Research
Cisco Talos Unveils CAIRN to Detect AI-Driven Malware
WIRED AIhigh confidence
Why it matters
- →CAIRN provides the first open-source method to detect digital fingerprints of AI integration in malware.
- →The discovery of CLOSEDQUORUM proves attackers are already using multi-model LLMs for fully autonomous command-and-control.
- →Security teams can now systematically classify and track emerging trends in AI-driven cyber threats.
Cisco Talos researchers have released CAIRN, an open-source framework for detecting and classifying malware integrated with artificial intelligence. Using this tool, they identified CLOSEDQUORUM, a Windows-based threat that autonomously coordinates attacks by polling multiple large language models, including DeepSeek and Google Gemini, to determine its next steps. The malware operates without human input, creating a redundant 'hive mind' infrastructure for credential theft and cryptocurrency fraud. While most AI-integrated malware remains experimental, CAIRN allows the security community to systematically track these emerging artifacts and identify trends in autonomous cybercrime.
Read originalMore from WIRED AI
© WIRED AIAgentsagents
Rabbit Launches OS3 Agentic Platform
Rabbit pivots from failed hardware to a cross-platform agentic operating system called OS3. It runs locally on desktops while being controlled via phone or browser, using your own API keys for models like OpenAI or Anthropic. The system executes local tasks and integrates third-party agents through a simple chat interface. This marks a significant shift from proprietary hardware dependency to an open software ecosystem that leverages existing devices.
WIRED AI
© WIRED AIInvestment · $50M
Models & Labsvideo
TBC Brings Rat Brain AI to AWS for Video Gen
The Biological Computing Company is moving from stealth to a limited AWS preview, offering a video generation model powered by living rat neurons. This isn't just wetware hype; they claim five times faster inference than frontier open-source models by mapping visual data onto multi-electrode silicon arrays. With $50 million in funding and Jeff Dean’s backing, TBC is betting that biological neural patterns can solve the efficiency bottlenecks of current generative video pipelines. The real test is whether this 'wetware as a service' scales beyond short clips without losing fidelity.
WIRED AI
© WIRED AIModels & Labsother
Viture Launches Privacy-Focused AI Glasses
Viture is pivoting from bulky VR displays to a sleek pair of smart glasses that map your life through audio alone. The Vonder device uses bone-conduction sensors to capture speech without cameras, building a personal memory graph that connects your daily experiences and goals. By routing queries through multiple LLMs like Claude and Gemini, it aims to act as an external brain rather than just a recorder. This approach sidesteps the privacy backlash plaguing Meta’s Ray-Bans by keeping data local and visual indicators subtle. It represents a distinct shift toward passive, audio-first AI companions that prioritize user comfort and discretion over constant surveillance.
WIRED AI
More in Research
© The AI Daily BriefResearchresearch
OpenAI and Anthropic Abandon Cross-Testing Deal
A previously planned cross-testing agreement between OpenAI and Anthropic has been abandoned.
The AI Daily Brief
© Hugging Face BlogResearchresearch
AISI releases verified evals for GPT-5 and Claude Opus 4
The UK AI Security Institute has published verified benchmark results for GPT-5 and Claude Opus 4 using EvalEval’s standardized schema, solving the reproducibility crisis in frontier model testing. By releasing raw configuration data alongside scores from benchmarks like SWE-Bench Pro and Humanity's Last Exam, they prove that inference-time compute drastically alters performance curves. This moves evaluation beyond opaque leaderboards into auditable science, allowing researchers to see exactly how protocol choices skew reported capabilities. It sets a new standard for transparency in high-stakes AI security assessments.
Hugging Face Blog
© Hugging Face BlogResearchmodels
Physics-based pruning beats heuristics for LLM compression
Multiverse AI reframes block removal as an Ising glass optimization problem, capturing the hidden couplings between transformer layers that mean-field methods ignore. By mapping block importance to spin interactions via a Hessian matrix, they turn model compression into a search for low-energy states rather than independent block scoring. This approach yields a massive 23-point MMLU gain over existing baselines when compressing Llama-3.3-70B by half, proving that many-body physics tools can unlock deep compression without retraining. The method scales to large models using classical and quantum-inspired solvers, offering a rigorous alternative to heuristic pruning.
Hugging Face Blog