#security News & Analysis

On-Policy Self-Evolution via Failure Trajectories for Agentic Safety Alignment

Seeing the Needle in the Haystack: Towards Weakly-Supervised Log Instance Anomaly Localization via Counterfactual Perturbation

Continuous Discovery of Vulnerabilities in LLM Serving Systems with Fuzzing

Natural Language based Specification and Verification

Can a Single Message Paralyze the AI Infrastructure? The Rise of AbO-DDoS Attacks through Targeted Mobius Injection

IPI-proxy: An Intercepting Proxy for Red-Teaming Web-Browsing AI Agents Against Indirect Prompt Injection

Elliptic raises $120 million backed by Nasdaq, Deutsche Bank as AI reshapes crypto security

Three Tennessee men indicted in alleged $6.5 million crypto robbery spree

Three Tennessee men have been indicted for allegedly orchestrating a $6.5 million cryptocurrency robbery spree across multiple California cities. The case highlights the physical security vulnerabilities in the crypto space, where high-value digital assets attract violent criminals despite being theoretically non-custodial.

Formal Policy Enforcement for Real-World Agentic Systems

SkCC: Portable and Secure Skill Compilation for Cross-Framework LLM Agents

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

You Have Been LaTeXpOsEd: A Systematic Analysis of Information Leakage in Preprint Archives Using Large Language Models

Simulating Complex Multi-Turn Tool Calling Interactions in Stateless Execution Environments

Mitigating Watermark Forgery in Generative Models via Randomized Key Selection

SAR-RAG: ATR Visual Question Answering by Semantic Search, Retrieval, and MLLM Generation

Identifying Backdoored Graphs in Graph Neural Network Training: An Explanation-Based Approach with Novel Metrics

When Prompts Become Payloads: A Framework for Mitigating SQL Injection Attacks in Large Language Model-Driven Applications

To Redact, or not to Redact? A Local LLM Approach to Deliberative Process Privilege Classification

Threat Modelling using Domain-Adapted Language Models: Empirical Evaluation and Insights

BEACON: A Multimodal Dataset for Learning Behavioral Fingerprints from Gameplay Data

Engineering Robustness into Personal Agents with the AI Workflow Store

Clear Signing: Making Transaction Approvals Safer on Ethereum

Repeated Deceptive Path Planning against Learnable Observer

Researchers introduce Repeated Deceptive Path Planning (RDPP), a framework addressing how agents can conceal destinations from learning adversaries who adapt over time. The proposed Deceptive Meta Planning (DeMP) algorithm uses two-level optimization to sustain deception against evolving observers, outperforming existing static-observer approaches while maintaining reasonable path costs.

Olympic Sprinter Can't Outrun Charges in UK Crypto Fraud Investigation

British Olympic sprinter CJ Ujah faces charges in a cryptocurrency fraud investigation involving wallet seed phrase theft and fraudulent impersonation calls. The case highlights how high-profile individuals can become entangled in crypto scams and underscores persistent security vulnerabilities in the digital asset ecosystem.

Governed Metaprogramming for Intelligent Systems: Reclassifying Eval as a Governed Effec

Researchers propose governed metaprogramming, a language design framework that reclassifies the eval function from an unrestricted primitive into a controlled effect subject to governance and inspection. The approach aims to address security and authority risks in AI systems that synthesize executable code at runtime, with implementation demonstrated in MashinTalk, a DSL for AI workflows.