Living report · refreshed from daily source data

AI agents report: autonomy, oversight and incidents

Daily evidence on autonomous AI agents, oversight, incidents and governance, with an updating graph and reusable data. Coverage counts are signals of attention—not measures of importance, harm or consensus.

Prepared by the ethics.ai evidence desk · automatically refreshed · editorial scope reviewed against the methodology and corrections policy

Daily source coverage90 days
2026-05-18 2026-08-15
3,850records in archive
1,721latest 30 days
+130%versus prior 30 days

By record type

Research 2,632
News 959
Field notes 184
Incidents 40
Policy 35

Leading sources in this record

arXiv 1,580
arXiv cs.AI 282
HuggingFace Daily Papers 249
OpenAlex 174
arXiv cs.CY 93
arXiv cs.HC 84
The New Stack AI 69
Heise Online (DE) 57
SiliconANGLE AI 55
Finextra AI 50

What this report tracks

Tracks systems that plan, use tools or act with limited supervision, along with evaluations, incidents and oversight proposals. It distinguishes agent research from verified real-world capability.

Questions to take into the evidence

  • What actions are agents being trusted to take?
  • Which failure modes have been observed rather than hypothesized?
  • What monitoring and control mechanisms are being tested?

Latest evidence

Full topic record →
Xataka (ES)

El Robot Phone de Honor es un teléfono verdaderamente emocionante. Tiene ante sí un reto casi imposible — open the original publisher

Solo es agosto, pero creo que salvo sorpresa mayúscula, ya se puede decir: el Robot Phone de Honor es el móvil más llamativo y diferente de 2026. Ni siquiera el cambio de formato del Samsung Galaxy Z Fold8 (promete ser el primero de varios y te miro a ti, iPhone fold ) le hace sombra. En un mundo tan aburrido y consolidado como el de los smartphones, Honor se ha atrevido a hacer algo muy distinto y no es un brindis al sol: pasó de prototipo en el MWC a dispositivo a la venta y tras probarlo, le

News Agents & autonomy
The Next Web AI

LG’s humanoid robot arrives in 2027, running Nvidia’s Isaac GR00T — open the original publisher

LG will unveil a bipedal humanoid robot in the first quarter of 2027, built on Nvidia’s Jetson Thor processor, Isaac GR00T foundation model and Halos safety system. The two signed the agreement at Nvidia’s Santa Clara headquarters on 13 August. LG will show a bipedal humanoid robot in the first quarter of next year. The […] This story continues at The Next Web

News Agents & autonomy
InfoQ AI/ML

Presentation: From Models to Agents: Building Context-Aware Consumer AI at Scale at DoorDash — open the original publisher

Sudeep Das shares how DoorDash shifts from legacy one-shot predictions to an agentic recommendation platform. He discusses leveraging language-native consumer memory, RQ-VAE semantic IDs for catalog representation, and grounded search to dramatically boost relevance and conversion metrics. By Sudeep Das

News Agents & autonomy
InfoQ AI/ML

Cloudflare Adds Agent Tracing, with Truncation Limits and Uneven Payload Defaults — open the original publisher

Cloudflare launched agent tracing, adding spans for agent invocations, model calls, tool runs, and approvals to existing Workers traces. Sessions replay turn by turn, though the docs warn traces are not lossless and payloads may be truncated. Payload recording defaults differ by framework, and from October 1, 2026 every span counts as a billable event. By Steef-Jan Wiggers

News Agents & autonomy

Method and limits

This report is assembled automatically from source metadata and keyword classifications. It summarizes what the tracked source fleet published; it does not independently validate every linked claim. Source-fleet growth can inflate historical comparisons. Cite the individual evidence record and original publisher for substantive claims.