August 11, 2026 Research Brief

Evaluation goes live.

Today’s strongest papers replace static benchmarks with prospective or longitudinal tests, while safer agents rely on explicit state, hard gates, and domain-grounded verification rather than orchestration alone.

Why it matters Prospective and longitudinal evaluation replaces static benchmarks

Static benchmarks increasingly miss contamination, memorization, and long-horizon failure modes. Live and longitudinal protocols reveal whether models can update on new evidence, sustain performance over time, and improve from experience rather than recall.

What changed Explicit state, memory, and verification outperform free-form agenting

Many agent failures come from stale context, invalid intermediate outputs, or inability to reuse prior corrections. Papers in this cluster show that making state and checks explicit is often a bigger win than adding more agent roles.

What to watch Safety is moving toward auditable gating and action admissibility

In high-stakes settings, correctness of diagnosis or intent is insufficient; what matters is whether the proposed action is safe, authorized, and verifiable. Several papers operationalize this with hard gates, digital twins, or cryptographic controls.

Read the brief 中文版本

Recent Briefs

Each issue should tell you why the day is worth reopening, not only when it was published.

Evaluation goes live. Today’s strongest papers replace static benchmarks with prospective or longitudinal tests, while safer agents rely on explicit state, hard gates, and domain-grounded verification rather than orchestration alone.
EN 中文
Agent control planes matter. Today’s strongest papers shift agent progress from bigger models to runtime verification, interface design, and evidence-bounded evaluation, especially in high-stakes or long-horizon settings.
EN 中文
Agent control gets audited. Today’s strongest papers show that agent progress depends less on raw capability than on auditable control: better evaluation, explicit execution checks, and new defenses for stateful memory failures.
EN 中文
Agent safety gets structural. Today’s strongest papers shift agent reliability from benchmark scores and output filters toward deployment-aware evaluation, pre-action controls, and audits of whether models actually use the evidence they claim.
EN 中文
Safety wrappers look brittle. Today’s papers show safety failures moving to adaptation and interfaces: fine-tuning can undo alignment, agent observations are easy to steer, and common monitoring signals break under dependence or weak calibration.
EN 中文
Agent safety moves downstream. Today’s strongest papers show frontier risk now lives in memory, tools, and serving infrastructure, while the best defenses add mechanism-aware auditing, privacy accounting, and stronger security evaluation.
EN 中文
Agent safety turns stateful. Today’s strongest papers show agent failures emerge across trajectories, memories, and sessions, while lightweight runtime checks and process-aware benchmarks expose and sometimes repair them.
EN 中文
Agent control moves outward. Today’s strongest papers shift reliability work from model weights to outer-loop control: deployment-readiness checks, selective escalation, and benchmark audits expose where agent capability still fails in practice.
EN 中文
Verification wraps the agent. Today’s strongest papers argue that reliable agents come from verifiable runtime structure, cleaner audits, and orchestration around frozen models rather than raw model upgrades alone.
EN 中文
Evaluation gets more surgical. Today’s strongest papers split capability into auditable parts: verification-first agents improve reliability, decomposed benchmarks expose hidden gaps, and open-ended autonomy still lags structured execution.
EN 中文
Agent reliability gets grounded. Today’s strongest papers argue that agent progress depends less on prompt cleverness and more on trustworthy evaluators, structured interfaces, and safety mechanisms that survive real attack channels.
EN 中文
Safety evaluation goes longitudinal. Today’s strongest papers replace single-turn safety claims with trajectory, memory, and lifecycle evaluation while exposing how brittle current defenses remain to surface-form attacks.
EN 中文
Agent safety moves runtime. Today’s strongest papers shift agent reliability from prompt obedience to runtime control, while new benchmarks show long policies and context alone still fail under realistic tool use.
EN 中文
Agent reliability gets stateful. Today’s strongest papers argue that safe agent performance depends on explicit state, provenance, and cost-aware evaluation, not just successful endpoints or higher benchmark scores.
EN 中文
Agent state gets transactional. The strongest papers treat agent safety as a runtime systems problem: permissions, memory commits, and trace-level evaluation now matter more than one-shot jailbreak scores.
EN 中文
Reliability gets more structured. Across agent, medical, and security papers, the strongest results came from auditable scaffolds: deterministic policies, evidence-use audits, and governed memory or verification loops.
EN 中文
Agent safety moves upstream. Today’s papers shift attention from model outputs to the systems around them: runtimes, retrieval, memory, and tool workflows now drive both capability gains and the sharpest failures.
EN 中文
Agent trust gets audited. Today’s strongest papers replace aggregate benchmark wins with deployment-shaped audits of provenance, hidden bugs, and research reliability, while stage-wise diagnostics expose where agent and safety failures actually arise.
EN 中文
Agent safety moves deeper. Today’s strongest papers push reliability from answer scoring to trajectory verification, predictive guards, and infrastructure threat models, showing that capable agents fail through state, tools, and serving layers as much as outputs.
EN 中文
Agent security goes systemic. Today’s strongest papers show agent risk and reliability moving from single prompts to workflow-level evaluation, provenance-aware defenses, and more selective safety controls that preserve utility.
EN 中文
Agent evaluation gets harder. Today’s papers push AI claims into executable, evidence-grounded, and budget-matched settings, while several control and post-training methods look weaker once realism replaces headline demos.
EN 中文
Agent oversight gets audited. Today’s papers argue that agent reliability depends less on final scores and more on traceable workflows, executable checks, and skepticism toward judge-based optimization signals.
EN 中文
Evaluation stops being optional. Today’s strongest papers show that deployment-shaped evaluation, hidden-instability audits, and verifier design now matter as much as model capability for agents, safety, and high-stakes use.
EN 中文
Agent safety gets structural. Today’s strongest papers argue that reliable agents need explicit control planes, evidence-bound execution, and new evaluations for persistent-context attacks, multilingual bias, and dynamic tool environments.
EN 中文
Agent oversight gets concrete. Today’s strongest papers turn agent safety and reliability into runtime systems problems: denser post-training signals, fairer evaluations, and deployable permission, routing, and attestation layers.
EN 中文
Agent control gets explicit. Today’s strongest papers shift agent reliability from end-task scores to explicit state, hard runtime constraints, and audited evaluation, with security work targeting workflow-specific attack surfaces rather than generic jailbreaks.
EN 中文
Agent safety gets operational. Today’s strongest papers replace abstract agent wins with runtime security, verifier audits, and structured control, showing that deployment failures often come from weak monitors, leaky judges, and flat orchestration.
EN 中文
Agent evaluation grows teeth. Today’s strongest papers replace surface benchmark wins with auditable, closed-loop, and leakage-aware evaluation, while verification and explicit control layers become core reliability tools.
EN 中文
Control moves outside prompts. Today’s papers show more reliable AI systems coming from explicit contracts, bounded actions, process-aware evaluation, and structural defenses around routing, tools, and validators.
EN 中文
Safety moves into architecture. Today’s strongest papers favor structural controls and realistic evaluation over prompt-only fixes, while targeted inference-time interventions and token-level training methods push reliability into deployment settings.
EN 中文
Agent safety moves inward. Today’s strongest papers shift reliability from bigger models to controllable agent scaffolds, while showing that visible reasoning and weak evaluation pipelines can become liabilities.
EN 中文
Agent safety moves upstream. Today’s strongest papers show agent reliability depends more on orchestration, interfaces, and deployment-grounded evaluation than on model weights alone, with silent multi-agent and tool-use failures driving the shift.
EN 中文
Agent safety moves downstream. Today’s strongest papers shift reliability from model outputs to execution boundaries, grounded verification, and realistic agent evaluation that exposes reward hacking and workflow failures.
EN 中文
Agent safety gets structural. Today’s strongest papers push agent reliability below the prompt layer: architectural trust boundaries, process-level audits, and verifier-backed training all target failures in memory, retrieval, and long-horizon action.
EN 中文
Agent evaluation gets harsher. Today’s strongest papers show that long-horizon agents, judges, and safety claims look weaker under realistic environments, deployment settings, and process-aware verification.
EN 中文
Safety moves to operations. Today’s strongest papers shift safety from average-case outputs to operating conditions: auditable deferral, judge reliability, multi-turn control, and ecosystem-level attack surfaces now dominate the research signal.
EN 中文
Agent safety moves runtime. Today’s strongest papers argue that reliable agents need execution-time authorization, memory integrity, and evaluation methods that expose security–fidelity tradeoffs and hidden proxy failures.
EN 中文
Agent safety gets stateful. Today’s strongest papers move agent safety beyond refusals toward evidence-grounded verification, governed state, and hybrid static-plus-runtime defenses as attacks exploit persistence, composition, and multilingual gaps.
EN 中文
Agent safety moves runtime. Today’s papers shift AI safety from prompt hardening to runtime control and behavioral audits, as realistic agent attacks and broken evaluation proxies expose weaknesses in deployed workflows.
EN 中文
Agent safety gets structured. Today’s strongest papers replace coarse end-to-end trust with gated execution, intermediate supervision, and production-like evaluation, while alignment work shifts toward controllable mechanisms instead of generic safety tuning.
EN 中文
Agent safety gets systemic. The strongest July 1 papers stop treating agent safety as prompt hygiene, pushing toward system-level benchmarks, runtime governance, and verification that surfaces hidden tradeoffs.
EN 中文
Agent reality gets harder. Realistic long-horizon benchmarks, dialogue-aware policy checks, and deployment-time defenses all point to the same fact: current agents stay brittle once hidden state, compression, and messy workflow constraints matter.
EN 中文
Agent safety moves to runtime. The day’s strongest abstracts argue that trustworthy agents come from explicit execution controls, least-privilege design, and external stop conditions—not from better refusals alone.
EN 中文
Agent safety moves runtime. The strongest papers treat agent safety as a runtime systems problem: they audit full action traces, expose real-world misuse on phones and terminals, and add lightweight checks before execution.
EN 中文
Agent safety goes structural. The best June 27 papers move agent safety out of prompts and into control planes, temporal memory rules, and tougher evaluations that test process, freshness, and adaptive attack resilience.
EN 中文
Agent safety gets operational. Today’s strongest papers push safety into runtime structure: external controls, unreliable-tool benchmarks, and repair-focused evaluations reveal how far agents still are from dependable execution.
EN 中文
Agent control gets explicit. Today’s strongest papers replace prompt-only agent design with governed memory, formal verification, and system-level security evaluation, while more realistic benchmarks expose where long-horizon agents still break.
EN 中文
Agent safety gets operational. Today’s strongest papers replace answer-only evaluation and static guardrails with verifiable agent checks, runtime authorization, and privacy-aware controls built for real enterprise environments.
EN 中文
Evaluation becomes infrastructure. Today’s papers argue that progress claims increasingly hinge on benchmark repair, process-level verification, and deployment-interface audits, while agent gains come more from structured scaffolds than larger models alone.
EN 中文
Evaluation goes process-first. Today’s strongest papers replace outcome-only scoring with verifiable process checks, while agent training and inference methods add finer-grained feedback for safer, more reliable systems.
EN 中文
Evaluation turns lifecycle-aware. Today’s papers push AI assessment into realistic workflows while exposing brittle safety, grounding, and training assumptions that cleaner benchmarks often miss.
EN 中文
Agent safety gets operational. Today’s strongest papers replace static agent scores with deployment-predictive evaluation and runtime control, while exposing safety failures rooted in tool privilege, orchestration, and execution boundaries.
EN 中文
Agent safety moves structural. Today’s strongest papers argue that prompt-only defenses are brittle: safer agents come from typed interfaces, privacy-aware benchmarks, and finer-grained training signals that constrain what models can access or emit.
EN 中文
Agent evaluation grows teeth. Today’s papers push agent research away from single-score demos toward process-aware evaluation, transactional runtimes, and realistic security tests that expose cross-step failures.
EN 中文
Agent security moves down-stack. Today’s strongest papers show agent failures increasingly come from infrastructure, process, and reward channels, pushing evaluation and defenses beyond prompt-level alignment alone.
EN 中文
Auditable agents take over. Today’s strongest papers favor process-aware verification, black-box auditing, and protocol-level agent design over monolithic accuracy claims, while multiple papers warn that current evaluation practice is too brittle to trust at face value.
EN 中文
Agent reliability gets audited. Today’s strongest papers favor evidence-bearing, executable agent workflows over answer-only performance, while puncturing default multi-agent assumptions and exposing new modular security risks.
EN 中文
Evaluation gets operational. Today’s papers push AI assessment and safety toward deployment-shaped tests, explicit control layers, and operational security for agents, RAG, and long-form oversight.
EN 中文
Agent safety moves upstream. Today’s papers argue that reliable agents depend less on bigger models than on containment, memory control, harder evaluation, and failure-targeted training loops.
EN 中文
Agent safety moves runtime. Today’s strongest papers argue that safer AI depends less on static alignment alone and more on process-aware evaluation, runtime controls, and finer-grained supervision for agents.
EN 中文
Agent security turns stateful. Today’s strongest papers show agent risk moving into memory, execution state, and post-training drift, while executable benchmarks and internal monitors expose failures that output-only checks miss.
EN 中文
Agent safety gets systemic. Today’s strongest papers argue that reliable agents need infrastructure-level controls, calibrated oversight, and harder long-horizon evaluation because weak judges, brittle verifiers, and prompt-only defenses fail predictably.
EN 中文
Reliability shifts to control. Today’s strongest papers treat reliability as a controllable systems property: richer evaluation, explicit verification layers, and security defenses that break attacker feedback loops rather than only filtering outputs.
EN 中文
Agent control gets concrete. Today’s strongest papers push agents toward governed memory, consequence-aware control, and more realistic evaluation, while exposing new attack surfaces in steering, context, and workflow artifacts.
EN 中文
Agent evaluation turns adversarial. Today’s strongest papers show that agent progress depends less on raw task wins and more on cheating-resistant evaluation, runtime defenses, and structured process signals for tool use and evidence.
EN 中文
Agent safety moves outward. Today’s strongest papers argue that agent safety now lives in interfaces and workflows: tool surfaces, memory gates, offline evaluation, and human oversight all expose failures hidden by clean benchmarks.
EN 中文
Agent safety turns stateful. Today’s strongest papers show agent risk and evaluation moving from single prompts and final answers toward persistent state, process tracing, and structured control surfaces.
EN 中文
Agent safety moves runtime. Today’s strongest papers shift AI safety from model-only alignment to runtime governance, realistic auditing, and trajectory-aware defenses as agent attack surfaces widen across the lifecycle.
EN 中文
Agent safety moves runtime. Today’s strongest papers argue that agent safety is now a systems problem: execution-boundary controls, process-aware evaluation, and supply-chain defenses matter more than prompt-only safeguards.
EN 中文
Agent control gets explicit. Today’s strongest papers replace monolithic agents with governed pipelines, adaptive context handling, and harsher evaluation that rewards traceability, calibration, and deployable safeguards over raw scores.
EN 中文
Agent reliability gets operational. Today’s papers push agents and safety systems toward deployment reality: process-aware evaluation, verifier-first scaffolds, and localized multimodal safety tests expose failures static benchmarks miss.
EN 中文
Agent benchmarks meet reality. Today’s strongest papers show agent capability claims are highly scaffold-dependent, while security and reliability increasingly hinge on pre-execution controls at routing, retrieval, and tool boundaries.
EN 中文
Agent safety moves runtime. Today’s strongest papers shift safety from end-score evaluation to runtime auditing and enforcement, while showing retrieval, memory, and judging pipelines create new structural failure modes.
EN 中文
Safety moves into systems. Today’s strongest papers show AI safety failures increasingly emerge from state, tools, memory, and evaluation design, pushing defenses toward structural controls and process-aware diagnostics.
EN 中文
Agent safety moves inline. Today’s strongest papers argue that agent safety now depends on runtime control, provenance, and long-horizon evaluation, because models often detect risk without changing unsafe behavior.
EN 中文
Agent safety turns runtime. Today’s strongest papers argue that deployment-grade agent safety comes from runtime control, long-horizon evaluation, and structure-aware training rather than prompt filters or static benchmarks alone.
EN 中文
Agent safety moves runtime. Today’s strongest papers argue that agent security and reliability depend less on detecting bad inputs than on controlling provenance, authority, and action at execution time.
EN 中文
Agent reliability gets structured. Today’s strongest papers improve agents and high-stakes AI systems by adding explicit control, state tracking, and evidence checks, while new benchmarks and attacks expose hidden deployment failures.
EN 中文
Evaluation turns adaptive. Today’s strongest papers push AI evaluation and control beyond static scores toward adaptive audits, explicit intermediate state, and deployment-minded hardening for agents, retrieval, and model supply chains.
EN 中文
Agent safety gets stateful. Today’s strongest papers show agent reliability now depends less on bigger models than on realistic security evaluation, runtime scaffolds, and explicit control of state, logs, and interfaces.
EN 中文
Agent safety moves runtime. Today’s strongest papers shift safety from prompt-level behavior to runtime audits, long-horizon reward-hacking evaluation, and system-level controls around tools, deployment, and optimization.
EN 中文
Evaluation gets executable. Today’s strongest papers replace heuristic scores with verifiable environments, uncertainty-aware auditing, and system-level safeguards, while new security results show agent risk is spreading across retrieval, multimodality, and reasoning workflows.
EN 中文
Agent safety shifts outward. Today’s papers argue that reliable AI depends less on bigger models than on external verification, auditable control layers, and broader threat models that include hidden attack channels and workflow failures.
EN 中文
Agent evaluation gets harsher. Today’s papers show a shift from static benchmark wins to adaptive attacks, process-aware reliability metrics, and realistic tool environments that expose large autonomy and safety gaps.
EN 中文
Agent safety moves downstream. Today’s strongest papers shift safety from output filtering to runtime structure, trace-level auditing, and post-deployment checks, with quantization and memory emerging as major failure surfaces.
EN 中文
Agent safety moves outward. Today’s strongest papers argue that reliable agents need external control layers, process-aware evaluation, and multi-turn threat models because prompt-level alignment breaks under history, peers, and persistent state.
EN 中文
Agent safety turns operational. Today’s strongest papers push safety from model claims to runtime evidence: real-environment jailbreak tests, formal guardrail guarantees, and benchmark audits that expose unsupported scores.
EN 中文
AI reliability gets real. Today’s strongest papers move beyond benchmark wins toward deployment evidence: harsher evaluation, validated agent workflows, and targeted robustness.
EN 中文