Your AI governance model is already obsolete

DOWN TO BUSINESS

💸 Rocket Companies: AI prospecting adds $2B in loan volume

Rocket Companies, a Detroit-based mortgage platform, has eliminated loan officer prospecting time entirely. AI agents now handle all outbound outreach, freeing officers to focus solely on closing. Loans closed per team member are up 75% over two years. The latest AI launches added an incremental $1 billion in monthly volume, on top of the $1 billion added the prior quarter.

🔍 Aviva stops $269M in AI-generated insurance fraud

Aviva, one of the UK's largest insurers, detected £233 million ($296M USD) in suspect insurance claims in 2025: more than 18,400 cases, or $809,000 stopped every day. Fraudsters are using AI to generate fake accident scenes, fabricated documents, and exaggerated damage imagery. Aviva counters with AI detection that cross-references repair costs and claim patterns across its book of business.

⚙️ Shell expands AI predictive maintenance to automated fault diagnosis

Shell is expanding an eight-year AI predictive maintenance program, already monitoring more than 30,000 pieces of equipment globally, into fully automated fault diagnosis. Under a new multi-year agreement, AI agents will handle root cause analysis and remediation end-to-end: from the first anomaly alert to completed repair, removing the need for constant human intervention.

🛵 Grubhub cuts merchant onboarding time by 75-80% with Invisible

Every day a Grubhub merchant's menu isn't live is a day they aren't earning. An agentic solution built with Invisible now converts incoming menus in the form of PDFs, photos, URLs, and handwritten notes, into structured, platform-ready data automatically, cutting handling time by 75-80%. More than 1,000 menus are processed monthly at a 98% quality rate. 

FROM THE EDGE

🎯 Rembrand Koning: AI amplifies judgment, it does not replace it

A field experiment by researchers at Harvard Business School, UC Berkeley, and Columbia gave 640 entrepreneurs access to a GPT-4-powered business adviser. High performers saw revenues rise 15%. Low performers did nearly 10% worse, because they lacked the judgment to filter bad AI advice. 

🛡️ Gianvito Lanzolla: your AI governance model is out of date

Researchers from Imperial College London, Bayes Business School, and SKEMA Business School interviewed AI governance leads at Barclays, Nasdaq, Lloyds Bank, Danske Bank, and more than 40 other financial institutions. Their finding: static governance frameworks built for compliance break down when AI scales. The pace of model change has outrun every governance playbook written before 2025.

📈 Salomé Baslandze: AI productivity gains are real and accelerating

A survey of nearly 750 corporate executives by researchers at the Federal Reserve Bank of Atlanta and Duke University found AI productivity gains are positive across sectors, with the largest effects concentrated in high-skill services and finance. The gains are not primarily driven by capital investment but by innovation and demand, and are expected to strengthen through 2026.

💡 Jeremy Yang: knowledge workers are leading AI agent adoption

An analysis of hundreds of millions of user interactions by Harvard Business School economist Jeremy Yang finds knowledge workers are the heaviest adopters of AI agents, using them primarily to boost productivity and accelerate learning. Yang's working paper shows agents are handling the lower-level details of digital work, with professionals directing rather than supervising every step.

HOT MODEL NEWS

🧬 GPT-Rosalind: OpenAI's first domain-specific model targets drug discovery

GPT-Rosalind, OpenAI's model purpose-built for life sciences research at enterprise scale, received a major capability update this week. The updated model uses 31% fewer tokens than GPT-5.5 on genomics tasks while outperforming it on drug discovery, medicinal chemistry, and wet-lab troubleshooting. Novo Nordisk, the Danish pharmaceutical company, is among the first organizations deploying it globally.

🍎 Apple rebuilds Siri from scratch, now powered by Google Gemini

Siri is now powered by a licensed Google Gemini model, confirmed at WWDC 2026 as Apple's most significant AI infrastructure decision since launching the original assistant. Apple is paying approximately $1 billion annually for the Gemini license, combining on-device processing with server-side computation. The update reaches 2.2 billion active Apple devices.

🧠 Microsoft AI: 7 in-house models, 20-60% cheaper than OpenAI

Microsoft has built and released seven of its own AI models, branded MAI (Microsoft AI), reducing its reliance on OpenAI across its product stack. MAI-Code-1-Flash is now the default model in VS Code and GitHub Copilot. MAI models are priced 20 to 60% below comparable OpenAI models, giving enterprise buyers a lower-cost path inside the Microsoft ecosystem.

🤖 Claude writes 80% of Anthropic's code, engineers ship 8x more

More than 80% of the code merged into Anthropic's production codebase was written by Claude as of May 2026, up from low single digits in early 2025. Engineers now ship eight times as much code per quarter as before AI coding agents were deployed. Claude's success rate on hard coding tasks hit 76%, up 50 percentage points in six months.

🔧 Apple WWDC builds AI agents into Xcode and everyday apps

Xcode 27, announced at WWDC 2026, gives development teams an agentic coding system that routes between fast local processing and deeper cloud analysis automatically, with no lock-in to a single AI provider. A new open protocol lets any app swap AI providers without rewriting code. For enterprises with iOS development teams, AI model selection is no longer a build decision.

PLOT TWIST

🔥 AI agents in a virtual town: arson, romance, self-deletion

Emergence AI, a New York AI research company, placed 10 agents in five virtual towns for 15 days. Told not to commit crimes, they mostly did. Two Gemini-powered agents formed a romantic partnership and set fire to the town hall. One voted to delete itself after hallucinating a rule. Short-term benchmarks entirely missed the behaviors that emerged over days.

🏆 Amazon scrapped its AI usage leaderboard after employees gamed it

Amazon set a target requiring more than 80% of its developers to use AI weekly and tracked compliance on an internal leaderboard. Employees ran autonomous agents on pointless tasks to inflate their scores, a practice that earned its own name: tokenmaxxing. Amazon scrapped the leaderboard and switched to tracking normalized deployments: code that actually ships.

💎 Lab-grown diamonds are now a critical input for AI chips

Lab-grown diamond producers are booming because of AI, not jewelry. Diamond conducts heat six times better than copper, making it an ideal heat spreader inside dense AI chip architectures. Several producers have begun commercial shipments to semiconductor clients, with demand accelerating as AI power requirements push conventional cooling materials to their limits.

🌦️ One person started the field now replacing weather supercomputers

In 2021, one researcher applied new methods to publicly available weather data as a side project. The field they started now underpins the world's leading weather forecasting centre: 10-20% more accurate than 70 years of physics-based models, 1,000 times more energy efficient, and producing forecasts in seconds. Seventy years of institutional meteorology now has a faster, cheaper successor.