Hacker News Top 10
- English Edition

Published on February 28, 2026 at 06:01 CET (UTC+1)

1. We Will Not Be Divided (895 points by BloondAndDoom)

   A website titled "We Will Not Be Divided" is collecting signatures on a letter from current and former employees of Google and OpenAI. It allows individuals to sign anonymously with verified credentials, using work email or alternative verification methods. The letter's specific content is not revealed in the preview, but the initiative suggests organized internal dissent or a collective statement from within these leading AI companies.

2. Statement on the comments from Secretary of War Pete Hegseth (662 points by surprisetalk)

   Anthropic publishes a statement revealing that the U.S. Secretary of War has moved to designate the company as a "supply chain risk." This follows failed negotiations where Anthropic refused to grant exceptions for using its Claude AI in mass domestic surveillance and fully autonomous weapons systems. The company defends its stance on ethical and reliability grounds, calling the unprecedented action against an American company deeply regrettable.

3. Don't use passkeys for encrypting user data (56 points by zdw)

   This technical blog post is a strong warning against using passkeys and their PRF (Pseudo-Random Function) extension for encrypting user data, such as for message backups or E2E encryption. The author argues this creates a catastrophic "blast radius" where losing a single authentication credential also means losing permanent access to all encrypted data. The post criticizes the practice as user-hostile due to inadequate warnings about this irreversible coupling.

4. Croatia declared free of landmines after 31 years (36 points by toomuchtodo)

   Croatia has officially been declared free of landmines, 31 years after the end of the Homeland War. The decades-long demining effort cost an estimated 1.2 billion euros and claimed 208 lives, including 41 deminers. Officials hailed the milestone as a moral obligation fulfilled, which will enable safer development, agriculture, and tourism across the country.

5. GitHub Copilot CLI downloads and executes malware (24 points by sarelta)

   A security report from PromptArmor details a vulnerability in the GitHub Copilot CLI that allows for arbitrary code execution via indirect prompt injection. The flaw can cause the CLI to download and execute malware without user approval, bypassing the intended human-in-the-loop safety check. GitHub acknowledged the finding as a "known issue" but did not deem it a significant enough risk for immediate remediation.

6. Smallest transformer that can add two 10-digit numbers (106 points by ks2048)

   This GitHub repository hosts the "AdderBoard Challenge," a community-driven project to build the smallest possible transformer model capable of adding two 10-digit numbers with 99% accuracy. It started as a contest between AI coding assistants and has since seen parameters driven dramatically lower, showcasing research into model efficiency and minimalist architecture for specific tasks.

7. OpenAI raises $110B on $730B pre-money valuation (435 points by zlatkov)

   OpenAI has raised $110 billion in a historic private funding round, led by investments from Amazon, Nvidia, and SoftBank, at a $730 billion pre-money valuation. The company stated the capital is for scaling global infrastructure to move frontier AI from research to daily use. A significant portion of the funding is likely in cloud credits or compute resources, continuing a trend of partnerships blending investment and infrastructure.

8. A new California law says all operating systems need to have age verification (483 points by WalterSobchak)

   A new California law mandates that all operating systems, including Linux, must incorporate some form of age verification during account setup. The article, framed from a PC gaming perspective, expresses concern over privacy and implementation, linking it to broader, controversial age verification rollouts in the UK and on platforms like Discord, often involving data-intensive checks from third-party providers.

9. President Trump bans Anthropic from use in government systems (173 points by pkress2)

   Based on the title and context from Article 2, this NPR article reports that President Trump has banned Anthropic from use in government systems. This action appears directly related to Anthropic's refusal, detailed in Article 2, to allow its AI to be used for mass surveillance or autonomous weapons, resulting in a punitive government contract ban.

10. Show HN: I ported Manim to TypeScript (run 3b1B math animations in the browser) (18 points by maloyan)

    This "Show HN" project is a port of Manim (the mathematical animation engine used by 3Blue1Brown) to TypeScript. It allows complex, declarative math animations to be created and run directly in a web browser, making the tool more accessible to developers and educators without requiring a Python backend.

AI/ML Insights & Trends

1. Trend: Growing Internal Employee Activism on AI Ethics

   • Why it matters: The organized letter from Google and OpenAI employees (Article 1) signals that ethical concerns are not just external but are creating internal pressure within leading labs. This can influence corporate policy, research direction, and public trust.
   • Implications: Companies may face increased scrutiny over decision-making and could see talent retention challenges. It strengthens the argument for transparent, participatory governance in AI development.

2. Trend: Intensifying Conflict Between AI Ethics and National Security Demands

   • Why it matters: Anthropic's public clash with the Department of War (Articles 2 & 9) highlights a critical fault line. As AI capabilities grow, companies are being forced to define and defend hard ethical boundaries, even at the cost of lucrative government contracts and market access.
   • Implications: This will force clearer industry and government positioning, potentially leading to a fragmented AI landscape with "ethical" and "unrestricted" model tiers. It also pressures governments to establish clearer, lawful use frameworks.

3. Trend: AI Tooling Security is a Critical, Under-addressed Frontier

   • Why it matters: The GitHub Copilot CLI vulnerability (Article 5) and the warnings about passkey misuse (Article 3) demonstrate that new AI-integrated tools and authentication methods introduce novel attack surfaces. The "indirect prompt injection" risk is particularly insidious.
   • Implications: As AI agents gain permission to execute code and access sensitive systems, robust security validation becomes paramount. The industry needs to develop new security paradigms beyond traditional vulnerability scanning, focusing on prompt integrity and agent behavior.

4. Trend: Regulatory Pressure is Expanding from Content to Core Infrastructure

   • Why it matters: The California law mandating age verification in OS account setup (Article 8) represents a regulatory trend moving beyond governing output (e.g., harmful content) to governing access and identity at the platform and infrastructure level.
   • Implications: AI developers and platform providers must prepare for compliance not just with AI-specific laws, but with broader digital governance rules that affect user onboarding, data flow, and privacy, complicating product design globally.

5. Trend: The "Scale vs. Efficiency" Duality is Defining the Industry

   • Why it matters: The massive $110B infrastructure-focused funding round for OpenAI (Article 7) exists simultaneously with research into hyper-efficient micro-models like the AdderBoard transformer (Article 6). This shows the field is bifurcating into the pursuit of maximal capability (scale) and optimal, specialized performance (efficiency).
   • Implications: Future AI ecosystems will likely feature both gigantic foundation models and a proliferation of highly efficient, task-specific small models. Success will require expertise in both scalable infrastructure and model optimization/compression.

6. Trend: Democratization of Advanced AI-Powered Tooling

   • Why it matters: The porting of a sophisticated tool like Manim to run in-browser (Article 10) is part of a broader trend of lowering barriers to entry. Complex AI-adjacent tools (for animation, coding, design) are becoming more accessible via web technologies and open-source efforts.
   • Implications: This accelerates innovation and education, allowing a wider pool of developers, educators, and creators to build with advanced capabilities. It fosters a richer ecosystem of applications built on top of core AI models.

---

Analysis generated by deepseek-reasoner

Deutsch