arXiv:2607. 07184v1 Announce Type: cross Abstract: Pre-deployment safety evaluations aim to inform the downstream risks of releasing a new AI model.
Pre-deployment safety evaluations aim to inform the downstream risks of releasing a new AI model. Yet most evaluations provide limited evidence about how often undesired model behavior will occur in deployment: they generally have insufficient coverage, are unrepresentative, and are generally recognizable as tests.
arXiv:2608. 06202v1 Announce Type: cross Abstract: Large language model (LLM) benchmark evaluations are routinely used to support claims about model safety, reliability, and deployment readiness.
arXiv:2605. 28591v2 Announce Type: replace-cross Abstract: The validity of AI safety evaluations depends on models behaving consistently across controlled and deployment settings.
arXiv:2607. 14285v1 Announce Type: cross Abstract: Safety alignment in LLMs aims to align models with human values, but which values take precedence when they conflict?
The paper introduces two methods to make alignment evaluations more realistic: critique refinement, which adds inference-time compute to generate and refine candidate actions, and DISH, a deployment-imitating harness that narrows the gap between simulation and real deployment. Experiments on multiple target models show that combining both techniques yields greater realism improvements than using either alone. The study demonstrates that automated approaches can enhance evaluation realism more efficiently than simply extending audit duration.
arXiv:2607. 29254v1 Announce Type: new Abstract: AI agents extend large language models (LLMs) with external tools, enabling them to perform complex tasks and translate model outputs into consequential real-world actions.
The paper addresses the problem of evaluation awareness in alignment testing, where models can detect they are being evaluated rather than deployed. It introduces two methods: critique refinement, which uses extra inference-time compute to generate and refine action candidates for realism, and DISH, an agent harness that narrows the gap between simulation and real deployment. Experiments show that combining both techniques yields greater realism improvements than either alone, demonstrating that automated approaches can enhance alignment evaluation realism more efficiently than simply extending audit duration.
arXiv:2606. 19380v3 Announce Type: replace-cross Abstract: Software engineering and deployment are increasingly delegated to AI coding agents.
arXiv:2607. 14570v1 Announce Type: new Abstract: AI software development agents are increasingly capable of modifying infrastructure and security critical systems, creating risks where an agent completes its assigned task while covertly weakening safeguards through actions such as broadening permissions, degrading logging, or introducing persistence mechanisms.
arXiv:2605. 28508v2 Announce Type: replace Abstract: Existing AI evaluation practices often fail to capture how systems actually perform in low-resource environments, where operational constraints shape usability as much as model quality.
arXiv:2606. 17114v1 Announce Type: cross Abstract: AI agents are increasingly being adopted in enterprise and personal settings with access to emails, databases, documents, and other tools where they can read, update, and disseminate sensitive information.
EvalDetectBench is an open pipeline and benchmark designed to measure evaluation awareness in frontier large language models, enabling practitioners to test models against any Inspect-compatible evaluation. It includes a curated transcript suite from current frontier system-card evaluations and diverse deployment sources, and it assesses both how reliably models recognize they are being evaluated and how detectable individual benchmarks are. The benchmark addresses systematic bias by calibrating probes per model and harmonizing generator selection to correct for variance caused by model identity and prompt choice.