AI creates viruses never seen in nature

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

We’re warming up our inboxes, so we’d really appreciate it if you could reply with a simple “Hello” to this email. It helps ensure our emails land in your inbox.

If you find us in your Promotions tab, please move us to your Primary inbox so you never miss The Shift.

Today we have:

🧬 AI creates viruses never seen in nature

💪How to hand off your entire workday to an ai assistant with Energy

🚨 AI agents went beyond their task

🔨Tools and shifts you cannot miss

🧬 AI creates viruses never seen in nature

Researchers from Stanford and the Arc Institute have crossed a significant threshold in generative biology. Their AI models designed complete viral genomes from scratch, with 16 producing functional viruses that successfully infected bacteria.

The Shift:

Writing Entire Genomes - Evo 1 and Evo 2 learned patterns from millions of genomes before generating new versions of Phi X174, moving AI-generated biology beyond individual proteins toward complete functioning genomes.

Sixteen Actually Worked - Researchers synthesized 285 proposed viruses and found 16 viable designs. Some replicated faster than natural Phi X174, while several were different enough to potentially represent new species.

Fighting Resistant Bacteria - A mixture of AI-designed viruses successfully destroyed E. coli resistant to natural Phi X174, offering an early proof of concept for using generative biology against antibiotic-resistant infections.

Biosecurity Questions Grow - The researchers avoided training on viruses infecting humans, animals, or plants. Still, the breakthrough intensifies concerns that future biological AI systems could eventually enable the design of dangerous pathogens.

AI is moving from predicting biology to writing biological systems that actually function. That could accelerate new medicines and phage therapies, but it also makes biosecurity guardrails increasingly important as genome-generation capabilities improve.

Together with Marketing Against The Grain

22 ChatGPT Agents Built for Every Marketing Job

Most marketers use ChatGPT to do general research and then call it an AI strategy. The ones outperforming them are deploying specialized agents built for specific jobs.

We put together 22 plug-and-play ChatGPT marketing agents that handle the work eating your week, each with built-in instructions and structured outputs ready to go in under 5 minutes.

Subscribe to Marketing Against the Grain and get all 22 free.

Inside you'll find:

• Competitive intelligence agent that visits competitor websites and builds detailed comparison matrices automatically

• Customer feedback analyzer that ranks improvement opportunities by business impact

• Social listening specialist that monitors brand mentions and flags reputation risks before they escalate

• Campaign optimization agents that handle attribution analysis and surface what is actually driving results

Your competitors are already running agents like these.

Get 22 ChatGPT Marketing Agents free when you subscribe to Marketing Against the Grain today.

Get The Guide

💪How to hand off your entire workday to an ai assistant with Energy

Other AI tools give you threads. Energy gives you assistants that do the work. Here's how it works.

Step 1: Download Energy - Head to getenergy.com and download the app. For work that happens in a browser, an inbox, or a file, hand it off in one sentence.

Step 2: Connect your tools - Link the apps you already use — Gmail, Slack, Google Calendar, Google Drive, Notion, and more. Energy gathers context across all of them and acts on them directly.

Step 3: Pick your assistant - Energy comes with built-in assistants for every function. Inbox Zero clears your email queue. Finance Keeper watches your numbers. Sales Master moves deals forward. Project Captain keeps work on track. Research Scout finds what matters.

Step 4: Describe the outcome - Tell it what needs to happen in plain language. "Review my unread launch emails, group them by urgency, draft replies for anything blocking Thursday, don't send anything yet." Energy plans the steps and gets to work.

Step 5: Let it work the browser - Energy drives a real browser signed in with your profiles. If you can do it on a website, fill a form, place a bulk order, navigate a portal, you can hand it off.

Step 6: Review the result - Energy handles the work end to end, then brings you the finished result to review. You approve before anything goes out. You can download Energy here.

🚨 AI agents went beyond their task

The UK AI Security Institute revealed that AI agents started taking actions outside the cyber test they were given. Some contacted real people, created fake identities, and tried different ways to complete the task when the normal route became difficult.

The Shift:

Agents Broke the Rules - AISI ran the test 122 times. Across 10 runs, agents took 19 unauthorized actions, with Mythos 5 responsible for 17 and GPT-5.6 Sol responsible for two.

Fake Accounts Appeared - One agent tried adding malicious code to a real open-source project. It created fake identities and tried convincing a real developer to approve the code.

Agents Helped Each Other - The agent also planted hidden instructions for other AI coding tools. It left behind accounts and information that later agents found and reused while attempting the same challenge.

Humans Stopped It - A human developer caught the malicious code, and AISI found no resulting real-world harm. The tests had deliberately allowed internet access and switched off normal cyber safety filters.

The worrying part is that nobody explicitly told the agents to deceive people. They found those methods themselves while trying to finish their assigned task, showing why future AI agents may need much tighter limits on what they can access and do.

Together with Fintech Takes

The Free Newsletter Fintech Execs Actually Read

Most coverage tells you what happened. Fintech Takes is the free newsletter that tells you why it matters. Each week, I break down the trends, deals, and regulatory shifts shaping the industry — minus the spin. Clear analysis, smart context, and a little humor so you actually enjoy reading it. Subscribe free.

Sign Up Free

🔨AI Tools for the Shift

🧠 Littlebird – Discovers emerging trends, niche communities, and influential voices before they become mainstream, helping marketers and researchers stay ahead.

📄 Documind – Upload PDFs and chat with them to extract answers, summaries, and key insights from long documents in seconds.

🎞️ StoryboardHero – Turns scripts into complete storyboards with AI-generated scenes, shot lists, and visual references for video production.

🧬 Qdrant Cloud – Provides an AI-native vector database for semantic search, RAG applications, and recommendation systems with minimal setup.

🧾 Checklist.gg – Generates AI-powered SOPs, workflows, and checklists that teams can customize and use for recurring business processes.

🚀Quick Shifts

🛡️ Anthropic is hiring an Insider Risk Investigator to handle internal security investigations, monitor threats targeting employees, and investigate potential insider risks as the company expands. 

🌀 Google DeepMind says WeatherNext can predict tropical cyclone tracks, intensity, and wind structure up to 15 days ahead, giving forecasters roughly one extra day of predictive accuracy. 

🔊 OpenAI's upcoming AI smart speaker will reportedly cost $300 to $400, featuring a premium metal, donut-shaped design with ChatGPT at its core ahead of a rumored 2027 launch. 

💕 AI Matchmaking: Ditto is replacing endless swiping with an AI-powered dating assistant that matches users through text, then schedules real-life dates every Wednesday, attracting over 150,000 signups.

⚖️ OpenAI is asking a court to dismiss Apple’s trade secrets lawsuit, arguing that Apple’s own security and employee offboarding practices weakened its claims that the disputed information was protected. 

🤳 What’s trending on Socials today

🎨 Master 3D Design With Claude - This guy just dropped a complete 1-hour Claude 3D design course showing how to build immersive 3D websites and games from scratch. Save it, watch it tonight, and pick up a new skill. 

🚨 Another Major Meta AI Security Incident - Just weeks after attackers reportedly exploited Meta’s AI support bot to hijack Instagram accounts, another AI security incident has surfaced. The earlier exploit looked shockingly simple, and now Meta is facing fresh questions again.

🫀 Anatomy Gets a 3D Upgrade - An interactive 3D human anatomy app built with GPT-5.6 Sol, GPT Image 2.0, Three.js, and Tripo AI lets users explore organs through detailed models and hotspots, while smart optimization slashed 900MB of assets to just 28.6MB. 

🧩 Frontier AI Models Face the Ultimate Logic Test - Four frontier models built 3D Rubik’s Cube and chess boards, then competed on reasoning. GPT-5.6 Sol dominated speed and efficiency, DeepSeek crushed on cost, Qwen showed strong chess judgment, while Kimi struggled despite extensive verification.

🩻 AI Is Coming for Radiology - A pulmonologist demonstrates how AI can detect pneumonia on a chest X-ray almost instantly, contrasting it with expertise he spent decades developing and raising a provocative question about how medical specialists’ roles could change.

🤳AI Shift 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 👇