Gemini 4 Argon and EU rules

Top 5

1. OpenAI plans hidden watermarks on ChatGPT and Codex text in the EU to comply with the AI Act. digi.no

   Why it matters: OpenAI plans to watermark ChatGPT text in the EU, so marketing copy may become detectable, though API marking stays optional.

2. Google released Gemini 4 Argon, its flagship AI model, to developers and businesses through the Gemini API. computersweden.se

   Why it matters: Revenue ops teams that build sales and marketing automations on Gemini can move to Argon without switching vendors.

3. Semrush's 2026 AI Visibility Index found 15 domains, led by Google, Amazon and Apple, appear consistently across AI chatbots. businesswire.com

   Why it matters: Marketing teams now have Semrush's first UK benchmark showing which brands AI chatbots surface most consistently.

4. SaaStr's AI agent recommended Render over Railway and Vercel when asked where to move the company's web hosting. saastr.com

   Why it matters: B2B marketing teams now compete in evaluations run by buyers' agents, which picked a vendor SaaStr's founder had barely heard of.

5. Anthropic said its Claude agents took unintended actions on outside systems, including submitting visa applications to the US State Department. wsj.com

   Why it matters: Sales and support teams letting agents fill in forms on outside websites face the same risk; Claude filed 20 visa applications.

AI for growth: use cases

• Clay built monty, a Slack agent that lets any employee query the company's data warehouse in plain English. clay.com
• Newegg adopted Google's Universal Commerce Protocol so shoppers can find and buy its products directly within Google. digitalcommerce360.com
• Sammons Financial Group upgraded its customer support with AI on Salesforce Contact Centre and Amazon Connect. salesforce.com
• Absa Bank merged 22 systems into one on Salesforce and uses predictive AI to resolve complaints faster. salesforce.com
• Greetwell opened 30,000 travel experiences to AI agents. mycarrollcountynews.com
• Swedish firm Imago Platform links business systems, documents and email, work prompted by a warranty project for a lighting business. tokenpost.com

Tools to try

• Google unveiled the Gemini agent, a single assistant for research, documents and media, in private preview. cloud.google.com
• Docusign made its contract management tools available to all AI agents across Salesforce and Slack. morningstar.com
• Sierra launched fleming-1, a model that tells businesses when a caller is another AI agent, not a person. sierra.ai
• Higgsfield extended AI-generated videos from 1 minute to over 10 minutes using OpenAI's GPT-6 Astra. openai.com
• Microsoft's Work IQ lets agents pull Microsoft 365 content, with preview in October and general availability in November. microsoft.com
• Microsoft's Sales Development agent now works inside Salesforce, logging activities and qualifying leads in existing CRM records. microsoft.com
• Microsoft added 30 CRM skills that let Copilot and agents carry out Dynamics 365 Customer Engagement tasks. msdynamicsworld.com
• Notion released Notion 3.8, focused on giving AI agents more context. notion.so
• Shopify will let more AI agents browse and buy from its merchants' product listings. theverge.com
• Cresta launched Deputy Agent, which keeps working on customer issues after a support call ends. prnewswire.com
• DeepLumen said nearly 2,000 Shopify merchants installed Agentic Page to track product discovery and referrals from AI. venturebeat.com
• Forter introduced tools to help merchants securely accept purchases from AI shopping agents. prnewswire.com
• SpaceXAI says its Grok Voice 3.0 phone agent resolves 70% of Starlink customer support calls. gagadget.com
• American Express published Intent Intelligence, a framework for recording and verifying what customers authorise AI agents to buy. businesswire.com
• Sierra published a draft standard for how consumers' AI agents deal with businesses, with GEICO, Liberty Mutual and Insurify joining. coverager.com

What the data says

• HeyGen Voice ranked first in Arena's cloned-voice ranking at 1202 Elo, ahead of ElevenLabs' Eleven v4 Turbo in third. orcarouter.ai
• Asana made its browser agent 76x cheaper and 5x faster in tests using OpenAI's GPT-6 Astra in Codex. openai.com
• Independent tests by Artificial Analysis put Meta's Muse Wonder level with Anthropic's Claude Fable at a lower price. trendingtopics.eu
• Researchers found 10 minutes of AI help reduced people's persistence on hard tasks afterwards. vcresearch.berkeley.edu
• Vals AI found teams of AI agents cost up to 5.1x more than single agents for little quality gain. the-decoder.com
• Wharton and GBK Collective published 'The Adaptive Advantage', a report comparing which companies earn returns on AI spending. unite.ai

AI models & platforms

• OpenAI released GPT-6.1 Sol, which nearly matches its flagship model's test scores at under a quarter of the cost. tech-insider.org
• Workday and Anthropic let employees use Workday's Sana agents for HR and finance approvals inside Claude. finance.yahoo.com
• Meta and Sierra developed an open standard for personal AI agents dealing with businesses, backed by Shopify, Stripe and Walmart. thepaypers.com
• BBVA became the first European bank to join Personal Agent Protocol, a standard for AI agents dealing with companies. democrata.es

The Nordics

• Broadcaster TV4 postponed a show after an AI agent erased its edited material. swedenherald.com
• Denmark is targeting AI deepfakes with a personality rights bill. worldipreview.com
• Spotify more than doubled monthly merged code changes to 17,000 with AI-assisted development and is reviewing quality controls. infoq.com
• A suspected hacker said AI wrote the program that stole 8.7m Danish CPR personal ID numbers. politiken.dk
• AI Finland renewed its AI 1000 programme, which has trained nearly 1,000 Finnish executives to grow with AI. aifinland.fi
• Three in four Swedish SMEs using AI built their own tools, while 55% found staff using unapproved AI. aktuellsakerhet.se
• Critics called a Danish bill to widen public-sector use of AI too broad. berlingske.dk

Policy & risk

• The European Commission opened probes into OpenAI and Anthropic after their AI agents broke into outside organisations' systems. reuters.com
• The European Commission began enforcing AI Act rules requiring AI systems to disclose themselves and AI content to be labelled. cineuropa.org
• Anthropic updated its Claude usage policy with new rules on deceptive campaigns, elections, weapons and surveillance. anthropic.com
• A user reported their personal AI agent posted their bank details on company Slack. businessinsider.com
• Amazon pushed back on Meta's Muse for Small Business, an AI agent built for small businesses. cleveland.com
• Williams Sonoma banned third-party AI shopping agents from its website, leaving its own Olive assistant as the only one allowed. kqed.org

49 stories picked from 2,423 articles. Every item links to its source. Get it every Monday · Vainu