The Handover
Thursday 15 October 2026
The day in 7 stories, as Asia signs off and New York wakes up.
Security
AI-Assisted Team Breaks Chrome Sandbox in Six Days, Earns $250,000 Bounty
Xbow used OpenAI's GPT-5.6 Sol to chain three Chrome zero-days into a Windows sandbox escape in six days, potentially affecting one billion users. Google patched the flaws and paid its top full-chain bounty: $250,000.
Australia
Online safety bill to penalise platforms promoting gender-based violence
Australia's government revised its online safety bill to penalise platforms that allow advocacy of gender-based violence. It added three harm categories and removed ministerial discretion; the Greens hold the key vote.
Geopolitics
Zelenskyy says he feels 'betrayed' by Trump over Starlink restrictions
Elon Musk denied any role in Starlink cutting speeds for Ukraine's military. Volodymyr Zelenskyy accused SpaceX and the US of acting deliberately, saying the restrictions cripple drones and air defences and warning that people will die.
AI
Torvalds hands Claude Code-assisted page cache fix to Kiryl Shutsemau after race flagged
Linus Torvalds handed his Claude Code-assisted Linux page-cache fix to Kiryl Shutsemau after reviewers flagged a race and memory-reclamation concerns. A revised 10-patch series was 23 times faster in a synthetic test; real gains may be smaller.
AI
Ricursive Intelligence bets on AI-designed chips to ease the compute crisis
Ricursive Intelligence founders Anna Goldie and Azalia Mirhoseini pitched AI that designs chips, learns from each iteration and improves the next design. They argue it could ease the compute crisis without building 10,000 times more data centers.
AI
AI startup deploys undercover agents to waste scammers' time and gather fraud intelligence
Australian startup Apate says thousands of its AI agents have posed as scam victims in 2.5 million conversations, occupying 17 years of scammers' time. The agents collect intelligence for banks, telecoms, governments and utilities.
AI research
Larger optimal transport batches make flow matching more stable
UC Berkeley's Chen and Levine developed an entropic transport approximation that expands online source-target coupling from hundreds to thousands of samples. They showed larger couplings reduce stochastic-gradient variance and numerical integration error.
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.