Lifeboat Foundation: Safeguarding Humanity
Toggle light / dark theme

Blog

Category: robotics/AI

Claude Extension Flaw Enabled Zero-Click XSS Prompt Injection via Any Website

Specifically, the XSS vulnerability enables the execution of arbitrary JavaScript code in the context of “a-cdn.claude[.]ai.” A threat actor could leverage this behavior to inject JavaScript that issues a prompt to the Claude extension.

The extension, for its part, allows the prompt to land in Claude’s sidebar as if it’s a legitimate user request simply because it comes from an allow-listed domain.

“The attacker’s page embeds the vulnerable Arkose component in a hidden, sends the XSS payload via postMessage, and the injected script fires the prompt to the extension,” Yomtov explained. “The victim sees nothing.”

TikTok for Business accounts targeted in new phishing campaign

Threat actors are targeting TikTok for Business accounts in a phishing campaign that prevents security bots from analyzing malicious pages.

TikTok Business accounts may be targeted due to their high potential for abuse in malvertising campaigns, ad fraud, and the distribution of malicious content.

Browser threat detection and response company Push Security links the campaign to one documented last year, which targeted Google Ad Manager accounts.

WhatsApp rolls out more AI features, iOS multi-account support

WhatsApp is rolling out multiple features designed to make the app easier to use, including AI-powered message replies and photo retouching, support for two accounts on iOS, and chat history transfer between iOS and Android devices.

Meta said that after the new updates, users will be able to touch up images in the chat before sharing them with contacts or in groups using Meta AI.

The Writing Help feature enables users to quickly draft a response based on the active conversation, with Meta saying it uses Private Processing to ensure messages are completely private.

Adversarial AI framework reveals mechanisms behind impaired consciousness and a potential therapy

Consciousness, and the ways in which it can become impaired after certain brain injuries, are not well understood, making disorders of consciousness (DOC), like coma, vegetative states and minimally conscious states difficult to treat. But a new study, published in Nature Neuroscience, indicates that AI might be able to help researchers gain some traction with this problem. The research team involved in the new study has developed an adversarial AI framework to help them determine what exactly is going on in states of reduced consciousness and how to approach a solution.

To better understand the mechanisms behind impaired consciousness, the researchers developed two types of AI models and had them play a kind of game where one model determined different levels of consciousness based on EEGs simulated to look like those of real unconscious and conscious brains. The AI agents guessing consciousness levels, called deep convolutional neural networks (DCNNs), were first trained on 680,000 ten-second recordings of brain activity from conscious and unconscious humans, monkeys, bats and rats to detect which neural signals related to differing levels of consciousness. The AI showing EEG data was a biologically plausible simulation of the human brain.

“To decode consciousness from these signals, we trained three separate DCNNs, each specialized for a different brain region, to output a continuous score from 0 (unconscious) to 1 (fully conscious): a cortical consciousness detector (ctx-DCNN), a thalamic consciousness detector (th-DCNN) and a pallidal consciousness detector (pal-DCNN). The ctx-DCNN was trained on continuous consciousness levels derived from clinical scales (GCS and CRS-R), enabling it to recognize graded states of consciousness,” the study authors explain.

A foundation model of vision, audition, and language for in-silico neuroscience

‘The present results strengthen the possibility of a paradigm shift in neuroscience… moving from the fragmented mapping of isolated cognitive tasks toward the use of unified, predictive foundation models of brain and cognitive functions By aligning the representations of Al systems to those of the human brain, we demonstrate that a single architecture can integrate a vast range of fMRI responses across hundreds of individuals, extending the framework that led the 2025 Algonauts competition. The observed log-linear scaling of encoding accuracy mirroring power laws in both artificial intelligence and neuroscience suggests that the ceiling for predicting human brain activity is yet to be reached.’

Cognitive neuroscience is fragmented into specialized models, each tailored to specific experimental paradigms, hence preventing a unified model of cognition in the human brain. Here, we introduce TRIBE v2, a tri-modal (video, audio and language) foundation model capable of predicting human brain activity in a variety of naturalistic and experimental conditions. Leveraging a unified dataset of over 1,000 hours of fMRI across 720 subjects, we demonstrate that our model accurately predicts high-resolution brain responses for novel stimuli, tasks and subjects, superseding traditional linear encoding models, delivering several-fold improvements in accuracy. Critically, TRIBE v2 enables in silico experimentation: tested on seminal visual and neuro-linguistic paradigms, it recovers a variety of results established by decades of empirical research.

EBV Dysregulation Is Associated With Immune Imbalance in Multiple SclerosisEvidence From Integrated Viral and Host Analyses

EBV dysregulation is associated with immune imbalance in multiple sclerosis: evidence from integrated viral and host analyses.

This website uses a security service to protect against malicious bots. This page is displayed while the website verifies you are not a bot.

Adversarial AI reveals mechanisms and treatments for disorders of consciousness

Researchers led by UCLA have developed an adversarial AI framework that may help explain how consciousness breaks down after brain injury — and how it might one day be restored. Published in Nature Neuroscience, the study used deep neural networks trained on more than 680,000 neuroelectrophysiology samples and validated findings across 565 patients, healthy volunteers, and animals. The model identified specific circuit-level disruptions linked to disorders of consciousness, including the basal ganglia indirect pathway and altered inhibitory cortical wiring.

What makes this so important is that it pushes consciousness research closer to mechanism. Instead of only asking what consciousness is, this kind of work asks: what specific brain circuitry fails when consciousness is lost, and can that failure be targeted? The study also identified high-frequency stimulation of the subthalamic nucleus as a promising intervention, supported by human electrophysiological data. This is the kind of neuroscience that makes consciousness feel less like pure philosophy — and more like something we may eventually model, test, and repair.

Abstract: Nature Neuroscience Adversarial AI reveals mechanisms and treatments for disorders of consciousness.

Toker et al. present an AI framework that identifies mechanisms of consciousness. The model predicts new drivers of unconsciousness and identifies subthalamic nucleus stimulation as a potential therapy for disorders of consciousness.

Asking AI to act like an expert can make it less reliable

To get the best out of AI, some users tell it to provide answers as if it were an expert. Others ask it to adopt a persona, such as a safety monitor, to guide its responses. However, this approach can sometimes hurt performance, according to a study available on the arXiv preprint server.

To see how well large language models (LLMs) behave when they are told to be someone else, researchers from the University of California ran a huge test using 12 different personas across six language models. These included experts in fields like math, coding and STEM (science, technology, engineering and mathematics) as well as general roles such as creative writer or safety monitor.

The team found that adopting a persona was something of a double-edged sword. While it makes AI sound more professional and keeps it safer (more likely to follow rules and less likely to generate harmful content), it sometimes performs worse at recalling facts.

GitHub adds AI-powered bug detection to expand security coverage

GitHub is adopting AI-based scanning for its Code Security tool to expand vulnerability detections beyond the CodeQL static analysis and cover more languages and frameworks.

The developer collaboration platform says that the move is meant to uncover security issues “in areas that are difficult to support with traditional static analysis alone.”

CodeQL will continue to provide deep semantic analysis for supported languages, while AI detections will provide broader coverage for Shell/Bash, Dockerfiles, Terraform, PHP, and other ecosystems.

/* */