CrowdStrike researchers analyzed AI infrastructure from four previously unreported threat groups operating between May and August, writing in Chinese, Spanish/English, Indonesian and Russian. Two built sophisticated AI 'harnesses' while two less technically advanced groups still compromised multiple targets, mostly relying on older versions of Anthropic's Claude and open-weight Deepseek models left exposed on public servers.
wired.com
· 2026-10-21
Security firm Claroc found that Meta's Muse AI agent generated shareable chat URLs using the conversation's title as a subdomain, which were recorded in public DNS and discoverable via passive DNS searches. Claroc said it enumerated roughly 6,800 shared Muse links, exposing personal and business data including meeting schedules, a loan application, a bank fraud alert and internal company documents. Meta confirmed to Ars Technica that it has paused the sharing feature while reviewing the report.
engadget.com
· 2026-10-20
AI cloud provider Modal has agreed to deploy Nvidia's new Arm-based Vera CPUs after preliminary internal testing showed roughly double the throughput of its existing AMD and Intel processors for agentic AI workloads. Modal says it will continue using its current CPU suppliers alongside Vera, rather than replacing them outright.
tomshardware.com
· 2026-10-20
Yelp has launched Check for Me, a new AI agent that calls restaurants on a user's behalf to check wait times and texts back updates within minutes. The feature expands Yelp Assistant, which already integrates with OpenTable for bookings and offers a Waitlist tool, alongside a new personalization feature called Keep Me Posted for scheduled recommendations.
cnet.com
· 2026-10-20
Arm announced results showing its AGI CPU delivers nearly double the performance per rack compared to AMD's EPYC 9755, based on modeled rack calculations across seven agentic AI workloads. The figures came from a new CPU evaluation methodology Arm developed with input from Fujitsu, Meta, and OpenAI, and were independently verified by Signal65.
tomshardware.com
· 2026-10-19
Contexte Tech's Tony argues that the Model Context Protocol (MCP), launched by Anthropic in November 2024, is increasingly used by AI assistants like Claude, ChatGPT and Cursor to fetch live answers directly from publishers' own servers rather than relying on frozen training data. The protocol was transferred in December 2025 to the Agentic AI Foundation, part of the Linux Foundation, and has progressed through several dated versions, with a public registry now listing available servers.
contextetech.com
· 2026-10-17
The IEEE Computer Society's Columbia Section Chapter (Region 3) held a technical session examining how AI in software development is moving beyond simple code-completion tools. Speakers discussed the emergence of agentic AI systems capable of more autonomous roles in enterprise software engineering workflows.
wpcms.computer.org
· 2026-10-16
IEEE Computer Society reports that agentic AI systems, which plan and execute multi-step tasks rather than just answering single requests, are now moving into real pilots and early production use across software engineering, operations, and business automation. These systems differ from traditional applications by taking autonomous actions over time, which is prompting operators to reconsider how cloud infrastructure and monitoring tools are built around them.
wpcms.computer.org
· 2026-10-16
Nvidia has launched Nemotron 3.5 Super VL, a 120-billion-parameter open-weight model aimed at agentic enterprise tasks. The company says benchmark testing by Artificial Analysis shows it is five times faster than the next-best open model in its class, outperforming InclusionAI's Ling 3.0 Flash and DeepSeek's V4.1 Flash. Nvidia also cited a 44% accuracy gain in a medical insurance claims task after post-training, surpassing Claude Opus 5 on that specific benchmark.
zdnet.com
· 2026-10-15
MacPaw has launched Eney, an AI assistant that monitors a Mac's apps, files, calendar, and system status to flag tasks needing attention, such as preparing for an upcoming call or warning about low disk space. The assistant can be invoked via a gesture called Eney Wand, which lets users select on-screen text and prompt the AI directly, or through the Option+W shortcut. Development began in 2019 as a way to simplify CleanMyMac through natural-language commands, evolving alongside advances in LLM and agentic AI technology.
9to5mac.com
· 2026-10-15
CedarDB, a Munich-based database startup spun out of Technical University of Munich research project Umbra, builds a database engine designed to handle both transactional and analytical workloads in one system. The company, led by cofounder and CEO Moritz Sichert, built CedarDB to be compatible with PostgreSQL to minimize switching costs for enterprise customers.
techcrunch.com
· 2026-10-14
Marc Brooker, an AWS engineer, responded to questions about why the strands-decider-2B model uses a modified decoder (LLM) architecture rather than a bidirectional encoder like BERT. He ran comparisons across five model designs, including the existing Qwen3.5-based decoder (v19), two encoder-decoder variants built on T5Gemma 2 at different parameter sizes, and encoder-only models based on T5Gemma. The post presents these as empirical tests of whether bidirectional encoders outperform causal decoders for this classification-style task.
brooker.co.za:443
· 2026-10-14
GoKawiil