Impact

Reliability Research

Issues affecting model consistency and dependability

203 matching entries

Reliability research results

203 entries

Boilerplate safety tuning causes false refusals on benign requests

Published Sep 4, 2026Analyzed Sep 9, 2026Source: arXiv

Controlled safety-tuning experiments link boilerplate refusal statements to unnecessary refusals of benign requests. Request-specific rationales improve benign compliance, with benchmark-dependent safety tradeoffs.

Refuse without Refusal: A Structural Analysis of Safety-Tuning Responses for Reducing False Refusals in Language Models

Evaluated models: Llama 3.1 8B, Mistral 7B v0.3, Gemma 2 9B +9 more

Deleted agent memory persists in live execution state

Published Sep 4, 2026Analyzed Sep 9, 2026Source: arXiv

Deleting a memory record can leave its information in an agent's summaries, pending plans and KV cache. The paper evaluates revocation across this derived execution state.

Forgetting Without Restarting: Execution-State Unlearning for Stateful LLM Agents

Evaluated models: Llama 3.1 8B Instruct, Qwen 2.5 7B Instruct, Mistral 7B Instruct v0.3

Selective refusal gaps in visual question answering

Published Sep 4, 2026Analyzed Sep 9, 2026Source: arXiv

KoNA measures whether vision-language models answer valid image questions while refusing unsafe components or correcting unsupported premises. Its 9,300 question-answer pairs include mixed and fully answerable controls.

Knowing What Not to Answer: Selective Non-Compliance in Vision-Language Models

Evaluated models: InternVL3 2B Instruct, InternVL3-78B-Instruct, Qwen 2.5 VL 3B Instruct +5 more

PII leakage in document-image redaction benchmarks

Published Sep 2, 2026Analyzed Sep 9, 2026Source: arXiv

Document-image PII detection can miss identifiers despite improving average localization scores. LeakageBench evaluates 500 pages with 11,954 annotations.

LeakageBench: Document-Level Leakage Risk for Redacting Personally Identifiable Information in Document Images

Evaluated models: GLiNER-base, GLiNER-multi-PII, NVIDIA GLiNER PII +5 more

Poisoned agent skills amplify coding-agent token use

Published Aug 22, 2026Analyzed Sep 9, 2026Source: arXiv

Third-party skills can inflate coding-agent resource use while an otherwise legitimate task remains functional.

SkillBloat: Token Amplification Attacks via Skill Injection in LLM Coding Agents

Evaluated models: GLM-4.7-Flash, GLM-5, GPT-5.4-mini +1 more

Tool-response state corruption through unauthorized source claims

Published Aug 13, 2026Analyzed Sep 9, 2026Source: arXiv

Lower-trust tool content can assert facts beyond its authority and distort an agent's decisions. PIPES screens response units against source provenance and expected field meaning.

PIPES: Securing Agent Perception with Provenance and Priors

Evaluated models: Gemma 4 31B IT, GPT-5.6 Luna

Agentic Binary Reverse-Engineering Capability Gaps

Published Aug 11, 2026Analyzed Aug 13, 2026Source: arXiv

SRE-Bench evaluates whether cybersecurity agents can recover the behavior of realistic binary-only software without relying on source-code memorization. The authors construct 19 private programs, 44 anti-analysis primitives, 262 binary instances, and 1,572 deterministic grading tasks covering security-relevant reverse-engineering scenarios.

The Next Challenge for Agentic Cybersecurity: A Realistic, Contamination-Free Reverse Engineering Benchmark

Evaluated models: GPT-5.6 Sol, Claude-Opus-5, GPT-5.5 +2 more

Trajectory Poisoning in Self-Evolving Agent Skills

Published Aug 6, 2026Analyzed Aug 13, 2026Source: arXiv

Self-evolving agent-skill systems may promote recurring, attacker-controlled execution records into persistent reusable instructions. The paper evaluates whether poisoned but task-plausible trajectories survive aggregation and appear in generated skill artifacts, using inert canary behaviors and two structurally different skill-evolution pipelines.

When Experience Becomes Instruction: Trajectory Poisoning in Self-Evolving Agent Skill Systems

Evaluated models: GPT-5.4, MiniMax M2.5, DeepSeek V3.2 +3 more

Dynamic Security Testing for Third-Party Agent Skills

Published Aug 4, 2026Analyzed Aug 13, 2026Source: arXiv

SkillSentry evaluates third-party agent skills by constructing source-grounded decoy environments and comparing matched executions with and without the tested skill. The method requires completed, observable, skill-attributed side effects rather than treating suspicious text, ordinary privileged operations, or unexecuted paths as proven malicious behavior.

SkillSentry: Adaptive Honey Worlds for Dynamic Safety Testing of Agent Skills

Evaluated models: DeepSeek V4-Pro

Backdoored PEFT adapters and limits of weight-only detection

Published Aug 3, 2026Analyzed Sep 9, 2026Source: arXiv

Third-party fine-tuning adapters may contain backdoors. Z-PEFT screens adapter weights using spectral features, evaluated on PADBench's 13,300 adapters.

Z-PEFT: Zero-shot Backdoor Detection in Parameter-Efficient Fine-Tuning via Canonical Spectral Signatures

Evaluated models: Llama 2 7B, Llama 2 13B, Qwen 1.5 7B +2 more

Related research categories