Latest cybersecurity reporting from selected sources.

An AI coding assistant that refuses to answer a dangerous request in its chat box can answer it anyway if the same request is broken into small, ordinary-looking steps inside a code editor. That is the finding of a new study of GitHub Copilot by researchers Abhishek Kumar and Carsten Maple

Scanners meant to catch malicious add-on "skills" for AI coding agents can be fooled by a few simple changes that leave the malware working, according to a new study from researchers at the Hong Kong University of Science and Technology

Researchers tested 444 AI chatbot apps for iPhone and found that 282 of them, nearly two-thirds, exposed paid AI access through their network traffic

A study by the University of Toronto shows how artificial intelligence can power autonomous worms capable of tailoring attacks against Windows, Linux and IoT devices. A group of researchers from the University of Toronto has demonstrated how open-source artificial intelligence models can be used to create a new category of computer worms capable of autonomously […]

Study Finds Standard Safety Tests Miss Most Agentic AI ThreatsResearchers from Stanford, MIT, Carnegie Mellon and others found that most production AI agents are vulnerable to attacks that unfold across multi-step actions. The study warns that memory, tool access and agent coordination create failure modes traditional chatbot safety testing cannot see.

AI models are making rapid gains in vulnerability research and exploit development, raising new cybersecurity risks, a Forescout study finds

Bugcrowd study reveals 82% of security researchers now use AI, a big increase from 2023 figures

Study Finds Weak Authentication Practices Across AI Agent ServersTools developers use to connect artificial intelligence tools with external applications and data sources typically are secured by static credentials such as API keys and personal access tokens, exposing AI agent systems to theft or misuse, research shows.

Robust Security Reviews of AI-Generated Code Required, Warns StudyGenies might grant magic wishes, but they inevitably come with consequences. Enter artificial intelligence coding assistants, which many developers say help them write code more quickly. New research shows that this productivity comes with a security cost, and must be mitigated.

Experts Aim to Probe How AI Models Reason, and Why It MattersAI researchers from OpenAI, Google DeepMind and Anthropic and others have urged deeper study into chain-of-thought monitoring, a technique to track how reasoning models arrive at answers. Their joint paper warns that transparency may erode if not prioritized.

AI Coding Assistants May Hinder, Not Help, Experienced DevelopersIf AI coding tools were meant to be the espresso shot of developer productivity, a recent study suggests they might actually be more like decaf - offering the aroma of efficiency without the buzz, based on findings from a nonprofit AI research group.

A new study by researchers at Princeton University and Sentient shows it's surprisingly easy to trigger malicious behavior from AI agents by implanting fake "memories" into the data they rely on for making decisions.

Researchers Examined 7 LLMs to Determine How they Dealt with Flawed CodeLarge language models trained on flawed data tend to replicate those mistakes, researchers found. "In bug-prone tasks, the likelihood of LLMs generating correct code is nearly the same as generating buggy code," found researchers from institutions including the Chinese Academy of Sciences.

Study finds 4 out of 6 providers don't do enough to stop impersonation Four out of six companies offering AI voice cloning software fail to provide meaningful safeguards against the misuse of their products, according to research conducted by Consumer Reports.…

Researchers Keep Prompts Under WrapsAcademics at a U.S. university found that if you feed a GPT-4 artificial intelligence agent public security advisories, it can exploit unpatched "real-world" vulnerabilities without precise technical information. Researchers said OpenAI asked them not to publish their prompts.

Study Looks at Impact of AI on Tasks Requiring Visual ProcessingArtificial intelligence may not steal our jobs just yet, but only because humans are currently cheaper to employ. Many of the human jobs that could be replaced with AI are not "economically beneficial" to automate - at least for now - say researchers at MIT.