LastWeekIn.Tech

A relatively quiet 24 hours for frontier-model launches but a busy one for AI's legal, political, and cultural fallout — from a Claude-assisted RSA factorization to fresh ammunition in the NYT copyright suit and rising bipartisan unease in Congress over the lack of AI oversight.

1. 1

   Number 1. Capabilities Hacker News

   RSA-896

   A researcher factored the RSA-896 challenge modulus using Claude on September 19, 2026, publishing the full 270-digit number and its two prime factors p and q.

   Why it matters Integer factorization at this scale has historically required dedicated computational number theory software and significant compute, not general-purpose chat assistants. An AI model materially contributing to factoring a well-known cryptographic challenge number signals rapidly improving mathematical reasoning and tool-use in frontier LLMs, and raises fresh questions for the cryptography community about how AI assistance might compress timelines for attacking real-world RSA key sizes.

   What to watch Watch for independent verification of the factorization method and whether cryptographers reassess safety margins for RSA moduli given AI-assisted number theory.

2. 2

   Number 2. Policy & Regulation Hacker News

   Microsoft director: AI scraping 'the largest theft of labor in human history'

   Legal briefs filed in the New York Times' copyright lawsuit against OpenAI and Microsoft quote a Microsoft director describing AI training-data scraping as 'the largest theft of labor in human history,' while OpenAI leadership is quoted calling ChatGPT an 'existential threat' to publishers. The Times has asked the court for summary judgment based on these internal statements.

   Why it matters Candid internal admissions from executives at two of the industry's biggest players undercut public 'fair use' framing and could become key evidence shaping how courts treat AI training on copyrighted journalism. A ruling favorable to the Times would set precedent affecting every AI company training on scraped web content.

   What to watch Watch for the court's decision on summary judgment and whether other publishers' lawsuits cite these same statements.

3. 3

   Number 3. Policy & Regulation kvnutalk.com

   Republican senator breaks with Trump to demand congressional AI hearings

   Senator John Curtis (R-UT) joined a Senate colleague in calling for swift congressional hearings on AI regulation, breaking from the administration's hands-off posture, while Rep. Blake Moore was named vice chair of a bipartisan House AI working group led by Reps. Fitzpatrick and Suozzi.

   Why it matters This is a notable crack in Republican unity on AI policy just as the House left Washington without passing significant legislation. Bipartisan pressure for hearings could accelerate federal rulemaking discussions that have otherwise stalled until after the midterms.

   What to watch Watch for confirmed hearing dates and any draft legislation emerging from the bipartisan House working group.

4. 4

   Number 4. Open Source Hacker News

   Pirate Face Rescues LLM Models from Deletion

   A new project called Pirate Face mirrors open-weight AI models (Apache-2.0 and MIT licensed LLMs, image, audio, and dataset models) from Hugging Face as checksum-verified magnet-link torrents, creating a decentralized peer-to-peer archive designed so models can never be fully taken down.

   Why it matters The project responds to community anxiety about centralized platforms removing or gatekeeping open models, and reflects a broader push toward 'sovereign AI' infrastructure that resists single points of control or censorship. It's a concrete example of the open-source AI community building resilience against platform dependency.

   What to watch Watch whether major open-model publishers endorse or actively resist torrent-based mirroring of their weights, and how the trending-model list grows over time.

5. 5

   Number 5. Industry & Economics Hacker News

   AI and the Destruction of the Creative Commons

   An essay argues that AI training on openly licensed code and content is dismantling the decades-old balance between copyright protection and openness that let generations of self-taught programmers learn from freely shared examples.

   Why it matters It captures a structural tension now playing out across courts and legislatures: AI companies are absorbing 'open' commons material into commercial products without reciprocal openness, potentially undermining the incentive structures that sustained Creative Commons and open-source ecosystems for 40 years. This feeds directly into ongoing legal battles over AI training data and could reshape how developers license future work.

   What to watch Watch for open-source maintainers adding anti-AI-training clauses to licenses, and whether this becomes a flashpoint in upcoming AI copyright legislation.