Anthropic AI Updates: July 28, 2026

1. Anthropic Clarifies Its Position on Open-Weights Models

Anthropic. Anthropic published a post stating it has never advocated for a ban on open-weights models, and framed its concerns around two national security risks: authoritarian governments building more powerful AI than the United States, and sufficiently capable models being misused for cyberattacks or biological attacks. Rather than restricting open weights, the company argues for three targeted measures, declining to sell powerful chips or chipmaking equipment to China, cracking down on industrial-scale distillation that accelerates rival model development, and requiring mandatory safety evaluations for all sufficiently capable models whether open or closed. The post maintains that the value of open-weights models should be judged through empirical testing rather than assumptions about safety tradeoffs. Source

2. Cognizant and Anthropic Expand Enterprise Partnership

Anthropic. Anthropic and Cognizant deepened their partnership to deploy Claude across enterprise clients in manufacturing, life sciences, insurance, and related sectors. Cognizant has trained more than 30,000 associates on Claude and is integrating it into platforms including Flowsource and Neuro AI Engineering, with its Spec-Driven Development module now using Claude Code to generate and evaluate code against project specifications. The company reported client production systems built on Claude, including a customer experience portal delivered in six months, a contract-intelligence system that cut review time by 40 percent, and a risk-navigation tool that saves underwriters roughly eight hours weekly per deployment. Cognizant is becoming a Global Premier Partner in the Claude Partner Network and is establishing a Frontier Certified workforce model to scale Claude expertise. Source