Anthropic reopens Pentagon talks after AI safety clash

Here’s what’s worth your time today.

Anthropic has resumed negotiations with the Pentagon after a dispute over how the military could use its AI models. Talks broke down when the company demanded assurances that its technology wouldn’t be used for mass surveillance of Americans or autonomous weapons, prompting the Defense Department to briefly label Anthropic a potential supply-chain risk. Discussions have now restarted, raising the possibility of a new agreement that would allow the military to continue using Anthropic’s models while addressing safety concerns and intensifying competition with OpenAI for defense AI contracts.

Supported by Descope

Customer SSO is never a “one and done” thing. For each customer you onboard, your engineers and support teams need to spend time configuring IdPs, mapping user and group attributes, testing domains…the list goes on.

The Descope SSO Setup Suite lets your customer admins set up SSO and SCIM connections entirely on their own.

Join 1000+ orgs and begin your “drag & drop SSO” journey.

OpenAI has introduced a new research framework designed to measure how using ChatGPT affects long-term learning. The system, called the Learning Outcomes Measurement Suite, aims to track how AI tools influence things like persistence, motivation, and creative problem-solving over time. While studies suggest AI tutoring can improve short-term recall, there’s still little data on whether heavy chatbot use strengthens or weakens deeper cognitive skills. As schools increasingly integrate AI tools from companies like OpenAI, Anthropic, and Google, the framework is meant to help educators understand whether these systems ultimately support learning or quietly undermine it.

A new BCG report shows companies are increasing spending on AI tools, with about one in four communications leaders planning to allocate more than 10% of their budget to AI in 2026. But most organizations still aren’t ready for real transformation. Around 88% say they aren’t prepared to lead AI adoption in their teams, and only 31% have scaled generative AI beyond pilot projects. While some companies are saving time and money with AI, few are reinvesting those gains into new capabilities, highlighting a growing gap between rising AI investment and meaningful organizational change.

OpenAI has released a new model, GPT-5.4, along with tools that let ChatGPT work directly inside Excel and Google Sheets. The update is designed to be more reliable and efficient, with stronger performance on workplace tasks like drafting documents and analyzing data. According to OpenAI, the model outperformed office workers 83% of the time on its internal GDPval benchmark. The company is also launching OpenAI for Financial Services, with partners including FactSet, MSCI, Third Bridge, and Moody’s, as it pushes deeper into enterprise workflows.