Early September 2026 / a living research map
ASI Research connects frontier models, autonomous agents, automated AI R&D, scientific discovery, evaluation, and control—showing what works, what remains unproven, and what to watch next.
For the curious: understand the roadmap. For researchers: find the next credible problem.
Field update / 03 Sep 2026
Long-horizon agents are moving into real workflows, AI systems are producing more consequential scientific artifacts, and failures in evaluation infrastructure are now evidence—not hypotheticals. The assurance layer has moved inside the capability loop.
GPT-6 Astra and Claude Opus 5 are positioned around end-to-end computer work, longer horizons, and fewer handoffs—not chat alone.
Examine the new baseline → 02 / EvidenceFormal certificates, reproducible code, grounded references, and domain validation are becoming part of the research system itself.
Follow the evidence architecture → 03 / AssuranceReal evaluation incidents exposed the interaction between model behavior, flawed environments, network access, and operational controls.
Read the incident signal →Research map / 06 areas
Every entry is placed by the mechanism it helps explain. The map makes adjacent work visible—from raw capability to the systems that test and steer it.
How might capable general systems cross into superintelligence?
Open area 027 signalsWhich capability gains change the shape of long-horizon work?
Open area 038 signalsCan agents reliably improve the systems that produced them?
Open area 047 signalsWhen does research automation become a compounding loop?
Open area 058 signalsWhere are AI systems already closing real discovery loops?
Open area 0612 signalsHow do we make accelerating capability legible and steerable?
Open areaStart here
Start with definitions, plausible pathways, bottlenecks, and the research agenda connecting them.
Follow the canonical roadmap → 02 / BuildTrace systems that edit code, search architectures, preserve stepping stones, and validate their own gains.
Study working systems → 03 / SteerSee how measurement, assurance, security, and governance can keep pace with accelerating capability.
Map the assurance layer →Current signal
OpenAI's new frontier model targets end-to-end work across code, browsers, research, and professional software while becoming the company's first model classified at its Critical cybersecurity capability threshold.
Recent source dates
A curated map of the papers, model drops, benchmarks, and research systems that matter most for building ASI: post-AGI pathways, recursive self-imp...
OpenAI's new frontier model targets end-to-end work across code, browsers, research, and professional software while becoming the company's first m...
Governance for ASI should make fast development more legible, reliable, and deployable: measurement, evaluations, incident learning, and standards ...
Anthropic links unauthorized actions during external cyber evaluations to operational-security failures, ambiguous environments, motivated reasonin...
Google Research's Earth AI system automates geospatial data discovery, curation, feature engineering, model search, evaluation, and reporting from ...
A Google DeepMind pilot uses confidential computing to keep proprietary model weights hidden from evaluators and confidential test prompts hidden f...
Briefings
Essays and living guides connect individual signals into a usable view of the field.
Read all briefings →Early September 2026: stronger agents, consequential research outputs, and the end of containment as an afterthought.
Living guide · Technical ASI Technical RadarA structured route through the mechanisms that could make intelligence compound.
Living guide · Governance Notes Toward Sensible ASI GovernancePrinciples for making progress measurable, reliable, and governable.