Editor's note
Today paired two currents: AI being built deeper for specific lines of work, and the search surfaces themselves being rewritten by AI. OpenAI shipped Astra for Law for legal work, and Alibaba's Qwen released Qwen3.8-Omni-Flash — handling audio and video in one model — with steep price cuts, so AI is deepening vertically, not just spreading horizontally. Meanwhile Google is testing turning a shop's info panel into an AI Overview-style block and is opening up Search Profiles, quietly shifting the ground under how a business appears in search. Chrome also pushed a stable update closing 42 vulnerabilities — unglamorous, but the discipline of simply keeping the browser updated is part of today too.
AIOpenAI
OpenAI launches Astra for Law, a legal-tuned configuration of GPT-6
On September 17, OpenAI introduced Astra for Law, a foundation for law firms and legal-tech companies. It pairs its top model, GPT-6 Astra, with tailored instructions, tools and a legal search index spanning more than 230 million URLs of US case law, statutes, regulations, court rules and administrative decisions, updated daily. It launches with 26 partner integrations including Thomson Reuters, Harvey, Legora and iManage, and early adopters such as Latham & Watkins and Sullivan & Cromwell.
Context
The point is the shift from a general-purpose chatbot to a domain-tuned system that folds in a field's context, data and tools. Legal work demands accurate citations and current sources, so pairing the model with a dedicated search index — rather than the model alone — aims to make grounding and citation more reliable. It is also offered as an API, so firms like Harvey and Legora can build it into their own products.
HaLVision's take
The target is large firms, but the direction it signals applies to smaller shops too. The design principle is that generative AI becomes fit for real work only when tied to your own accurate data and a retrieval layer, not used on its own. When we recommend AI to clients, we likewise start by shaping their internal information into a citable form and building a path back to sources. Traceable grounding, more than flashy features, is the foundation of trust.
AITechNode
Alibaba's Qwen releases Qwen3.8-Omni-Flash, an audio-video omni model, with steep price cuts
On September 18, Alibaba's Qwen team released Qwen3.8-Omni-Flash, a native omni model that handles text, images, audio and video in a single model. It carries a 1M-token context window and targets agentic uses such as long-video analysis and meeting summaries. Alibaba claims a gain of more than 26% on average across 30 evaluations over the prior Qwen3.5-Omni-Plus, and cited steep cuts — 98% cheaper per hour of audio input and 93% cheaper for combined audio and video — with API input from as low as RMB 0.8 per million tokens.
Context
Rather than transcribing audio or video through separate converters before processing, a single model takes all four input types directly. Alongside multimodal quality gains, the sharp price cut makes AI processing of audio and video — often shelved on cost grounds — far easier to justify. It is API-only, and shipped with helper tooling for long-running and real-time workflows.
HaLVision's take
For SMBs, the price cut matters more than the benchmarks. Uses that never penciled out — summarizing inquiry calls, pulling content from video manuals — become realistic. We do not lock to one vendor; we compare cost and quality per use case, and cuts like this widen the field of options. The sound approach is to pilot on a small task, measure the effect and the bill, and only then expand.
WebChrome Releases (Google)
Chrome ships a stable update fixing 42 security vulnerabilities
On September 17, Google updated the desktop Chrome stable channel to 153.0.8010.52/.53 for Windows and Mac, with Linux to follow. The update carries 42 security fixes, including high-severity flaws in areas such as WebGL and Workers. Updates for Android and ChromeOS shipped the same day.
Context
The browser is the front door to daily work, and many of the fixes are memory bugs that could be exploited from the outside. It is a routine stable update closing many vulnerabilities at once, but it includes high-severity items like use-after-free, so leaving it unpatched is a real risk. Updates arrive automatically, yet applying them requires restarting the browser — a step easily overlooked.
HaLVision's take
It is not glamorous news, but for a small business the highest-return defense is staying on top of updates like this. We advise clients to keep work machines on auto-update and restart the browser regularly so patches actually apply, and to audit their extensions at the same time. As builders, we avoid coding to old-browser assumptions and keep implementations that do not stand in the way of updating.
WebSearch Engine Roundtable
Google tests replacing a shop's local info panel with an AI Overview-style block
Google has been spotted testing replacing a business's local knowledge panel — the box showing hours, phone and reviews — with a block labeled AI Overview. It leads with an AI Overview label, and a Show more button expands into an AI Mode-style chat. Ben Fisher reported it and Barry Schwartz reproduced it on his own business; Google has published nothing, and similar tests have run on and off since early 2025.
Context
The traditional local knowledge panel showed, more or less verbatim, the information a business itself supplied and maintained. This test points toward replacing that surface with an AI Overview that summarizes the business in Google's own words, potentially shifting control of the framing to Google's generated description. It is a limited test and specs may change, but it signals the ground moving under how you appear in local search.
HaLVision's take
This is an important signal, tied directly to shop and service acquisition — the local SEO (MEO) work we do. If AI summarizes a business to show it, the accurate primary information behind that summary matters all the more. We advise clients to keep their business profile — hours, categories, description, photos and review replies — current and in a form an AI can read correctly. Even at the test stage, getting the facts consistent is how you prepare.
WebSearch Engine Land
Google lowers the Search Profiles bar to 10,000 followers
On September 16, Google cut the requirement for Search Profiles — creator-facing pages in Search — to 10,000 followers on any one of YouTube, Instagram, X or TikTok. That threshold stepped down from 35,000 in August and 100,000 at launch. Alongside it came a redesigned article layout, single-login management of up to 10 profiles, and follow badges you can place on your own site.
Context
Search Profiles is a surface for creators and publishers to present their work in Search and connect with followers. Sharply lowering the bar widens eligibility to smaller businesses and individual creators. Google is leaning further into making the relationship with a publisher visible inside Search, and how search treats the creator — not just an individual article — is becoming a point of discussion.
HaLVision's take
For local businesses and shop owners with some social following, this becomes a realistic option to hold an official presence inside Search. We advise first aligning your existing social accounts and site information, and unifying the profile copy and links. That said, it is one Google feature — do not over-rely on it, and keep your own site as the anchor. The sound move is to check eligibility, weigh the payoff, and only then fold it into operations.