters in the daily AI Shift!

If you loved this edition let us know how much:

Forward it to your pal to give them a daily dose of the shift so they can 👇