Daily Signals: Agents That Certify Their Own Exploits

A dated Agentic RL Daily snapshot with 6 verified primary-source signals across papers, official releases, deployment evidence, and safety or alignment findings.

EDITOR'S VIEW

Three judgments

The edition is based only on primary sources or official project releases.
Older signals remain visible as continuing observations, not as rewritten news.
Headline claims are constrained by the evidence included in this dated snapshot.

FULL EDITION

All signals in this edition

Archived / 2026-08-02

2026.07.30PapersPapers

Agents That Certify Their Own Exploits

An agent playing a Nash-equilibrium strategy in a two-player zero-sum imperfect-information game secures the game value but forfeits the additional value offered by a flawed opponent.

Training AlgorithmsAgent CapabilitiesData Loops安全与对齐

arXiv ->

2026.07.30PapersPapers

ORCA-bench

Large language models can write, patch, and search code, but oncall root cause analysis (RCA) demands something different.

Agent CapabilitiesSystems Engineering评估体系

arXiv ->

2026.07.30PapersPapers

OSReward

Computer-using agents (CUAs) are advancing rapidly across the digital world.

Reward and Credit AssignmentAgent CapabilitiesData Loops安全与对齐评估体系

arXiv ->

2026.07.30PapersPapers

Paying for Honesty Without Knowing the Truth

LLM agents increasingly act as autonomous merchants that write their own product listings, and under competitive pressure, they fabricate attributes to win sales.

Training AlgorithmsAgent Capabilities记忆与自进化评估体系

arXiv ->

2026.07.30PapersPapers

Tycho

ARC-AGI-3 turns abstraction into an interactive problem of skill acquisition.

Agent CapabilitiesData Loops评估体系

arXiv ->

2026.06.18Open SourceOpen Source

alibaba/ROLL v0.3.0

ROLL发布了v0.3.0版本，新增Video RLVR、AgentRunner 2.0、MTP训练、Router Replay、Multi-Teacher OPD等重要特性。

Training AlgorithmsReward and Credit AssignmentAgent CapabilitiesData LoopsSystems Engineering

GitHub ->