Editor's note
Today paired two questions: how to govern AI, and how far to let people use it. OpenAI, Anthropic and Google DeepMind were confirmed to have spent weeks in talks on safety and a possible industry standards body, while the European Commission lines up a Kids Act to restrict minors' use of social media and AI chatbots. On the product side Google pushed real-time voice into Search and Workspace with Gemini 3.8 Live, and quietly kept reshaping how search reports and sends traffic to sites. For SMBs, the practical move is to watch how measurement and click flows shift in AI-era search — and read it against your own site and lead generation.
AITechCrunch
OpenAI, Anthropic and Google confirm weeks of AI safety talks, eye an industry standards body
On September 15, OpenAI policy chief Chris Lehane confirmed the company has spent several weeks in talks with Anthropic and Google DeepMind on AI safety. The discussions center on a standards body for frontier models, reportedly including the option to coordinate a slowdown in development if risks escalate. Lehane acknowledged such coordination could raise antitrust questions but said the firms do not believe they need a government waiver to proceed.
Context
The move follows public backing from company leaders for safety collaboration, in the wake of an essay by Anthropic's Dario Amodei calling for industry cooperation. As the capability race continues, there is growing unease about leaving the pace of frontier development to individual labs, and the focus is now on how to build in standardization and third-party evaluation. Enforceability and antitrust alignment are the open questions — whether the pledges carry real weight.
HaLVision's take
Industry self-regulation is medium-term groundwork for a studio like ours. The more clearly standards spell out who can stop a model and how much autonomy is allowed, the easier it is to advise clients on whether to adopt one. We stay unattached to any single vendor and keep comparing each provider's safety posture against its actual behavior.
AIUnite.AI
Google launches Gemini 3.8 Live, a real-time voice model, rolling into Search and Workspace
On September 15, Google launched Gemini 3.8 Live, a voice-to-voice live dialogue model, alongside Gemini 3.8 Live Extended Thinking for more complex tasks. The models sustain a spoken conversation, understand what a camera sees, call tools and APIs while still talking, and switch among 97 languages mid-sentence. They roll out across the Gemini API, AI Studio, Gemini Enterprise, Search Live, Gemini Live and Workspace, with Search Live gaining step-by-step real-time help.
Context
The update pushes spoken interaction toward more speed and naturalness and folds it into search and work apps. Beyond text-first generation, an experience where the model talks, sees and acts at once makes uses like inquiry handling and on-site guidance more plausible. At the same time, real-time voice is harder to keep on the rails on cost and errors, so deciding how much to delegate becomes the adopter's task.
HaLVision's take
What we watch is how this lands in the inquiry flow on a client's site. Voice-based first response and FAQ guidance could ease the load for small shops and service businesses, but you need to fix the pricing and the tolerance for mistakes up front. The steady path is to trial it on a narrow use case and pair it with a clean hand-off to a human.
AIForbes
OpenAI reportedly weighing a new funding round at up to a 1.5 trillion USD valuation
OpenAI is reportedly weighing a new funding round at a valuation of up to 1.5 trillion USD, more than double the roughly 730 billion USD it commanded in its last round. The talks run alongside its stated plan to push an IPO into 2027 on safety grounds. Reports also note that the coordination around an AI standards body could draw antitrust scrutiny.
Context
The shape of it is raising a very large sum privately while staying in no hurry to list. As the cost of frontier development and operation swells, a higher valuation reflects strong investor expectations but also sharpens the question of how to close the gap with actual revenue and discipline. The figures are still at the exploratory stage, with terms and execution unconfirmed.
HaLVision's take
The valuation itself has no direct bearing on SMBs using AI day to day. But the scale and priorities of a major player's fundraising feed through, over the medium term, into model pricing and how services are provisioned. We avoid being swayed by any one firm's capital events, keep an architecture that can compare multiple providers, and stay ready for price hikes or spec changes.
AICNN
EU set to present a Kids Act restricting minors' use of social media and AI chatbots
The European Commission is set to present a Kids Act on September 17 that would restrict minors' use of social media, video-sharing, AI chatbots and online games. The plan is tiered by age: under-13s could not use social media, some gaming platforms or AI chatbots without parental supervision, 13-to-15s would need parental consent, and those 15 and over could set up their own accounts. Covered companies would have to avoid designs that foster addiction or unhealthy emotional attachment and verify users' ages through an official age-verification app.
Context
The proposal, advanced by Commission President von der Leyen and colleagues under a child-protection banner, is notable for extending the regulatory focus from social media to AI companions and chatbots. How to implement age verification while preserving privacy is the practical sticking point. The presentation is a starting point — the text can change as member states and the Parliament weigh in.
HaLVision's take
If you run services or media whose audience includes minors, age verification and disclosure rules are a design question to fold in early. For clients with EU-facing acquisition or apps, we would urge unhurried preparation on the assumption that requirements will firm up. We track the detail without overreacting, and implement only what is needed, in stages.
WebSearch Engine Roundtable
Google signals Search Console position and impression metrics may change as AI Mode and AI Overviews evolve
Google's John Mueller said that position and impression tracking in Search Console may evolve over time as AI Mode, AI Overviews and Search itself evolve. No timeline or concrete change was given, but he acknowledged that as AI-driven surfaces spread, the meaning of the traditional metrics can shift.
Context
In AI Overviews and AI Mode, sources are summarized and cited inside a generated answer, which looks nothing like the classic ten blue links. That erodes the assumptions behind measuring where and how often a page appeared, leaving room for Search Console's reports to be reworked to match reality. For site owners, the point is to track citation and traffic patterns on AI surfaces rather than fixate on an absolute rank.
HaLVision's take
This bears directly on our SEO and Search Console work. We need to move reporting away from leaning on rank as a single number, toward a read that includes AI-surface exposure, citations and actual inquiries. We advise clients to assume the metrics will move, and to judge by trend and lead quality rather than a raw rise or fall in a figure.
WebSearch Engine Roundtable
Google tests linking product results in Search directly to merchant sites
Google is testing linking product and shopping results within Search directly to the merchant or retailer website, rather than opening an in-Google detail panel. Product titles appear as ordinary blue links, and a click sends the user to the product page outside Google. Google ran a similar test within Shopping in June and has now extended it to core web search results.
Context
Traditionally, clicking a product opened a detail panel inside Google, keeping users within Google's own surface. This test points toward favoring direct hand-off to the merchant site over that on-page dwell. It is only an experiment, not a full rollout, but if shipped it could change the traffic path to retail sites and the assumptions behind measuring it.
HaLVision's take
For e-commerce and shop clients, this is a change that could make well-built product pages matter again. Structured data, clean product information, and the speed and clarity of the landing page will decide whether you benefit from the direct hand-off. We will watch where the test goes, but first advise prioritizing the basic quality of product pages.