Daily AI Newsletter: OpenAI Introduces Apps in ChatGPT

In today’s email:

☠️ Florida student asks ChatGPT how to kill his friend, ends up in jail: deputies
🇩🇪 Historian uses AI to help identify Nazi in notorious Holocaust murder image
🔥 Mira Murati’s Stealth AI Lab Launches Its First Product
🧰 11 new AI-powered tools and resources. Make sure to check the online version for the full list of tools.

OpenAI Introduced Apps in ChatGPT

Key Takeaway: OpenAI is baking third-party, interactive apps directly into ChatGPT, plus a preview Apps SDK so chats can call services like Spotify, Figma, Coursera, Zillow, and more without ever leaving the convo.

More Insights:

Launch lineup includes Booking·com, Expedia, Spotify, Figma, Coursera, Zillow and Canva. DoorDash/Instacart/Uber/AllTrails “coming soon.”
You can summon apps by name (“Figma, turn this sketch into a diagram”) and ChatGPT may auto-suggest relevant ones (goodbye, tab-juggling cardio).
Built on Model Context Protocol (MCP): apps can trigger actions, render full interactive UIs, and even play modifiable videos pinned in chat.
Accounts & $$$: sign in to existing services inside ChatGPT; monetization paths planned via features like Instant Checkout because even chatbots have rent.

Why it matters: This turns ChatGPT from “answer box” into a real platform/OS for everyday tasks, great for convenience, distribution, and new workflows, but it raises big questions about data scope, neutrality between competing apps, and whether your future internet starts (and stays) inside one chat window. (On the bright side, copy-paste just lost its gym membership.)

Turn Data into Ready-to-Share Content with AI

Still copying and pasting dashboards into slides? Matik’s AI connects to your BI tools, CRMs, and data warehouses to create presentations, documents, and reports directly from your data.

Auto-generate text, visuals, and tables
Always on-brand and up-to-date
Generate insights and align messaging

See Matik in Action

OpenAI, Jony Ive struggle with technical details on secretive new AI gadget

Key Takeaway: OpenAI and Jony Ive are aiming to launch a palm-sized, screenless AI assistant next year, but they’re wrestling with “always-listening” behavior, privacy guardrails, and the raw compute needed to make it useful at scale.

More Insights:

Concept: a phone-sized puck with mic/speaker (and possibly multiple cameras) that can sit on a desk or travel with you—designed to watch, listen, and build memory without a screen.
Biggest software headache: giving it a helpful “personality” that chimes in only when it should—and knows when to stop talking. (Yes, devices can overshare too.)
Compute crunch: serving ChatGPT-class models on a mass device is pricey and capacity-limited; that’s a major source of delay.
Supply chain: OpenAI has tapped Apple supplier Luxshare to build early hardware.
Stakes: OpenAI’s ~$500B valuation raises pressure to ship something that justifies the hype.

Why it matters: If they nail “ambient” AI that’s helpful-not-creepy and cheap enough to run, this could be the first real post-smartphone hit; if they miss, it risks joining the Humane AI Pin in the gadget afterlife—and proving that your “AI friend” still needs boundaries (and a budget).

Microsoft says AI can create “zero-day” threats in biology

Key Takeaway: Microsoft’s red team used generative AI to redesign known toxins so they dodged DNA-order screening software—purely in silico—prompting partial patches and a fresh biosecurity arms race.

More Insights:

Microsoft (using models incl. EvoDiff) showed AI can tweak toxin proteins to evade vendor sequence checks while keeping predicted function.
No wet-lab work: researchers kept the test digital and withheld sensitive code/protein details.
Vendors and U.S. officials were alerted; patches rolled out, but some AI-designed sequences still slip through.
Debate rages: some say DNA-synthesis screening is a practical chokepoint; others argue it’s leaky and want model-level controls and limits on outputs.
Bottom line from industry: this is ongoing “red team vs. blue team”—defenders must keep testing as AI design tools spread in drug discovery.

Why it matters: When code can rapidly remix the code of life, “Patch Tuesday” needs a lab coat—security can’t rely on one gate; we need layered defenses (model guardrails + synthesis screening + policy oversight) or the bad guys will keep finding creative speedruns.