Anthropic's AI agents outnumber humans 6 to 1

OpenAI Blog

OpenAI released MultiLangBench, a new evaluation benchmark that tests AI models on coding tasks in low-resource languages like COBOL and Fortran by having them build a SQLite-compatible database across 20 languages.

OpenAI

OpenAI's GPT-5.6 model discovered critical PHP vulnerabilities before attackers, prompting the company to propose a new coordinated disclosure framework for the AI era and commit $1 billion to subsidized AI access for security defenders as AI-discovered vulnerabilities are expected to surge.

Anthropic

Anthropic launched Claude Duet, a proactive AI agent app that autonomously manages daily tasks like scheduling and email prioritization while competing directly with OpenAI's Dots and Grok Bot in the persistent AI assistant market.

TechCrunch

Google and Amazon are simultaneously launching PyTorch variants for their proprietary AI chips to challenge Nvidia's GPU dominance, allowing developers to switch hardware platforms with minimal code changes. This signals an industry-wide push to reduce vendor lock-in and commoditize AI hardware infrastructure.

Anthropic

Anthropic's workforce study shows AI agents now work 6x more hours than humans, driving a sharp decline in junior technical hiring (27% to 9%) while productivity surged across roles. The company is managing disruption through early warning systems, internal redeployment, and shared ownership, as work shifts from execution to managing AI agents and strategic decision-making.