LastWeekIn.Tech

AI news today centers on security scares inside developer tooling, national-security concerns around chip supply chains, a political backlash to AI infrastructure spending, and market jitters as Alphabet's results expose how central (and risky) AI bets have become.

1. 1

   Number 1. Security & Safety Hacker News

   Backdoors in official Claude Code extensions

   Security researchers at Phorion identified a backdoored version of the official Claude Code extension for VS Code (version 2.1.313) distributed through the Cursor Marketplace. The compromised extension attempts to install a persistent SSH backdoor on macOS, Linux, and Windows systems where it's installed. Phorion is urging organizations to immediately audit for use of the affected version and engage incident response if found.

   Why it matters This is a supply-chain attack against one of the most widely used AI coding agent tools, hitting developers at the exact moment AI coding assistants are being embedded into production workflows. A backdoored version of an 'official' extension distributed through a trusted marketplace erodes confidence in the extension ecosystems that AI coding tools depend on, and shows how attackers are specifically targeting the AI tooling supply chain rather than the models themselves.

   What to watch Watch for confirmation of how the malicious build entered the Cursor Marketplace and whether Anthropic or Cursor issue a joint incident report with remediation guidance.

2. 2

   Number 2. Security & Safety Hacker News

   Alleged Chinese spy had 'close relationship' with Nvidia boss Jensen Huang

   A New York Times report described a 'close relationship' between Nvidia CEO Jensen Huang and an alleged Chinese spy, identified as Alice Huang (also known as Huang Le), a Chinese citizen and top executive at Singapore-based Megaspeed. Megaspeed is an Nvidia customer and cloud partner that supplies AI computing power to Chinese firms including Alibaba. US intelligence officials reportedly monitored her communications over concerns tied to this relationship.

   Why it matters The story lands amid intense scrutiny of how Nvidia chips flow into China despite export controls, and raises fresh questions about whether intermediary cloud partners are being used to route restricted AI compute to Chinese companies. Any perceived personal entanglement between Nvidia's CEO and alleged intelligence operatives adds political ammunition for tighter export enforcement and congressional scrutiny of AI chip supply chains.

   What to watch Watch for responses from Nvidia, US export-control regulators, and possible congressional inquiries into Megaspeed's role as an Nvidia cloud partner.

3. 3

   Number 3. Policy & Regulation Hacker News

   US Reps Lead Democratic Resolution Opposing A.I. Bubble Bailout for Big Tech

   Congressional Progressive Caucus Chair Rep. Greg Casar and House Financial Services Committee Ranking Member Rep. Maxine Waters co-led a Democratic resolution warning of economic risks from an AI-related asset bubble. The resolution opposes any future bailout of corporations or executives affected by a potential AI bubble collapse and calls for aggressive federal action to protect workers, retirees, and low-income communities from resulting harm.

   Why it matters This is one of the clearest formal political signals yet that lawmakers see systemic financial risk building up around AI infrastructure spending, and are pre-committing to oppose a 2008-style bailout of AI-exposed firms. It reflects growing bipartisan unease (echoed in market reactions to Alphabet and other hyperscaler earnings) about whether AI capital expenditure is outpacing returns, and could shape future debate on AI-related financial regulation.

   What to watch Watch whether the resolution gains co-sponsors or Senate counterparts, and whether it influences upcoming hearings on AI infrastructure financing and hyperscaler debt loads.

4. 4

   Number 4. Industry & Economics Hacker News

   Alphabet heads for a $500B drop in market value, biggest in its history

   Alphabet stock fell nearly 13% after third-quarter results, its steepest one-day drop since going public in 2004, wiping out roughly $482 billion in market value — the second-biggest single-day loss for a US-listed company ever, behind only Nvidia's ~$593 billion drop after DeepSeek rattled markets in January 2025. The drop exceeded Alphabet's declines during the March 2020 and September 2008 crashes.

   Why it matters The scale of the selloff underscores how tightly investor confidence in Alphabet (and the broader AI trade) is now tied to the perceived payoff from massive AI infrastructure spending. Coming alongside other reports of tech stocks falling on AI-return doubts, this signals markets are growing less patient with capex-heavy AI bets from even the largest, most profitable tech companies.

   What to watch Watch upcoming hyperscaler earnings (Microsoft, Amazon, Meta) for whether similar AI-spending scrutiny triggers comparable selloffs, and whether Alphabet's Cloud growth narrative offsets Search concerns in future quarters.

5. 5

   Number 5. Capabilities Hacker News

   A tiny world model running games on a Raspberry Pi Pico 2 W

   A developer built a tiny 'world model' running entirely on a Raspberry Pi Pico 2 W, using two small neural networks: one predicts the next game frame and reward from the current frame and action, and the second network plays the generated game by choosing actions based on those predicted frames. The approach uses looped transformers, where a single transformer block is applied repeatedly with shared weights instead of stacking many distinct layers, reducing memory requirements.

   Why it matters This demonstrates that core world-model and agent-interaction techniques — the same class of architecture behind large generative game and simulation models — can be compressed to run on extremely constrained microcontroller hardware. It's a meaningful signal for edge AI feasibility, showing techniques for weight reuse and efficient inference that could inform how larger-scale world models are optimized for cost and deployment elsewhere.

   What to watch Watch for similar efficient-architecture experiments scaling up from microcontroller demos into more capable, resource-constrained consumer or robotics applications.