AI Round-up: 17 October 2026

Anthropic reports that Claude Mythos 5 designed more resource-efficient quantum circuits using automated evaluation, a research result rather than a demonstration of practical quantum computing. A separate Claude investigation traces the origins of standard ECDSA curve parameters; it does not break the encryption. OpenAI's GPT-6 Astra Pro update brings parallel agents and improved tool use to ChatGPT, while Google demonstrates Gemini agents collaborating in Minecraft.

The FTC chair discusses scrutiny of Anthropic and OpenAI over AI safety incidents, alongside reporting on AI borrowing and CoreWeave-linked valuation pressures. Anthropic's workplace research examines changing hiring and the burden of reviewing agent output. YouTube mixes AI-market commentary with browser and coding workflows; Reddit discussions cover Copilot integrations, protecting production data from agents, Claude Code explanations and coding-plan comparisons.