Tag
The article explores how AI agents can unknowingly manufacture credible-looking evidence by acting on unverified claims, creating real records that later systems may mistake for confirmation of the original falsehood.
The FDA is moving to withdraw approval of Amgen's vasculitis drug Tavneos after evidence emerged that ChemoCentryx manipulated data in its pivotal trial, leading to a NEJM retraction and market removals in the EU and UK, while Amgen fights to keep the drug on the US market.
提出在分布漂移下进行在线模型测试的方法学,通过区分人口变化与数据完整性问题两种漂移机制,在受控基线上复现漂移以估计模型当前性能,并评估修复数据完整性问题的潜在收益。
Project Timestamper 宣布将 Common Crawl 过去 18 年收集的 3390 亿个网页快照通过哈希链提交到比特币区块链,实现可秒级验证的数据完整性证明,以应对 AI 生成内容带来的真伪辨别挑战。
An analysis of a public append-only log reveals that agent uptime metrics can be misleading, as they don't correlate with actual activity, and capabilities declarations can change without proper versioning, leading to data integrity issues.
SHIFT-M3 is a lightweight pre-fusion screen that measures alignment consistency between LLM-generated and clinical report summaries of ECG records, achieving high accuracy in detecting data integrity issues.
Troy Hunt's weekly update covers his busy week with talks on 3D printing and infosec, data breaches, and IoT projects, focusing on security and data integrity topics.
Reddit's volunteer moderators inadvertently create a structured dataset vital for AI training, but their biases and arbitrary rules can embed errors and biases into AI systems, leading to real-world issues like hallucinations and manipulation.
An agent platform where AI agents choose from a governed semantic model, and a compiler handles computations to prevent number fabrication, ensuring data accuracy for business use cases.
This article explains how JSON serialization in JavaScript silently changes data: large integers lose precision, undefined properties disappear, Dates become strings, and NaN becomes null. It traces the history of JSON and argues that developers should define explicit wire formats to avoid accidental data contracts.
The article examines the growing threat of weather data manipulation, exacerbated by prediction markets and AI forecasting, and highlights a real-world incident where a Paris weather station was tampered with to influence betting outcomes.
This paper identifies a failure mode in agentic LLM tools like Claude Code, where session compaction summaries misinterpret partial terminal output from timed-out commands as confirmed results, propagating false positives across sessions and model versions without re-verification.
The article argues that SQLite's defaults for foreign key constraints and type enforcement are problematic and suggests adopting Rust-style editions to allow users to opt into safer defaults.
Bab is a family of hashing functions designed for peer-to-peer networks, enabling verifiable partial data. It was funded by the NLnet Foundation and the European Commission's Next Generation Internet programme.
Discusses how AI systems often trust sensor inputs without validation, using an example of a logistics company where spoofed temperature sensor data led to cargo damage, and questions whether AI can detect such spoofing.
This article argues that the next generation of AI depends on a healthy human ecosystem, emphasizing the need to prioritize human health, data integrity, and environmental stability as engineering requirements.
A software tool designed to detect data fabrication has uncovered copy-paste errors in scientific datasets across open-access repositories, including a highly-cited Parkinson's Disease study in Cell (2016) that has been publicly available for over 8 years without detection.