Digital CxO

"Keep a human in the loop" is the most trusted safety answer in institutional AI, and at agent speed and agent scale it becomes one of two things, a bottleneck or a fiction. The real question was never whether humans belong in the loop but where their authority should sit, and the answer is calibrated placement at the reasoning layer, where aligned interpretations proceed in milliseconds and only the divergent few reach a reviewer with real evidence and real authority. Read the full argument on why the blanket version of human oversight is obsolete: buff.ly/AftYi2F #AI #AIAgents #AIGovernance #HumanInTheLoop #EnterpriseAI

SAP used its TechEd event to embed AI deeper into enterprise workflows, extending its Knowledge Graph to third-party data, making more than 300 AI agents discoverable via MCP and A2A, and adding an Enterprise AI Gateway to centrally manage authentication, routing and spend. As SAP CTO Philipp Herzig put it, the goal is a truly autonomous enterprise that doesn't just add AI agents but reinvents the machinery under the hood, and the bar has never been higher. Read the full article on what SAP announced and why general-purpose models alone won't get organizations there: buff.ly/AzHeJ5m #SAP #AI #AIAgents #EnterpriseAI #MCP #AIGovernance

Anthropic has acquired Mithril Medicine, a one-person biotech startup whose founder used Claude to take an experimental LAMA2 muscular dystrophy therapy from ideation to positive preclinical mouse trials in under 18 months, with human trials possible by late 2028. Alongside the deal, Anthropic published early results from six partners, including Generate Biomedicines engineering a CAR-T component about 70% shorter than standard analogs and Enveda processing 1,061 pages of manufacturing records in under three hours instead of 25. Read the full article on what those collaborations produced and where Anthropic's life sciences push goes next: buff.ly/vs7Z335 #AgenticAI #Anthropic #DrugDiscovery #LifeSciences #Claude

Political campaigns are using generative engine optimization to influence how AI chatbots describe candidates ahead of the midterms, raising concerns about misinformation and voter trust. Read more: buff.ly/YHp34Qc #AI #ElectionSecurity #GenerativeAI #PoliticalTechnology #AIGovernance

I would like Nick Cammarata to be right, and I would also like us to resist turning a promising research argument into permission to declare the safety problem solved. The Goodfire researcher and former OpenAI staffer set off a substantial debate with his essay, "Neural Networks Are Already Clean," arguing that these systems may be considerably more understandable than researchers, including himself, previously believed. Knowing how something works can make it easier to defeat, so if we can see more of what these systems are doing, we should expect more from the companies building them and the organizations putting them to work, not less. Read the full piece on why useful progress doesn't require complete understanding, and why it shouldn't excuse us from building the controls we already know how to build: buff.ly/qxnko7y #AI #AISafety #Interpretability #AIGovernance #AIAgents

A Cloudera survey found 89% of enterprises have limited or redesigned AI use cases due to rising costs, while only 23% say AI delivers more value than it costs. Read more: buff.ly/sAZU8QP #AI #EnterpriseAI #AIInfrastructure #AIGovernance #CloudComputing