Anthropic Flags >60% Chance AI Will Self-Train by 2028
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- Jack Clark predicted a greater than 60 % chance that an AI system will be able to autonomously train its successor by the end of 2028.
- Anthropic released a five‑page research agenda for the Anthropic Institute warning of a possible intelligence explosion and emphasizing recursive self‑improvement.
- The Anthropic Institute organized its agenda into four research buckets: economic diffusion, threats and resilience, AI systems in the wild, and AI‑driven R&D focused on recursive self‑improvement.
- Anthropic will publish monthly reports on AI’s impact on jobs and productivity as an early‑warning signal for significant change.
- Anthropic is exploring coordinated industrial‑policy levers, such as sector‑by‑sector throttling of AI diffusion, potentially in partnership with governments.
- The institute proposes a tabletop “fire drill” exercise to test decision‑making for an intelligence explosion, drawing parallels to Cold‑War hotlines for crisis communication.
- Anthropic commits to publicly sharing detailed information on how new AI tools have accelerated its own R&D and the implications of recursive self‑improvement.
Why it matters: Anthropic’s public warning and detailed agenda give regulators and industry a concrete signal of imminent AI self‑improvement, prompting potential policy coordination and safety measures, while competitors may face pressure to adopt similar transparency and risk‑mitigation practices, reshaping the competitive landscape.
Ask SkimNews
