The Handover

Tuesday 3 November 2026

The day in 7 stories, as Asia signs off and New York wakes up.

Illustration: Anthropic swings to operating loss as revenue growth slows, revises S-1 ahead of IPO
  1. Business

    Anthropic swings to operating loss as revenue growth slows, revises S-1 ahead of IPO

    Anthropic reported $17bn in third-quarter revenue but a $567m operating loss as compute costs rose, its amended IPO filing shows. Compute and infrastructure spending commitments total roughly $895.4bn.

  2. Security

    Google takedown leaves XCPE malware on infected TV boxes as Synthient founder says scale may be higher

    Synthient found the XCPE backdoor in a MECOOL Android TV box; Google attributes it to manufacturer Videostrong. Synthient's founder called Google's count of over 1.3 million infected devices 'highly conservative'.

  3. Security

    FT says phone numbers were enough to target German officials with spyware

    Germany is investigating possible spyware attacks on lawmakers and federal officials after Apple warned their iPhones may be compromised. The FT reported that attackers needed only targets' phone numbers.

  4. AI

    LLMs can read and write neural network weights, researchers demonstrate

    Researchers including Alex Zhang showed that language models can read and edit another model's weights, fix bugs in one step and write weights from scratch. The work remains limited to simple tasks.

  5. AI

    Developers report Claude refusals disrupting work

    Anthropic added Claude Code restrictions to prevent game cloning. Developers report stalled paid projects; the company has not confirmed whether reverse-engineering refusals stem from the change.

  6. Business

    Marvell's $115B Lumentum deal: shares fall as valuation raises questions; CEOs tout AI connectivity

    Marvell and Lumentum announced a $115bn all-stock acquisition deal focused on AI connectivity. Marvell shares fell 8% and Lumentum shares fell 13% after the announcement.

  7. AI research

    AI agents cut prompt injection success from 79% to 9%

    UC Berkeley and Anthropic researchers cut prompt-injection success from 78.8% to 8.8% on an agent benchmark under the AttackerMovesSecond attack. They reported no noticeable losses on nine benchmarks of legitimate tasks.

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.