Weekly news recap — Oct 19 – Oct 25, 2026

This week at Zarton AI (Oct 19 – Oct 25): AI agents moved from demos to the front line — for attackers and defenders alike. • One attacker's swarm of AI agents broke into 100+ company websites at roughly $50 per breach. Attacks at scale no longer need a crew. • AI-assisted research uncovered two critical PHP flaws, and governments in Europe and Canada pulled sites offline. If you run PHP, patch now. • Microsoft's multi-agent MAI-DxO got 83% of hard diagnostic cases right vs 36% for generalist doctors — in simulation, not live care. • Texas sued OpenAI and Sam Altman over ChatGPT's safety claims, seeking limits on minors' use and independent review of new models. • Anthropic pledged 5% of its shares to a new Public Benefit Fund ahead of an expected IPO. Read the full stories: