AI archive
Every AI story, newest first. 91 in this archive.
Sep 20, 2026
Enforcement Gap Paper Shows Agent Audits Fail When Controllers Ignore Them
An arXiv paper argues Reflexion style agents detect unsafe plan steps yet lack a pathway to block them, and a small enforcement hook cuts attack success more than fourfold.
Sep 19, 2026
AgentAudit Scores Full Agent Traces and Surfaces Unsafe Compliance
A new open framework evaluates planning, tools, memory, and security across entire agent runs. Trust scores diverge sharply from task success on adversarial work.
Sep 18, 2026
Agent Harness Plans Beat Sham Guidance by Seven Points on Retail Tasks
An arXiv study of stateful LLM agents on tau bench finds task specific plans raise oracle verified success by 7.17 percentage points, while a read only terminal verifier blocks most invalid completions at under one cent per episode.
Sep 17, 2026
HazardAuditor Audits Full Agent Trajectories Not Static Prompts
A 14 September 2026 arXiv paper introduces HazardAuditor, an execution grounded guard for computer use agents that improves safety accuracy by up to 16.5 points over prior guards using GuardPO training.
Sep 16, 2026
Blindspot Benchmark Treats Agent Safety as a Trajectory Property
A 14 September arXiv paper introduces Blindspot, scoring long horizon tool using agents on safe completion, correct refusal, and over refusal across more than 2,500 trajectories, exposing calibration gaps proprietary models miss in single turn tests.
Sep 15, 2026
Tau Tau Bench Shows Coding Agents Fail Real Client Style Agent Builds
A 4 September 2026 arXiv paper finds the best coding agent configuration passes only 23.9 percent of realistic customer service agent builds, against an 82.2 percent expert ceiling.
Sep 14, 2026
LLM Information Geometry Is Shared Across Architectures and Can Steer Safely
A 10 September 2026 arXiv paper reports that large language models share Fisher Rao output geometry across transformer, state space, and recurrent architectures, enabling minimum disturbance control for steering and fine tuning.
Sep 13, 2026
T1 Terminal Agent RL Hits 64 Percent on Terminal Bench 2.1
Researchers introduced T1, a 122 billion parameter mixture of experts model trained with reinforcement learning in real cloud shells, reaching 64.0 percent on Terminal Bench 2.1.
Sep 12, 2026
AlphaGenome Atlas Precomputes Nine Billion Variant Effects for Genomics Teams
Google DeepMind released AlphaGenome Atlas on 8 September 2026 with precomputed predictions for every single nucleotide change in the human genome, plus an AVI score to rank variants across coding and noncoding DNA.
Sep 11, 2026
Looped Flows Lift Recurrent Reasoning to 58.8 Percent on ARC AGI 1
A 10 September arXiv paper trains looped recurrent models with local denoising objectives and reports state of the art looped model scores on ARC AGI 1 and ARC AGI 2.
Sep 10, 2026
uFlowCSP Cuts Crystal Structure Search From Thousands of Steps to Five
Researchers report uFlowCSP, a MeanFlow crystal structure predictor that matches prior generators in fewer network evaluations, shrinking wall clock time for high throughput materials screening.
Sep 9, 2026
NeoHorse 1 Closes an Evaluation Selection Update Loop for Agent Native Models
A 8 September arXiv paper introduces NeoHorse 1, an agent native model family that records routing decisions and converts them into validated training data, raising macro average scores on eleven agent benchmarks.
Sep 8, 2026
Harbor Index Condenses 54 Agent Benchmarks Into 82 High Signal Tasks
Harbor Index 1.0 selects 82 difficult tasks from more than 6,600 candidates across 54 benchmarks, capping the best model harness pass rate at 28 percent and cutting eval cost to a few hundred dollars per run.
Sep 7, 2026
EVOHARNESSBENCH Shows Tool Growth Can Erase Agent Competence
EVOHARNESSBENCH tests agents when tools, skills, and subagents change over time, finding harness expansion can erase prior competence even as new capabilities arrive.
Sep 6, 2026
CivBench Exposes Long Horizon Agent Planning Gaps in Civilization VI
CivBench tests language agents across 300 plus Civilization VI turns with 76 MCP tools, showing aggregate scores hide plan execution failures.
Sep 5, 2026
WorldBench Shows Frontier Agents Fail Half of Culturally Grounded Tasks
A 1 September 2026 arXiv paper reports frontier LLM agents reach only 49.2 percent Constrained Task Success on 1,600 multilingual, persona grounded workflows. The gap between pass rate and environment preservation is the decision signal for product teams shipping global agents.
Sep 4, 2026
KC Bench Shows Agent Knowledge Conflicts Still Break Across Nine Frontier Models
Researchers posted KC Bench on 3 September 2026 with 238 interactive tasks that test how LLM agents reconcile conflicting instructions, memory, and tool observations. No evaluated model handled factual correction, identity checks, and temporal conflicts reliably.
Sep 3, 2026
HarnessDev Tests Whether Models Can Build Their Own Agent Scaffolding
HarnessDev shifts agent evaluation from task outputs to whether models can build and evolve their own execution harnesses. Platform teams should treat orchestration as part of the capability stack, not invisible plumbing.
Sep 2, 2026
CoBRA Teaches Agents When Retrieval Beats Answering From Memory Alone
A 1 September arXiv paper introduces CoBRA, which trains agents to call retrieval tools only when counterfactual reward margins justify the cost. Applied teams can treat routing as a first class optimization target.
Sep 1, 2026
AdaptiveFlow Cuts Billion Molecule Screen Cost for Drug Discovery Teams
A Nature Biotechnology paper introduces AdaptiveFlow, an open platform that screens billions of molecules with far lower cloud cost and demonstrated hits on PARP1 and FSP1 oncology targets.