Good morning. The job market is tough enough without AI recruiters offering you yet another life changing opportunity on LinkedIn. But one dev found a way to make the spam worth opening. He hid a prompt in his profile telling any AI that scanned it to call him “My Lord” and write in Old English. The bots read the instructions and sent back recruiter messages that sounded like someone was hiring for a startup in the year 900. Thy background hath impressed our hiring council.
The Pope and Anthropic are teaming up to decide where people belong in AI’s plan. Pope Leo is releasing one of the Vatican’s most important teaching documents next week and presenting it alongside Christopher Olah from Anthropic. The document focuses on protecting people as AI reshapes jobs and moves deeper into autonomous warfare. Anthropic makes sense here because they are the poster child for AI safety. Now they get to stand next to the world’s largest moral authority while the Pope tells 1.4 billion Catholics that AI is a matter of human dignity. That’s one hell of a brand partnership. (The Guardian)
The next killer app for smart glasses might be actual killing. Anduril and Meta are building smart glasses for the battlefield. The idea is to put military software directly in front of a soldier’s eyes, so they can identify a target and direct a drone to blow it up. Anduril wants the system to use eye tracking and small taps, so commands can happen without speaking. In one example, the glasses could send a drone to search an area, spot artillery, and tee up the next move before anyone says a word. Amazing how fast decision support starts sounding like “please tap approve.” (MIT)
Most intelligence tools scrape social after the fact and call it insight. Quid has direct data partnerships with top social platforms, so you get cleaner signals, faster.
AI-native from the ground up. Built to find what consumers actually think, not what the algorithm let through.
👀 closer look
Voice AI has a security problem hiding in plain sound. Researchers found a way to hide commands inside audio that sounds normal to people but gets heard by AI voice systems as instructions. The attack worked against leading open audio models with success rates up to 96%. In tests, hidden audio pushed models to run sensitive web searches, download files from attacker controlled sources, and send emails containing user data. Even asking the model to double check the user’s intent caught only 28% of attacks. Next time an AI takes your drive thru order, consider it a field test with fries. (IEEE Spectrum)
6.7 million people mistook a real Monet for AI slop. An anonymous artist posted a cropped image of Monet’s Water Lilies on X, slapped a “Made with AI” label on it, and asked people to explain why it was worse than a real Monet. People jumped in with museum grade confidence. They said the composition lacked coherence, the greens looked wrong, and one person wrote more than 700 words explaining why the “AI generated Monet” was a disaster. Then came the reveal. It was a real Monet hanging in a Munich museum. A few artists recognized it, but most people saw the AI tag and graded Monet like a failed Midjourney prompt. (Fortune)
Could AI coding make GitHub obsolete? Microsoft execs worry Cursor, Claude Code, and Codex could replace GitHub entirely. Copilot had the early lead, but newer AI coding tools let devs build and manage code without making GitHub the center of the workflow. If the AI coding app handles the work and keeps the files, GitHub loses its role. That’s a problem because Microsoft wants to train its coding models on GitHub customer data. Meanwhile, GitHub keeps buckling as AI tools flood it with uploads and edits, spiking traffic by 14x. It’s hard to believe GitHub might share Stack Overflow’s fate, but who wants to park their code with a company that treats it like training data? (The Information)
fun stats
🫰 $420. Bonus xAI promised employees for handing their tax returns to Grok. Two months later, Musk hasn’t paid up.
💸 $240 billion. Debt issued by tech hyperscalers since September 2025 to fund data center buildouts. 48 of these projects were blocked or stalled by local opposition last year.
🧶 94%. Enterprises struggling to manage unstructured data while deploying AI agents. Over half of agent projects are failing because the data is a mess.