Anthropic confronts the RSI clock
This edition of The Rundown AI newsletter covers Anthropic's report on recursive self-improvement (RSI), noting that Claude already authors over 80% of Anthropic's merged code and is accelerating AI development faster than expected. It also covers OpenAI's new 'dreaming' memory system for ChatGPT, a coalition of AI CEOs urging Congress to regulate synthetic DNA to prevent bioweapon risks, and several other AI industry developments.
Summary
The newsletter's lead story covers Anthropic's newly published report titled 'When AI builds itself,' which examines recursive self-improvement (RSI) in AI systems. Anthropic disclosed that Claude now authors more than 80% of the company's merged code as of May 2026, and that engineers are pushing 8x as much code per day in Q2 2026 compared to 2024. Co-author Jack Clark described a trajectory where each new version of Claude could theoretically be built by the previous version without human involvement. Anthropic clarified that full RSI is not yet here or inevitable, but acknowledged the pace is faster than anticipated. The company also stated it would consider slowing or pausing frontier AI development if peer labs agreed to do the same, with policy discussions planned for coming months. OpenAI echoed similar RSI concerns in its own 'Democratic Governance of Frontier AI' blueprint.
The newsletter also covers OpenAI's major memory update for ChatGPT, dubbed 'dreaming.' This system runs in the background to build a continuously updated, category-sorted profile of each user based on past conversations. The update significantly improved ChatGPT's factual recall from 41.5% to 82.8% and preference-following from 31.4% to 71.3% in internal evaluations. The feature is rolling out initially to Plus and Pro users in the U.S.
On the AI safety front, CEOs from OpenAI, Anthropic, Google DeepMind, and Microsoft — including Sam Altman, Dario Amodei, Mustafa Suleyman, and Demis Hassabis — signed an open letter to Congress warning that AI systems now outperform PhD-level virologists in technical knowledge, enabling bad actors to potentially design bioweapons. They urged Congress to mandate that synthetic DNA and RNA sellers screen orders, verify buyers, and log sales.
Additional highlights include: Nvidia releasing the open 550B Nemotron 3 Ultra reasoning model; Canadian PM Mark Carney unveiling a five-year AI national strategy targeting $200B in growth; the U.S. and Japan announcing a $1B AI research partnership; CloudFlare co-founder Matthew Prince revealing that bot traffic now surpasses human traffic on the internet; and a reader workflow featuring AI-assisted parenting tips for managing toddlers.
About this episode
PLUS: Stress test business ideas with Perplexity
Key Insights
- Anthropic's internal data shows Claude authored over 80% of the company's merged code as of May 2026, with engineers shipping 8x more code per day in Q2 2026 than in 2024, suggesting RSI dynamics are already influencing real development pipelines.
- Anthropic stated it would slow or pause frontier AI development if peer labs agreed to do the same, representing a notable shift toward conditional, coordinated restraint from a leading frontier lab.
- OpenAI's 'dreaming' memory system nearly doubled ChatGPT's factual recall (41.5% to 82.8%) and more than doubled preference-following (31.4% to 71.3%), suggesting background profiling significantly outperforms the previous fact-list approach.
- A coalition of top AI CEOs warned Congress that 'knowledge barriers which have historically prevented bad actors from obtaining biological weapons will meaningfully erode,' framing AI-enabled bioweapon design as an imminent regulatory gap rather than a distant risk.
- CloudFlare co-founder Matthew Prince revealed that bot traffic on the internet has already surpassed human traffic, a milestone he expected to arrive a full year later than it actually did, indicating AI-driven automation is outpacing even expert forecasts.
Topics
Transcript
Good morning, {{ first_name | AI enthusiasts }}. Recursive self-improvement used to sound like a far-off sci-fi phrase. Now, frontier labs say the signs are showing up internally faster than anyone expected. Anthropic's new report shows AI already accelerating the work that builds its own successors, and lays out the risks of systems that improve themselves… While also floating the jarring scenario of a coordinated development pause across the industry. Anthropic charts path to self-improving AI OpenAI’s memory overhaul lets ChatGPT ‘dream’ Stress test business ideas with Perplexity Rival AI labs unite behind bioweapons risks 4 new AI tools, community workflows, and more ANTHROPIC The Rundown: Anthropic just published "When AI builds itself," a report on recursive self-improving (RSI) systems…
Full transcript available for MurmurCast members
Sign Up to AccessMore from The Rundown AI
Runway's Solaris previews the no-code internet
Runway unveiled Solaris, an AI-powered interface that renders websites and apps as real-time video with no underlying code, while Imperial College researchers developed an AI model that detects heart disease from ECGs in under two seconds with superior accuracy to human doctors.
OpenAI cuts out SpaceX-owned Cursor
OpenAI is removing its models from Cursor coding editor by mid-November following SpaceX's acquisition of the platform, citing Elon Musk's history of contract violations as justification. The move escalates the ongoing feud between Sam Altman and Elon Musk while putting developers in the middle of a high-profile corporate conflict.
Every machine is about to speak Claude
Anthropic unveiled the Model Hardware Standard, enabling AI agents to control physical machines like microscopes and robot arms with minimal setup time. The newsletter also covers voice-controlled AI workflows gaining mainstream adoption, brain surgery assisted by live AI video analysis, and various AI industry developments from major tech companies.
Grok 4.6 storms the AI frontier
SpaceXAI's Grok 4.6 has emerged as a competitive frontier AI model, scoring 61 on benchmarks and undercutting rivals on price at 60% less cost, with Grok 4.7 teased for release in 3-4 weeks. The newsletter covers AI trends including practical guidance on using AI effectively, a personalized cancer vaccine startup for dogs built with AI tools, and updates on competing models and AI products.
Anthropic slips an invisible signature into Claude
Anthropic has announced the implementation of invisible watermarks in outputs from its Claude models to enhance transparency, aligning with the EU AI Act. The move has sparked mixed reactions, as stakeholders navigate the implications for open and private AI models.