DIGITAL AGENDA · September 25, 2026

A four-agent hunter for AI slop, Shopify putting return shipping tax on automatic, and two speech models built for Turkish

Today’s 3 verified stories: Search Engine Journal reported a three-page paper in which seven Google-affiliated authors set out SAFE, a four-agent architecture for investigating mass-produced synthetic media and coordinated bot-net channel clusters, with no performance figure anywhere in the text. Shopify announced that from October 23, 2026 Shopify Tax and Tax Platform will calculate and apply tax on return shipping fees themselves, for merchants in the United States and orders shipped to US addresses. Voice AI startup Patientdesk.ai introduced two Turkish-focused speech models, Alania for text to speech and Duyu for speech to text. The detailed look at each story, its sources and what it means for your business lives on its own page.

NEWS 1 · AI

Google researchers put a four-agent hunter for AI slop on paper, without numbers

Search Engine Journal, in a piece by Roger Montti dated September 25, 2026, reported that Google researchers had published a three-page paper describing a system called SAFE. Written by seven Google-affiliated authors, it sets out a four-agent architecture for investigating mass-produced synthetic media and coordinated bot-net channel clusters: a root agent that synthesizes signals and renders the verdict, plus content, behavior and channel-cluster agents. The content agent runs on LoRA-adapted large language models with few-shot learning and targets violations of the spirit of the policy that slip past traditional classifiers, while the behavior agent reads infrastructure signals such as ASN, device fingerprints, synchronized upload timestamps and burst publishing. The paper also defines the synthetic gap as the time between a new generative attack vector emerging and a counter-measure going live. Although the abstract says early deployment results indicate SAFE significantly accelerates identification of novel synthetic threats, the text reports no performance figure anywhere: the Evaluation and Impact section is written in future tense and only names the metrics it will use. Search Engine Journal treats that absence of shared test results as highly unusual, and neither source says anything about Türkiye.

NEWS 2 · E-COMMERCE

Shopify puts tax on return shipping fees on automatic from October 23

Shopify announced in a changelog entry dated September 25, 2026 that tax on return shipping fees will be calculated automatically. From October 23, 2026, Shopify Tax and Tax Platform will work out and apply that tax themselves, for merchants in the United States and orders shipped to US addresses. Existing shipping tax settings still hold, and in states where shipping is untaxed the return shipping stays untaxed too. An estimated amount is shown when the return is created, the final amount is recorded at processing, and it flows into tax reports automatically. Shopify asked merchants running a manual workaround today to review it before October 23 so the tax is not counted twice, and said no action is required for everyone else. The entry contains nothing about Türkiye.

NEWS 3 · STARTUPS

Patientdesk.ai introduced two speech models built for Turkish: Alania and Duyu

Voice AI startup Patientdesk.ai introduced two Turkish-focused speech models on September 25, 2026: Alania for text to speech and Duyu for speech to text. Webrazzi reports that Alania supports tone variations such as appointment, question and calming delivery, ships a single voice output, and can stream audio without waiting for the sentence to finish, which points it at low-latency situations such as phone calls. Duyu transcribes both pre-recorded files and live speech in real time, covering harder cases including phone-quality audio and conversations where an email address or phone number is spoken aloud. On the Turkish FLEURS test Duyu was reported at a 4.71 percent word error rate against 5.04 percent for OpenAI Whisper large-v3, and on the company’s own phone-quality test set at 9.87 percent against 10.4 percent for ElevenLabs Scribe v2. Both models use the OpenAI API format and are free for one month from the sign-up date as part of the launch. The company also published Antalia-1, an experimental open source Turkish text to speech model, on Hugging Face with its weights, code and technical report; Webrazzi recalls a 1 million dollar pre-seed round led by Y Combinator in February.

The agenda moves fast, the system stays calm.

Let us talk about what these shifts mean for your business and work out together which step fits you.