LFDT Sweden

Zhiyang Chen (University of Toronto & ETH Zurich) will give a seminar on 04 Sep 2026 at 15:00 in KTH Room 1440 (Henrik Eriksson) and on Zoom.

Title
Scam2Prompt: A Scalable Framework for Auditing Malicious Scam Endpoints in Production LLMs

Abstract
Large language models are increasingly used for software development, but their generated code can sometimes incorporate malicious URLs and endpoints originating from scam websites, even in response to seemingly benign developer requests. In this talk, I will present Scam2Prompt, a scalable framework for auditing this behavior in production LLMs. Scam2Prompt starts from known malicious websites and automatically constructs realistic developer-style prompts to test whether LLMs reproduce scam infrastructure in generated code. Using this framework, we build Innoc2Scam-bench, a renewable benchmark of 1,377 prompts, and evaluate several state-of-the-art production LLMs. Our results show that malicious endpoint generation remains prevalent across recent models. Interestingly, the framework also uncovered previously unknown active scam websites, 62 of which were subsequently confirmed and added to MetaMask’s scam database. More broadly, this work highlights a security risk beyond explicit harmful prompting: LLMs may unintentionally propagate unsafe web content when assisting users with executable code.

The tokenisation of regulated real-world assets has led to multiple institutional implementations of the same financial instruments, each operating on its own infrastructure with few shared patterns for how those infrastructures interoperate. As a result, tokenised assets remain confined to the environment they were issued in, accessible only to counterparties already inside the same institutional perimeter.

An online session introducing the Personal Data Portability Protocol (PDPP), following its public announcement at the Global Digital Collaboration Conference in Geneva. This is the online counterpart to Geneva for everyone who could not be there: what the specification defines (user-authorized access to specific personal data, built on OAuth 2.0 and RFC 9396), what the August working sessions changed in the draft, and how to take part in the public comment period. Format is roughly 40 minutes of presentation and demonstration, followed by 20 minutes of questions.

Join us for a discussion about the new VeltoraCore lab that is building a pair-coin, system-agnostic optimization engine that jointly coordinates user execution and LP participation in a single feasibility framework. A lightweight Proof-of-Optimization (PoO) mechanism cryptographically commits optimized batch outcomes to external system states, enabling verifiable cross-system settlement. Simulations on the ETH–USD pair show superior retained execution and materially lower user fees relative to benchmarks, while delivering liquidity-provider returns higher with reduced volatility.

This is a great opportunity to talk to the developers about what they're building and to learn how to get involved if this sounds interesting.