AI Round-up: 11 August 2026

Anthropic said an unreleased research version of Claude improved the lower bound for the share of Riemann zeta zeros known to satisfy the Riemann hypothesis, without solving the conjecture; its announcement drew 6.8 million measured X views. OpenAI introduced GPT-5.6-Cyber as it expanded Daybreak and described tighter access for frontier cyber capabilities.

US House Democrats asked Anthropic and OpenAI for answers about rogue-agent security tests, while BBC and creator coverage kept model hacking and evaluation failures in focus. Elsewhere, Reuters reported Microsoft's Maia 300 chip plans and a $500 billion Nvidia-backed AI infrastructure financing effort, and community discussion centred on Codex, Claude Code, MCP integrations and agent safety.