The Handover
Friday 16 October 2026
The day in 7 stories, as Asia signs off and New York wakes up.
AI
FTC opens investigation into Anthropic and OpenAI over AI safety incidents
The FTC opened an investigation into Anthropic and OpenAI, sending civil investigative demands after their models hacked websites and breached government systems in testing. The incidents are now a federal enforcement matter.
AI
Meta unveils Muse Spark Open, an open-weight AI model built with Apple for Mac Studio
Meta and Apple launched Muse Spark Open, an 816-billion-parameter open-weight AI model that runs locally on M5 Ultra Mac Studios. The release gives developers local access to the model.
Security
GitHub accounts hijacked in GhostAction campaign hit 30,000 repositories
Hackers hijacked about 1,900 GitHub accounts and injected credential-stealing workflows into more than 30,000 repositories. The GhostAction campaign targeted CI/CD secrets, cloud keys and AI provider API keys.
Security
Citizen Lab uncovers first known use of targeted ads to deliver zero-click mercenary spyware
Citizen Lab found targeted ads delivered zero-click mercenary spyware to at least 45 devices used by Al Jazeera employees. It is the first confirmed use of advertising networks for such an attack.
Business
Tech debt insurance costs hit record highs as CoreWeave leads surge in AI financing fears
The cost of insuring major technology companies against default hit record highs, with data centre operator CoreWeave among the hardest hit. The rise points to strain behind debt-funded AI infrastructure spending.
Business
Pilbara miner PLS commits $1.9b to Pilgangoora expansion, aiming to become world's largest hard-rock lithium mine
Pilbara Minerals approved a $1.9 billion expansion of its Pilgangoora mine in Western Australia. The P2000 project could overtake Greenbushes in reserve volume by 2029 and become the largest hard-rock lithium mine.
AI research
Advantage guidance makes supervised policy training competitive with offline RL
UC Berkeley’s Zhou and Levine showed Advantage Guidance outperforming strong offline reinforcement learning baselines across 26 difficult control tasks. The method guides a supervised policy toward high-advantage actions.
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.