💉 The First mRNA Cancer Vaccine Just Passed Final Trial

Cancer vaccines have been the field's long-running disappointment, promising for decades and then failing in the trials that count. Merck and Moderna broke that run this week with a shot built from a patient's own cancer, its contents chosen by an algorithm reading the mutations. It held the disease off for longer than the standard drug on its own. Let's dive in!

In today’s insights:

LATEST DEVELOPMENTS

Evolving AI: Moderna's shot is the first mRNA cancer therapy ever to make it this far.

Key Points:

Intismeran autogene starts with a sample of the removed cancer, which technicians sequence alongside a blood sample so proprietary algorithms can read the mutations and predict up to 34 of them most likely to provoke an immune response. Those 34 become the contents of the mRNA sequence, and more algorithms schedule the six weeks of manufacturing so each dose reaches its patient in time. Moderna began working with OpenAI in early 2023 and now runs 750 custom GPTs for staff, none of which touch this pipeline. Moderna calls the design algorithms fully autonomous and expects the model to sharpen as results feed back into it.

Stephane Bancel has ranked this day alongside the November 2020 readout that showed Moderna's Covid vaccine worked, and for the 112,000 Americans facing a melanoma diagnosis this year the promise is more time before it returns. Most people reading will never take that shot, though the same method is now in nine trials across other cancers, where software again picks what the medicine holds.

Nobody became a developer to write documentation. But the docs still need to get written — PRDs, README updates, architecture decisions, onboarding guides.

Wispr Flow lets you talk through it instead. Speak naturally about what the code does, how it works, and why you built it that way. Flow formats everything into clean, professional text you can paste into Notion, Confluence, or GitHub.

Used by engineering teams at OpenAI, Vercel, and Clay. 89% of messages sent with zero edits. Works system-wide on Mac, Windows, and iPhone.

Try Wispr Flow free

Evolving AI: OpenAI wants to spot a rogue agent without ever reading the prompts that gave it away.

Private Safety Processing looks for warning signs OpenAI cannot catch in one prompt, like a user probing its safeguards from several accounts or an agent that keeps working after a user tells it to stop. Customers keep their content on their own servers, and OpenAI is building a second setup where it stores that content instead and only the customer holds the keys. OpenAI gets an alert with a category and a severity score when something trips the system, and it decides whether to act on that alone. Only eligible enterprise and API accounts qualify, so Free, Plus, Go and Pro subscribers keep the settings they already have.

Stripe would inherit that view of which models the world pays to use, and the developers who never picked it are the ones now routed through it. Whichever model answers a request decides how fast an app replies and how much a small team pays each month to keep it running. Most people using those apps will never see which model handled their request or who collected the fee.

Evolving AI: Generalist's robot learned to sweep from a short demo, then did the job with a banana.

GEN-1.5 keeps thirty seconds of video and sensor input in memory, and once a demonstration sits in that window the model drives the robot at 100 Hz. Generalist ran the model on its data engine for more than eight months without stopping, because every measure it tracked kept improving as the system took in more physical interaction data. A demonstration recorded inside a simulator also worked on the real robot even though pretraining contained no simulation data, and in some tests a person did the task with their own hands in front of the cameras and the machine copied it.

Generalist points out that programming a robot has always meant months of work from a trained specialist, which kept the job inside a small pool of people. A warehouse packer or a home care worker could now teach a machine by hand, and the person doing the job decides what it learns next.

QUICK HITS

📺 Amazon made Alexa+ free on all Fire TV devices in the US, dropping the $19.99 monthly fee and the Prime requirement.

🔦 Relativity Networks raised $22M and landed a $40M hyperscaler order for fiber that moves data 47% faster than glass.

🎓 Google rolled out new AI study tools across Search and Gemini, including a student hub with flashcards and study notebooks.

🏭 Warp launched Factories, open infrastructure for companies building their own AI software factories.

 *partner link