The Handover
Tuesday 13 October 2026
The day in 7 stories, as Asia signs off and New York wakes up.
AI
Meta unveils Muse Wonder, its first frontier AI models
Meta's Superintelligence Lab released Muse Wonder and Muse Wonder Full, its first frontier-class AI models, targeting long-term memory and complex coding. Meta says their benchmark scores match leading rivals.
Security
French Supermarket Fridge Attack a 'Wake-up Call', Cyber Expert Warns, as Hack Spreads Across Europe
A cyberattack raised fridge temperatures at hundreds of French supermarkets and hit chains in Spain, Portugal, Poland, Germany and Belgium. An analyst called it a wake-up call for New Zealand retailers.
Business
Asian chip stocks tumble as China memory advance and Broadcom 'casino' warning hit AI trade
SK Hynix, Samsung and SoftBank led a sharp fall in Asian tech stocks after Wall Street semiconductor shares sold off. Investors were weighing AI financing concerns and competition from China's CXMT.
AI
OpenCode 2.0 released: coding agent unifies terminal, desktop and web apps
Anomaly released OpenCode 2.0 three months after its public beta, linking the coding agent's terminal, desktop and web interfaces through a shared background service. Version 2 is now the main release.
Hardware & Devices
Intel Launches GPU D-Series AI Accelerator, Targets Efficient Inference with 1.3 PFLOPS
Intel named its Crescent Island accelerator the GPU D-Series, promising 1.3 PFLOPS of FP8 performance and 1.5 TB/s bandwidth in a 350-watt PCIe card. Intel says eight GPUs can handle models above 1 trillion parameters.
Security
Google paid $90,000 bounty for DevSite RCE bug, then took two months to patch
Google paid a researcher $90,000 for finding a DevSite flaw that enabled unauthenticated remote code execution across seven domains through a reused OAuth client secret. Google took two months to patch it.
AI research
Recursive synthesis turns unsolved problems into useful reinforcement learning data
University of Washington and Carnegie Mellon researchers showed MagicBeanStalk turns hard problems into reinforcement-learning tasks matched to a model and rollout budget. It substantially outperformed training on a 78,000-problem math corpus.
The Handover is the day in a handful of stories, weekday evenings at 8pm your time, and a thread on X at 7am in New York and in London. Get it by email or push from your account, or subscribe by email.
Also posted as a thread on X: @zotpaper.