AI Round-up: May 17, 2026

Anthropic's Mythos dominated the snapshot through a claimed Apple M5 memory exploit, with Reddit threads, security-focused posts and high-view YouTube coverage turning it into the clearest cross-source story. The same community layer stayed developer-heavy: Codex users compared workflows against Claude Code, Claude Code users traded operational tips, and broader research safety posts about AI psychosis and ArXiv's AI-generated-paper ban kept the trust and misuse angle live.

OpenAI supplied the main official story by partnering with Malta to give citizens ChatGPT Plus access after an AI literacy course, backed by both its own announcement and Reuters coverage. Around that, the day was more fragmented but still agent-centred: xAI extended Hermes Agent with X Premium subscriptions and X search, Google signals ranged from Gemini Spark leaks to Fitbit and smart-glasses speculation, and news coverage tracked Greg Brockman's product role, AI education effects, Cerebras market momentum and data-centre energy bets.