DeepMind wants to test models before launch

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

Today we have:

🤖 DeepMind Calls for AI Safety Body

🧐How to Actually Understand the AI Code You're Shipping With LoWisa

⚠️ 'We Must Act Now' on AI

🔨Tools and Shifts you cannot miss:

🤖 DeepMind Calls for AI Safety Body

DeepMind CEO Demis Hassabis proposed creating a new U.S. standards body to evaluate frontier AI models before release, aiming to establish a formal safety framework as AI systems become increasingly powerful.

The Shift:

Independent Standards Body - Hassabis proposed a FINRA-style organization that would work with government agencies, national labs, and independent experts to assess frontier AI models before deployment.

Frontier Model Testing - The proposed body would evaluate advanced models for deception, cybersecurity risks, biological threats, malicious capabilities, and attempts to bypass safety guardrails.

Capability-Based Oversight - The framework would apply based on a model's capabilities, not whether it is open or closed source, with safety benchmarks updated regularly as AI advances.

Voluntary Before Mandatory - Frontier AI labs would initially submit models for review up to 30 days before release, with Hassabis suggesting the process could later become a formal requirement.

Hassabis said the standards body should begin operating this year, arguing that AI capabilities are advancing quickly enough to require independent safety testing, common evaluation standards, and a framework that could coordinate responses if frontier development begins outpacing safeguards.

Together with Gladly.ai

Same AI software. Wildly different results.

Every company in this dataset bought the same AI capabilities. The difference in results came down to one thing: whether someone inside CX owned it.

One beauty retailer made 202 workflow updates in 30 days — refining as policies changed and new questions came in. Companies without a named owner saw performance stall or decline.

Read the data on what separates AI deployments that work from the ones that stall, and the four questions worth asking before your next AI investment.

See the data

🧐How to Actually Understand the AI Code You're Shipping With LoWisa

AI wrote it in seconds. But can you explain it? Debug it? Own it? Here's how LoWisa closes that gap.

Step 1: Download LoWisa - Free on Windows, macOS, and Linux. No API key needed to start, 20 free credits included out of the box.

Step 2: Open your project or paste a GitHub repo - Drop in your existing codebase or paste a repo URL. LoWisa creates a safe sandbox copy, your real project never gets touched.

Step 3: Start the tutor - Hit Start Tutor. It scans your project, builds a live map of your architecture, and starts talking you through it in real voice, entry points, data flow, and the decisions hiding in the code.

Step 4: Watch it open your files and highlight what matters - The tutor doesn't just explain in the abstract. It opens specific files, highlights the exact lines it's talking about, and draws the system on an interactive whiteboard as it goes.

Step 5: Run the terminal with it - Every claim gets tested with real commands in the sandbox. Break things safely, watch what production would do, and fix it until it's muscle memory.

Step 6: Prove you own it - The tutor tests whether you can actually explain the system yourself before it hands the keys over. Not a quiz, a real conversation.

AI builds the code. LoWisa makes sure you're the engineer behind it. You can download LoWisa for free here.

⚠️ 'We Must Act Now' on AI

More than 200 AI researchers, economists, and technology leaders signed "We Must Act Now," a Stanford-led statement calling on governments to prepare for the economic impact of increasingly capable AI systems.

The Shift:

200+ Experts Sign - The statement was signed by more than 200 researchers and economists, including 16 Nobel Prize winners, alongside leaders from Google DeepMind, Anthropic, and OpenAI.

Decade of Change - The group said AI could become radically more powerful within 10 years, potentially driving the largest and fastest economic transformation since the Industrial Revolution.

Focus on Jobs - The statement warns that rapid AI adoption could reshape labor markets and calls for governments to begin developing worker protections, safety nets, and updated economic policies.

Act Before Disruption - UVA economist Anton Korinek said previous technological revolutions gave societies decades to adapt, while AI may compress that adjustment period into just a few years.

The statement reflects growing consensus across AI researchers, economists, and industry leaders that planning for AI's economic impact should begin now, rather than after large-scale disruption has already started.

Together with Kalshi

Turn Your Opinions Into Profit

Join millions of traders putting their knowledge to work on real-world events—from inflation to elections. Buy “Yes” or “No” shares and earn if you’re right.

No house. Peer-to-peer. Cash out anytime.

Get a free $10 to start. Claim it and start trading now.

Start Predicting Now

_{Trade responsibly.}

🔨AI Tools for the Shift

📊 htmlslides – Creates AI-powered presentations and exports them as interactive HTML slide decks that are easy to share online.

🖼️ Image to Prompt Generator – Converts any image into a detailed AI prompt optimized for recreating similar visuals in image generation models.

📞 Rocalyn Edge – Answers calls around the clock with an AI concierge that represents your business in a professional, brand-consistent way.

🔍 AI Image Detector.info – Screens images using multiple AI detection models and provides evidence-backed conclusions instead of unreliable probability scores.

📢 AdPulse Studio – Imports your website to generate, publish, and measure AI-powered ad creatives from a single streamlined workflow.

🚀Quick Shifts

🔒 SpaceXAI disabled Grok Build’s codebase uploads after researchers found it was sending entire project repositories to the cloud, raising privacy concerns over source code, credentials, and other sensitive development data. 

⚡ AI's growing energy demand is driving up electricity bills, with the PJM grid operator adding $6.3 billion in costs across 13 U.S. states over the next two years as data center expansion continues to strain the power grid. 

🌍 Meta expanded AI voice dubbing on Instagram and Facebook, adding more language support so creators can automatically translate and lip-sync their videos for wider global audiences. 

🔊 OpenAI's first AI hardware device is reportedly a screenless smart speaker, designed as a mobile, voice-first ChatGPT companion that can learn about its user over time and even includes moving mechanical elements.

🧬 OpenAI researcher Miles Wang is reportedly launching an AI drug discovery startup, with the company said to be seeking $200 million at a $2 billion valuation to build AI models that help identify new uses for existing medicines. 

🧩 Prompt of the Day

Compare Job Offers Beyond Just Salary

Most people compare offers based on compensation alone. The best career decisions consider growth, culture, stability, learning opportunities, and long-term upside alongside pay.

Choose the offer you'll be happiest accepting years from now.

Paste the prompt: Drop this into ChatGPT, then fill in your offer details.

Prompt to paste

Help me compare my job options using [Insert offer details], [Insert what matters most to me], and [Insert alternative options]. Evaluate each opportunity across compensation, equity, benefits, career growth, learning potential, company stability, leadership quality, work-life balance, culture, mission, and long-term earning potential. Highlight the key trade-offs, identify any risks or hidden considerations, and recommend which option best aligns with my priorities. Explain the reasoning clearly so I can make a confident, well-informed career decision.

🤳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 👇