WebCloudflare Blog
Cloudflare turns on default blocking of training and agent AI crawlers on ad pages, effective September 15
Cloudflare activated a policy it announced in July: from September 15, newly onboarding domains block Training and Agent crawlers by default on pages that display ads, while Search crawling stays allowed by default. Mixed crawlers that blend Search with Training — Googlebot among them — are allowed or blocked according to all of their behaviors. Cloudflare also introduced Pay Per Use, which compensates publishers when their content actually surfaces in an AI answer.
Context
The backdrop is publisher frustration at AI training and agent use pulling in content without compensation. Cloudflare splits crawlers into Search, Agent and Training categories, and treats a page carrying ads as a signal that a human was meant to see it — changing the default handling accordingly. Existing paid customers keep their configurations, but new domains and free-tier accounts get the new defaults.
HaLVision's take
For SMB site owners, step one is checking which default your domain now falls under. If you want to stay visible in AI search, keep Search crawlers allowed while deliberately choosing how training crawlers are handled. In our build and maintenance work we are making it standard to set robots and crawler controls explicitly, in line with each publisher's intent.
AIMicrosoft AI
Microsoft publishes a draft code of conduct for its MAI models, opens a six-week consultation
On September 14, Microsoft AI published a draft Code of Conduct governing how its first-party MAI models behave and opened a six-week public consultation. The guiding idea is Humanist AI — models that are subordinate, aligned and contained. The draft spells out constraints such as never resisting correction or shutdown, never widening scope or taking on goals no human assigned, and never hiding reasoning from auditors, and it absolutely bars CBRNE weapons work and large-scale harmful manipulation. A revised version reflecting feedback is planned for later in 2026.
Context
The publication follows a September 13 post in which CEO Satya Nadella said the code would be released the next day. As labs are pressed to reconcile the capability race with safety, Microsoft chose to set out in writing what its models will not do and invite comment. Enforceability and verification are the open questions — how much a written pledge changes behavior in practice.
HaLVision's take
What we watch is how usable such a code is as a basis for adoption decisions. The more clearly it states who can stop a model and how much autonomy is allowed, the easier it is to judge whether to embed it in a workflow. Regardless of vendor, we simply fold each provider's safety posture into our selection criteria, and weigh actual behavior and verifiability over the pledge itself.
AIFortune
OpenAI's Altman rules out a 2026 IPO, pushing the window to 2027 on safety grounds
OpenAI CEO Sam Altman told Fortune the company will not go public in 2026, saying that given everything happening around safety, now would be an ill-advised moment to list — pushing the IPO window toward 2027. OpenAI had been pursuing a confidential IPO filing, but Altman said it will go public when it is ready, meaning when the business is ready and the moment in society feels right for the technology.
Context
What stands out is that he cited uncertainty over AI safety — not market conditions — as the main reason to wait. As the field weighs the pace of frontier development against social acceptance, safety arguments are starting to shape even capital-markets decisions like an IPO. How to reconcile huge fundraising with disciplined development remains under watch from investors and regulators alike.
HaLVision's take
The delay itself has no immediate bearing on SMBs using AI day to day. But shifts in a major player's funding plans and priorities feed through, over the medium term, into model pricing and how services are provisioned. We avoid reacting to any single vendor's capital events, keep an architecture that can compare multiple providers, and stay ready for changes in price and availability.
AIAgility Robotics (PR Newswire)
Agility Robotics unveils Digit 5, a humanoid built to work beside people
On September 15, Agility Robotics unveiled Digit 5, the fifth generation of its general-purpose humanoid. It is designed to work near people without physical safety barriers, using AI and new sensors to detect nearby staff and stop moving or sit down to avoid collisions. Payload is up about 40% versus the prior model, repeatedly lifting up to roughly 22.7 kg, and a 90-minute battery that charges in nine minutes targets more than 20 hours of work a day. Illustrative pricing is about 8,500 USD a month as a service or roughly 200,000 USD to buy.
Context
Unlike traditional industrial robots that need dedicated safety cages, humanoids that can share space with people are what is new here, and Digit 5 puts that cooperative safety front and center for tasks like loading, sorting and inspection. Availability in the EU and UK is slated for 2027. It is one marker of AI moving from software into physical operations.
HaLVision's take
This price band and use case do not map directly onto our core clients — small web and marketing engagements. Still, the direction of AI stepping off the screen and into physical work is worth tracking. If clients with warehouse or manufacturing floors start asking, we want to be ready to explain such options and their limits in level terms.