Top stories
July 11, 2026 · 5 stories (site) · 5 stories (base)
July 9 was a turning point for enterprise AI: OpenAI locks its alliance with Microsoft (GPT-5.6 in Word, Excel, PowerPoint), launches ChatGPT Work to challenge Claude Cowork, and loses its number two. Meanwhile, Alberta has Claude Code audit 466 million lines of government code, and Cursor teaches its agents to self-correct on the fly.
🔥 Top stories
01
OpenAI launches GPT-5.6 and installs it as the default engine for Word, Excel and PowerPoint
When you open Word tomorrow to draft a report, or Excel to crunch numbers, the help that suggests rewrites or formulas will come from a new AI model: GPT-5.6, unveiled by OpenAI on Thursday, July 9 in three versions (Sol, Terra, Luna). Microsoft confirmed the same day that GPT-5.6 becomes the default engine for the entire Microsoft 365 Copilot suite (Word, Excel, PowerPoint, Cowork). It puts to rest the divorce rumours Bloomberg revived last week. Good news on price: the Luna tier is priced at 1 dollar per million words read, cheaper than the previous version, and OpenAI promises 54 percent better efficiency on code. For businesses, it confirms that your favourite software vendor pays OpenAI to deliver your AI; for the curious, the price war keeps grinding down.
02
ChatGPT Work: OpenAI takes the fight to Claude Cowork on the white-collar desk
Picture a new invisible colleague, plugged into your daily tools (Slack, Salesforce, Jira, Databricks, Oracle), who turns your scattered notes and rough drafts into finished documents. That's ChatGPT Work, launched by OpenAI on July 9. It's already in use at Virgin Atlantic (analysis cycles cut from weeks to hours), Zapier (leads sorted automatically) and NVIDIA (40 percent of post-event crunching replaced). It is a direct frontal challenge to Anthropic's Claude Cowork and Microsoft's Copilot Cowork, in the nascent category of AI coworker for office workers. For a small French business, it's confirmation that OpenAI is moving away from consumer products and betting hard on the enterprise segment - where it was lagging behind Anthropic.
03
A government scans 466 million lines of its code with Claude Code in 20 hours
Alberta's Ministry of Technology, in Canada, just did something new: it handed Claude Code (Anthropic's AI software engineer) the task of combing through 466 million lines of government services code. Result in 20 hours: vulnerabilities identified, then internal tools built automatically to patch them. For an individual user, nothing changes tomorrow. But for a municipality, a ministry or a large administration, it's proof that a frontier coding agent can now work at the scale of a real software estate - not just a 200-line script. The same day in the United States, the federal cyber agency CISA used Mythos (Anthropic's red-team model) to audit agency code, and the UK FCA floated the idea of directly regulating AI models. First wave of governmental usage.
04
Cursor 3.11: your coding agent learns to self-correct while it works
Cursor, the editor that popularised agents that code on your behalf, released version 3.11 on July 10. Three concrete new features for everyday users. First the 'Side Chats': you can ask a question in parallel with your main agent without interrupting it, like asking a colleague a quick aside while they keep typing. Then 'Conversation Search': press Cmd+K to find a specific detail across the thousands of conversations you've had with the agent - a search bar for your memory. And most importantly, new programmable 'hooks': you can now write rules telling the agent 'review yourself after each reply, if quality is insufficient, try again'. It's the missing piece that turns a good assistant into one that improves itself on the fly.
05
Anthropic recruits the former Fed chair to strengthen long-term trust
Ben Bernanke, the former chair of the US Federal Reserve (2006-2014, the man who steered the 2008 financial crisis), officially joins Anthropic on July 9. He takes a seat on the 'Long-Term Benefit Trust', the independent body that appoints board directors and advises on economic decisions. For a regular user, it sounds abstract. But for an enterprise customer or a regulator, it's a strong signal of institutional seriousness, three months ahead of a likely IPO. Anthropic sends a clear message the same day: 'we welcome your hard questions' - with a public collection where anyone can submit their criticisms of AI. On a day when OpenAI loses its number two for medical reasons, it's a strategic communications coup that repositions Anthropic on the 'trusted enterprise' side.
📡 To watch
OpenAI's number two steps back to part-time adviser for health reasons
Fidji Simo, the executive overseeing OpenAI's consumer and pro apps (ChatGPT, API, Sora), has been on medical leave for three months due to a relapse of a neuro-immune condition (POTS). She stays on as a part-time adviser. It's OpenAI's fourth executive loss in six months, in a year of a potential IPO at 852 billion dollars. For enterprise customers, it's an internal turbulence signal worth watching over the next six months.
Tencent Hy3: 50 internal products adopt the cheapest Chinese model in 48 hours
On July 6, Tencent released Hy3, an open-weight model (permissive licence) priced at 0.14 dollar per million words read - 45 percent cheaper than Anthropic's equivalent. Within 48 hours, the official account confirmed 50 Tencent products (messaging, browser, office suite, code) already run on it. For a team looking for an open, low-cost alternative, it's validation at scale by the largest Chinese model deployer in the world.
IPO pipeline: Anthropic expected Q3, OpenAI within 90 days?
Two structural IPOs are closing in. Anthropic has filed preliminary documents for Q3 2026. OpenAI is reportedly 90 days from an S-1 filing. For competing products such as Agent Wealthy, these target valuations will redefine market ranges - worth tracking closely to calibrate price positioning.
Alberta, CISA, FCA in 72 hours: the first wave of regulatory use of agentic AI
Three regulators in three different countries, in the same week, used or announced they would regulate AI models directly: Alberta (Canada) audits its code with Claude, CISA (USA) uses Mythos for federal agencies, FCA (UK) floats regulating ChatGPT, Claude and Gemini as general-purpose tools. For French or Norwegian public-sector clients, it's a signal that a regulatory precedent now exists: using a frontier coding agent in a ministry is justifiable.
📊 Trend
July 9, 2026 will be remembered as a turning point for the enterprise AI industry, with three signs aligned on the same day. First sign: OpenAI locks its alliance with Microsoft by making GPT-5.6 the default engine of the Office suite for hundreds of millions of users, and announces alongside it ChatGPT Work, its answer to Claude Cowork on the agentic desktop. Second sign: the head of OpenAI product steps back from her role the same day, confirming the 'kill what does not work, bet everything on business' strategy. Third sign: Anthropic recruits Ben Bernanke, the former chair of the US Federal Reserve, and sends a clear signal of institutional seriousness three months ahead of an IPO. Meanwhile, the government of Alberta in Canada becomes the first in the world to have a real production AI agent audit 466 million lines of its code - followed the same day by the US federal cyber agency CISA. The cross-cutting lesson: AI agents in July 2026 have left the 'impressive demo' phase and entered the 'critical infrastructure' phase, with regulatory trust being built in real time and a price war pushing rates systematically downward.