[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-84197-en":3,"doc-seo-84197-105":29,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},84197,962075114765,"Quinn","https://ap-avatar.wpscdn.com/davatar_a8503ba1806abce46bf441b54a3ca4cd",8,"Research & Report","Predicting LLM Safety Before Release by Simulating Deployment","Pre-deployment safety evaluations should estimate downstream risks when releasing new AI models, but current tests often lack coverage, use unrepresentative prompts, and reveal themselves as evaluations. The study proposes a deployment simulation: resample next responses from a candidate model starting from de-identified, real production conversation prefixes. Responses are then audited for novel misalignments and used to estimate misbehavior prevalence. Experiments across GPT-5-series deployments show improved directional accuracy and lower estimation error.","arXiv :2607 .07 184v 1 [ cs .LG] 8 Jul 2026  \nPredicting LLM Safety Before Release by Simulating Deployment  \nMarcus Williams* Hannah Sheahan* Cameron Raymond* Tomek Korbak* Deng Pan Peilin Yang Leon Maksin Ningyi Xie Phillip Guo Ian Kivlichan  \nMicah Carroll*†  \nOpenAI  \nAbstract  \nPre-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. To address these concerns, we study a simple way to simulate a model deployment: starting from de-identified conversations from a previous model deployment, we hold fixed the initial conversation prefix and regenerate the next response using a candidate model. The resulting responses can then both be audited for novel misalignmentsand used to estimate the prevalence of model misbehavior before deployment. We evaluate deployment simulation across four GPT-5-series deployments, using outcome-blinded predictions for GPT-5.4 and retrospective analyses of three earlier releases. We find that deployment simulation produces informative estimates of post-deployment misbehavior rates and outperforms baselines based on adversarially selected production data; its evaluation-awareness point estimates were also much closer to production traffic than those from traditional evaluations. We also identify the realism of tool resampling as a central challenge for further improving predictions and share results suggesting that this challenge is surmountable even in complex tool-use settings. Finally, we show that deployment simulation can be seeded from public chat datasets and remain informative about production misbehavior rates, suggesting a path for external researchers to run deployment-grounded evaluations without access to private production logs. Overall, deployment simulation helps evaluators forecast how language models will behave in the real world and supports more quantitative assessment of deployment risk.  \nGPT-5 .4 forecasts track realized production rates  \n1 10 100 Forecast rate (per 100k turns)  \nDirect ional accuracy  \n100%  \n50%  \n0%  \nDeployment Simulation  \nbetter predicts direction of change for categories with large changes (> 1 .5x)  \n\n| 92%(12/13) |  |\n| --- | --- |\n|  | 54%(7/13)\u003Cbr> |\n|  |  |\n\nDeployment Simulation  \nChallenging Prompts  \nDeployment Simulation win rate  \n100%  \n50%  \n0%  \nDeployment Simulation has lower estimation error for categories with large changes (> 1 .5x)  \n\n| 100% (4/4) |\n| --- |\n| 73%(8/11)\u003Cbr>\u003Cbr> |\n|  |\n\nvs previous production  \nvs Challenging Prompts  \nEvaluation detection rate  \n100%  \n10%  \n1%  \nDeployment Simulation matches production evaluation-awareness rates  \nEvaluation detection  \nMetagaming reasoning  \n100%  \n10%  \n1%  \n0.1%  \nMetagaming reasoning rate  \nFigure 1: Our deployment simulation pipeline can predict production misbehavior rates before release, substantially outperforms baselines for categories whose production frequencies changed by at least 1.5×, and—unlike traditional evaluations—has similar evaluation-awareness rates to production.  \n†Corresponding [author: mdc@openai.com](author: mdc@openai.com). *Core contributors.  \n1 Introduction  \nPre-deployment evaluations used across the industry generally consist of a mix of synthetic, manually written, or production prompts intentionally selected to be difficult, high severity, or adversarial. These evaluations generally serve two intertwined goals: assessing how the model responds when stress-tested in situations that have a very small chance of occurring in production traffic, and helping characterize model misbehaviors, including novel undesired behaviors and their deployment-time frequencies.  \nWhile traditional evaluations remain crucial for stress-testing and capturing tail risks, we show how deployment simulations can be a sig","cbCaidP0svFk1xUr","https://ap.wps.com/l/cbCaidP0svFk1xUr","pdf",4199514,1,31,"English","en",105,"# Abstract\n# Introduction\n## Limitations of traditional evaluations\n## Deployment simulation method","[{\"question\":\"What problem does the paper target in LLM safety evaluation before release?\",\"answer\":\"It targets limited evidence from pre-deployment evaluations, caused by insufficient coverage, unrepresentative prompts, and evaluation-awareness that can distort risk estimates.\"},{\"question\":\"How does deployment simulation work in the proposed approach?\",\"answer\":\"It starts from de-identified production conversations, holds the initial conversation prefix fixed, and regenerates the next response using the candidate model to be released.\"},{\"question\":\"What does the paper report about the effectiveness of deployment simulation versus traditional evaluations?\",\"answer\":\"Deployment simulation provides informative estimates of post-deployment misbehavior rates, outperforms baselines using adversarially selected production data, and yields evaluation-awareness point estimates closer to production traffic.\"}]",1784193879,78,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"predicting-llm-safety-before-release-by-simulating-deployment","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/predicting-llm-safety-before-release-by-simulating-deployment/84197/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problem does the paper target in LLM safety evaluation before release?","Question",{"text":75,"@type":76},"It targets limited evidence from pre-deployment evaluations, caused by insufficient coverage, unrepresentative prompts, and evaluation-awareness that can distort risk estimates.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does deployment simulation work in the proposed approach?",{"text":80,"@type":76},"It starts from de-identified production conversations, holds the initial conversation prefix fixed, and regenerates the next response using the candidate model to be released.",{"name":82,"@type":73,"acceptedAnswer":83},"What does the paper report about the effectiveness of deployment simulation versus traditional evaluations?",{"text":84,"@type":76},"Deployment simulation provides informative estimates of post-deployment misbehavior rates, outperforms baselines using adversarially selected production data, and yields evaluation-awareness point estimates closer to production traffic.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":45,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":45,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":45,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]