0:04
AI agent crawlers now need permission. Here’s how to get it
AI agent crawlers, the bots that fetch pages in real time on behalf of a person waiting for an answer, will be blocked by default on a slice of the web from September 15 onwards. Cloudflare announced the change on July 1, and most of the coverage since then has focused on Google. The more useful part is what it asks of everyone building agents, and what it offers them in return.
Cloudflare has replaced its single block-AI-bots switch with three categories. Search covers bots that index a page to answer questions about it later.
0:20
Building a Foundation Stack for General-Purpose Robots
This article is brought to you by X Square Robot.
Large language models gave artificial intelligence a working recipe. Pretrain a large model on broad data, and general capability follows. Robotics has no such recipe. Robotics systems have long been assembled from separate perception, planning, and control parts that rarely add up to intelligence a robot can carry from one task to another, or one machine to another.
0:33
Anthropic extends free Fable 5 access for subscribers as OpenAI's GPT-5.6 Sol heats up the pricing war
Anthropic is keeping Claude Fable 5 in its subscription plans through July 19, 2026. The model was supposed to switch to pay-per-use today. Subscribers can use up to 50 percent of their weekly limit for Fable 5.
0:41
Google’s SensorFM turns messy wearable sensor data into a general-purpose health intelligence layer
Google Research's SensorFM is a foundation model trained on more than a trillion minutes of wearable data from five million Fitbit and Pixel Watch users. It beats existing benchmarks on 34 of 35 health and behavioral tasks.
0:50
German AI consortium releases Soofi S, an open 30B model that tops benchmarks in both English and German
A German research consortium has released Soofi S 30B-A3B, an open language model trained entirely on Deutsche Telekom's cloud infrastructure in Munich. The model uses an efficient hybrid architecture that activates only a fraction of its 31.6 billion parameters per token, keeping throughput steady even at very long contexts.
1:02
The Three Dimensions of Custom Agentic Alignment: Purpose, Principles and Practices
I’ve been chewing on this new alignment framework and the most unexpected bit is how they split the problem into three moving parts that talk to each other in real time. First, they treat purpose not as a static goal but as a continuously updated intent vector that the agent samples every few seconds, letting the system shift focus without a hard re‑program.
Next, the principles layer isn’t a rule list; it’s a set of constraint functions that get weighted by the current purpose vector, so the same rule can be stricter or looser depending on what the business is trying to achieve at that moment.
Finally, the practices side closes the loop with a feedback engine that watches outcomes, nudges the purpose vector, and refines the constraints. The result is an AI that stays on‑track across scenarios while still being free to improvise where it makes sense.
