SPEAKER_00 0:00

This is Sumo Growth AI Insights. Operator Briefing. Here's what actually moved and what you should do about it this week. Lead story. The European Union just put ChatGPT under its toughest platform rules. Monday, the European Commission dropped ChatGPT into the Digital Services Act's top tier as a very large online search engine. Reddit and Roblox were designated very large online platforms. The trigger is the same for all three. Each declared at least 45 million average monthly users in the EU. The Commission's supervision page has the provider-reported numbers. ChatGPT 159.1 million.2 million.6 million. ChatGPT is not sitting with the social platforms. It joins Google Search and Bing in the search engine category because it answers prompts in part by searching the web. That is the designation that matters for anyone running chat widgets, ads, or client funnels on top of it. The clock is four months. By January 2027, these three services have to deliver systemic risk assessments, independent audits, ad repositories, and researcher data access. Miss it, and Digital Services Act fines can hit 6% of global turnover. If you sell into Europe on Chat GPT, Reddit, or Roblox, that clock is now on your stack, not just OpenAIs. Plus, Bank of England Governor Andrew Bailey, writing as chair of the Financial Stability Board, just warned the G20 on Frontier AI cyber risk. In an August 31st letter to G20 finance ministers and central bankers, Bailey said Frontier AI shows increasingly sophisticated autonomy and problem-solving abilities, as well as threat capabilities. For finance, the most immediate concern is cyber risk. He warned that Frontier AI may materially alter the speed, scale, and economics of cyber risk, undermining confidence system-wide giving concentrated third-party providers. He also said many jurisdictions lack protocols for safe model release. The letter flags stretched AI-related valuations and other market fragilities ahead of G20 meetings in Asheville, read that as the finance version of the Brussels move. A concentrated third-party AI stack is a cyber risk item, not a vendor footnote. Three more items you need on the board. First, OpenAI cuts Cursor after SpaceX's $60 billion deal. Anthropic piles on clawed compute. OpenAI said it will wind down supplying models to Cursor after SpaceX's roughly $60 billion AnySphere acquisition. Proposed cutoff November 12th, 2026. No future models, including Astra. The contract is exiting through a change of control clause. Cursor's Michael Trule said OpenAI is about 5% of traffic. Anthropic's Tom Brown pledged to continue to increase compute for Claude and Cursor. Do not treat model access as permanent. OpenAI just showed a change of control clause can yank a frontier model with about 11 weeks' notice. Second, Broadcom launched VMware Private AI Cloud for on-prem agents. At VMware Explorer 2026, Broadcom bundled VCF9, AI Factory, and more than 150 validated models. Tanzu deny by default agent sandboxes go generally available in the fall. Agent Minder is generally available now. It is an identity and intent control plane between agents and tools, APIs and MCP. Broadcom claims internal agent minder traffic of about 36 million customer and 7 million workforce API calls daily. If you are in regulated shops that cannot put agents on a public chat endpoint, this is the on-prem package they are selling. Third, Clawed Code's quote-unquote permanent plus 25% weekly limits still cut today's headroom by about 17%. Anthropic will permanently raise Claude Code's standard weekly limits by 25% on September 14th for Pro, Max, Team, and Seat-based Enterprise. At the same time, it is ending the temporary 50% boost. Net versus what users have today, about a 17% cut. Raise on paper, cut in practice. If your delivery teams are running clod code at the current cap, plan for less room, not more, after September 14th. Trending AI tools today. 3. No pitches. What they do. Video agent by Fotor at Photor.com creates and edits precision motion graphics and video with chat. Interactive sessions at revolt.ai lets you drive the full software lifecycle with AI agents step by step. Brandjet at brandjet.ai turns public buying signals into sales pipeline. Business AI tip of the day. Put a January 2027 clock on every ChatGPT funnel. Brussels just drop ChatGPT into the same Digital Services Act bucket as Google search. If you run EU client ads, chat widgets, or Reddit campaigns through AI, you now have four months by January 2027 for risk assessments, ad transparency, and researcher access. Bailey's financial stability board letter is the finance version of the same warning. A concentrated third-party AI stack is a cyber risk item, not a vendor footnote. This week, before you sell the next AI-powered workflow, do four things. Inventory every client funnel that hits Chat GPT, a single frontier API, Reddit, or Roblox. Flag EU-facing chat and ads that need logs, source tracking, and a human override. For finance and insurance clients, write a one-page what if this API goes dark playbook. Dual provider, exportable prompts, kill switch. Do not treat model access as permanent. OpenAI just showed a change of control clause can yank a frontier model with about 11 weeks' notice. A clever agent on an untracked ChatGPT wrapper is not a growth system. It is an unlogged vendor with 159 million EU users and a ticking compliance clock. Pro tip: Do not wait for a client to ask. Send a five-bullet regulated AI stack memo this week. Lead with ChatGPT's very large online search engine designation and Bailey's Cyber Risk Line as the why. Then list the three tools or APIs you would actually fail over to if one got shut off. That is the briefing. Four months to January 2027. Inventory the funnels, write the failover, send the memo, operator out.