New research on AI and security, selected for relevance to language models, AI agents and cybersecurity. Summaries are the authors’ or publisher’s own words, linked to the original.
Peer-reviewed
A multimodal foundation model for molecular and morphologic data in oncology
Large language models' complicit responses to illicit instructions across socio-legal contexts
Generative molecular design needs oracle grounding
Anthropic IPO: four ways the AI giant wants to disrupt science
Anthropic IPO: four ways the AI giant wants to disrupt science
AI is turbocharging cyberattacks — can computer scientists catch up?
New preprints (not yet peer-reviewed)
Hardening JavaScript Programs against LLM-Based Vulnerability Detection
JavaScript programs are often distributed as source code, so attackers can analyze them for vulnerabilities. LLM-based vulnerability detection agents now make this analysis much easier.
Safeguarding LLMs via Model-Agnostic Latent Safety Signals from Dark Knowledge
LLMs have advanced rapidly, raising growing concerns about their safety. Recent work has proposed approaches to detect and defend against attacks including defenses at decoding stage that leverage models' hidden states.
ChainAgent: Automated Chainable Vulnerability Discovery in PHP Web Applications
High-impact attacks on PHP web applications often chain limited-capability vulnerabilities with other flaws or intended behaviors. Discovering these chains requires identifying flaws and the dependencies enabling capability composition.
Privacy-Utility Tradeoffs with Quantum Local Differential Privacy
When sensitive information is encoded in data, it is important to ensure the privacy of information when attempting to learn useful information from the data. There is a natural tradeoff whereby increasing privacy requirements may decrease the utility of a learning protocol.
Post-Reasoning: Improving the Performance of Non-Thinking Models at Negligible Cost
As the widespread adoption of Large Language Models (LLMs) accelerates, token consumption from intermediate reasoning traces increasingly contributes to inference latency and operational cost. Recent studies suggest that many real-world tasks require little to no explicit reasoning, with additional reasoning sometimes even degrading performance.
ToolPulse: Side-Channel Eavesdropping on Tool-Using LLM Agents
Large language models (LLMs) are increasingly deployed as agents that translate user requests into sequences of local tool invocations. While prior cache-based side-channel attacks target the locally deployed models, they typically require the adversary to share model-related components or hardware resources with the victim's inference process.
Verifiable Logic in Medical LLMs: A Critical Review of Agentic Clinical Reasoning Optimization and Neuro-Symbolic Grounding
Clinical large language models are usually judged by whether the final answer is right. Yet a model can reach a fluent, plausible output through invalid reasoning that static benchmarks do not detect, and confident hallucination, causally invalid inference, and spurious demographic shortcuts can pass conventional evaluation.
Replication of Agentic Security Events: Audit and Investigation (ROGUE-AI)
Agents based on large language models (LLMs) can now delete files and send payments on behalf of their users. News media reports routinely describe failures of these agents (e.g., Replit's coding agent deleting a production database), but a report often only says what went wrong and rarely (in)validates the incident or finds its root cause.
