This week’s links show AI moving further into the environments where delivery professionals work. Agents can now operate connected devices, draft campaigns and act on behalf of users, while specialist models are being built around legal, CRM and structured decision workflows. At the same time, policy proposals, financing conditions and vendor disclosures are making the control layer harder to ignore. The useful question is no longer whether a tool can produce an impressive result. It is whether its permissions, evidence and review path are fit for the work.
Agent access and enterprise controls
Apple launches Siri AI in beta: Siri AI adds personal context and app actions, but Apple says it will initially be unavailable in the EU on iOS, iPadOS and watchOS.
Meta introduces Meta One subscriptions: Meta is packaging higher AI usage and business tools into paid plans, a useful signal for software-cost scrutiny.
OpenAI tests Sponsored Agents: The trial points to a new commercial interface where a clearly labelled business agent continues the conversation after an advert.
Specialist models and structured work
OpenAI introduces Astra for Law: A 230-million-URL legal index and Word integration show how specialist context and document workflows are becoming part of model positioning.
Salesforce introduces Koa: A CRM reasoning model trained on synthetic workflow scenarios keeps customer data inside Salesforce’s trust boundary, according to the vendor.
TypeSafe introduces Jev: Jev is designed for structured decisions rather than free text, offering a different route to automation where outputs must be validated by software.
Google introduces Gemini 3.8 Live: Google is adding near-real-time voice, visual grounding and background task execution across API, Search, Workspace and the Gemini app.
How a 1941 Enigma message was recovered: This reproducible case study is a useful reminder that AI-assisted discovery still needs source preservation, independent checks and explicit limits.
Governance, disclosure and public accountability
The EU proposes restrictions for children under 13: The proposal would include AI chatbots, with parent-chaperoned accounts suggested for some older children.
The DeepMind Institute launches: Google DeepMind is creating a forum for technical, social and governance questions around advanced AI.
Trump proposes an AI Force and AI czar: The announcement remains a proposal, but it shows how quickly AI governance is becoming an executive and political portfolio.
Sam Altman to brief the UN Security Council: The 23 September briefing places frontier AI and regulation on a diplomatic agenda beyond the technology sector.
Economics, infrastructure and the cost of rework
The Federal Reserve raises rates to 3.75% to 4%: Dearer money matters for data-centre buildouts and every scheme whose financing assumed easier conditions.
Unredacted filings expose AI scraping claims: The New York Times case raises licensing, paywall and substitution questions for firms whose data suppliers may be training sources.
Superhuman Academy: A free, email-gated learning path offers a structured way to turn general AI interest into daily workplace practice.
All content reflects our personal views and is not intended as professional advice or to represent any organisation.
