recursive-self-improvement
Weco reports a research agent that improved its own harness, without proving recursive takeoff News
Weco says its AIDE² outer loop produced seven accepted harness rewrites over eight days and beat a human-engineered baseline under fixed budgets, while its test for compounding self-improvement remained inconclusive.
Dream-RSI improves a coding agent’s search policy without changing its weights News
Dream-RSI uses replayed discovery histories to improve the exploration policy around a fixed coding agent, reporting large search-efficiency gains while stopping well short of autonomous model-weight self-modification.
Dario Amodei asks AI labs to pace the frontier, not pause it. Altman agreed to match one step. News
Anthropic CEO Dario Amodei published an essay on 12 September 2026 calling for AI labs to slow the rate of capability gains through outside evaluators, government rules and international deals; the only step Anthropic commits to on its own is embedding third-party evaluators, which OpenAI's Sam Altman said OpenAI will also do.
OpenAI now says it will 'slow or stop' at an unacceptable risk. It has not said where that line is. News
OpenAI wrote on 9 September 2026 that it will slow or stop developing systems it cannot sufficiently safeguard and will not pursue fully autonomous recursive self-improvement until it can be done safely, but set no threshold or date; the same week, more than 70 UK lawmakers urged a superintelligence ban and President Trump said he had no extinction concerns.
OpenAI says it is prioritising RSI and alignment over making models better at math research News
OpenAI says it could push math-research capability harder but is prioritising recursive self-improvement and automated alignment research instead, without publishing a formal slowdown trigger.
OpenAI reports 3.1 agent-workdays for every human research workday News
OpenAI says internal research agents generated 3.1 normalized eight-hour workdays per human workday by mid-August, a preliminary throughput metric rather than an independently audited replacement claim.
A self-improving coding agent that compares notes with a rival lineage News
Most self-improving coding agents rewrite themselves after a single failure, throwing away the archive of everything they have already tried; a new method adds two edit operations that use multiple trajectories and a competing agent's evidence instead.
Models that rewrite their own harness gain 16 points and flunk office work News
Evo-Bench holds the model and budget fixed and measures only what improving its own scaffolding is worth, finding gains of up to 16.6 points that come close to human-engineered baselines everywhere except tasks with prescribed workflows.
Greenblatt puts his median at five years of progress in one News
Redwood Research's Ryan Greenblatt told Dwarkesh Patel that once AI matches top human AI researchers the feedback loop could compress four or five years of progress into a single year, and that what models still lack is not deep insight but hands-on experimental taste.
An agent edited its own runtime for 161 days News
Ouroboros is a coding agent whose tools, prompts and core implementation change through reviewed commits that become the runtime for its next task, and its longest public deployment ran live for 161 days across seven surfaces.
An open 35B model trained to evolve its own machine-learning code nearly doubled its base model's medal rate News
Frontis-MA1, released with full weights and stack, raises its base model's medal average on a machine-learning engineering benchmark from 39.4% to 60.6%, and to 71.2% with a stronger search - all within a 12-hour budget on a single consumer GPU capped at 12GB.
DeepMind Sketches Four Roads From Human-Level AI to Superintelligence News
A new report from senior DeepMind researchers lays out four ways AI could push past human-level ability -- and argues the leap is more likely to be a steady climb than a single dramatic jump.
The AI That Now Writes Most of Its Maker's Code News
Anthropic says more than 80 percent of the code it ships is now written by its own model, Claude, and the more interesting numbers are about judgment.
Recursive self-improvement: when AI starts building AI Lesson
The idea that an AI good enough at AI research could improve itself, and the improved version could improve itself again, faster each round. Here's what it actually means, why a major lab now says we're getting close, and why "close" is not the same as "here."