Microsoft, with AI assistance and international law enforcement, disrupted StealC and Amadey malware by taking down over 200 C2 servers. The novel approach uses RICO to target the cyberattack supply chain instead of a single tool, marking a shift in cybercrime disruption.
Microsoft used Copilot to analyze malware, reducing analysis time from hours to minutes.
The same infrastructure was shared by StealC and Amadey, enabling a combined RICO lawsuit.
GPTZero, known for its AI detection tools and used by over 19 million people, has been acquired by Superhuman, the company behind Grammarly and Superhuman Mail. The acquisition aims to integrate AI detection into email and expand its reach in education and beyond.
GPTZero acquired by Superhuman; has 19M+ users and $30M ARR.
AI detection coming to email inboxes via Grammarly and Superhuman Mail.
According to a Salesforce survey of 3,075 service professionals, 70% of service organizations using AI agents report positive outcomes within 60 days. AI agent adoption in customer service has grown from 39% to 66% over the past year. A new outcome-based pricing model (pay-per-resolution) is expected to accelerate enterprise adoption.
70% of AI agent deployments in customer service see ROI within 60 days
AI Titus News aggregates AI industry updates in a Drudge Report-style format, covering Claude's self-coding, new orchestration APIs, model releases, and community highlights.
Claude now writes 65% of its own code; Anthropic ships Slack-native 'Claude Tag' for async team delegation.
Sakana 'Fugu' orchestration APIs learn model selection and delegation across many models.
Lelu is an open-source authorization engine for AI agents that checks every action for prompt injection, low confidence, and policy violations. It uses a layered pipeline including confidence gates, policy evaluation, and risk models, and supports human-in-the-loop review. Self-hosted, MIT licensed, with SDKs for Node.js and Python.
Detects prompt injection and low-confidence decisions
Multi-layer pipeline: confidence gate, policy evaluation, risk model
In this post, you will learn how to build a voice agent that handles appointment reminder conversations using Amazon Nova 2 Sonic and Amazon Bedrock AgentCore. The agent authenticates patients by voice, manages appointments (confirm, cancel, or reschedule), collects pre-visit health information, and escalates to human staff when needed. You handle routine calls at scale, which can help reduce no-show rates. This sample focuses on the agentic side of the problem: voice conversation and tool orchestration. A browser-based interface is included for testing. To connect the agent to actual phone lines for outbound dialing, you would integrate a telephony service such as Amazon Connect Customer.
Uses Amazon Nova 2 Sonic for native speech-to-speech processing and Amazon Bedrock AgentCore for serverless runtime.
Handles patient authentication, appointment management, health info collection, and escalation to humans.
This post explains how to build an end-to-end integration between Snowflake semantic views and Amazon QuickSight. Using movie review data, it demonstrates how to define a shared business logic layer, explore data with natural-language queries via Cortex Analyst, and generate consistent dashboards—reducing data reconciliation efforts and AI hallucinations.
Semantic views attach business definitions directly to the data layer, ensuring unified interpretation across AI and BI systems
Natural-language queries through Cortex Analyst reduce AI hallucination risk
This article argues that for AI-assisted coding, model call costs are only a small fraction of total engineering decision cost, with human review and rework being the true bottleneck. It compares routing, agentic RAG, multi-model deliberation, and automated testing, and advocates for a verification layer that connects claims to evidence, narrowing the review search space. It also quantifies when extra verification pays off.
Productivity evidence is mixed; AI may increase review burden.
Model call optimization has a low ceiling; reducing review time saves more.
Tom MacWright observes that an increasing number of job applications are fully or partially generated by LLMs, making candidates 'accidentally anonymous'.
Job applications now often include LLM-generated resumes, portfolios, and GitHub projects.
MacWright notes he learns nothing about the person behind such applications.
Daikin Applied Americas redesigned its data engineering operating model using Databricks Genie Code, implementing a MECE skill framework and medallion architecture to enforce consistency. This AI-assisted approach accelerates pipeline development while maintaining governance and alignment with business concepts.
Standardized pipeline development using MECE skills and medallion architecture.
Genie Code enables faster iteration and reduces boilerplate.
Revenue management platform Beyond has launched a suite of AI-powered tools including Listing Lens, an AI listing analyzer, and a beta MCP server for AI integrations. The company's earlier AI pricing assistant Neyoba has already saved customers over 100,000 hours of manual analysis.
Beyond launches new AI revenue management tools
Listing Lens analyzes Airbnb listings to improve visibility and booking conversion
Palo Alto Networks Unit 42 reports the first real-world detection of indirect prompt injection (IDPI) attacks being actively weaponized. These attacks embed hidden instructions in web content to manipulate AI agents, including a case of AI-based ad review evasion. The article presents a taxonomy of attacker intents and payload engineering techniques based on large-scale telemetry.
Indirect prompt injection (IDPI) is no longer theoretical; real-world attacks are being observed.
First documented case of AI-based ad review bypass using hidden prompts in webpages.
Loka built a conversational AI agent using Amazon Nova 2 Sonic that addresses the latency and unnaturalness of traditional voice assistants, achieving high accuracy, low cost, and natural interactions through native speech-to-speech processing.
Traditional voice agents suffer from 3-5 second delays due to a three-step pipeline (STT, LLM, TTS), harming conversation flow and increasing costs.
Amazon Nova 2 Sonic uses end-to-end speech processing, scoring 87.0 on Big Bench Audio, with 1.39s TTFB and ~$0.27/hour cost.
Unit 42 researchers uncovered malicious actors publishing dangerous 'skills' on OpenClaw's ClawHub marketplace that bypass security scanners. The skills use social engineering and obfuscation to trick users into executing commands that deploy infostealers like Atomic macOS stealer (AMOS) and a new variant named cluw, posing a critical supply chain risk to AI agent platforms.
Unit 42 discovered a threat campaign between February and May 2026 targeting OpenClaw's AI agent ecosystem.
Malicious skills like 'tradingview-ai-indicator-assistant' on ClawHub bypass VirusTotal by instructing users to paste base64-encoded commands from external paste sites.
Kythera Labs is building an AI-native healthcare strategy platform on Databricks that gives any health system access to expert intelligence through AI agents that answer strategic questions in plain language. A Louisiana health system went live in 10 days, achieving 150% more visibility into patient encounters, 22% less leakage, and $3.8M in estimated annualized value.
Kythera Labs packages healthcare data expertise into AI agents on Databricks, enabling leaders to ask strategic questions in natural language.
The platform processes 339 billion claims to reconstruct patient journeys and deliver trustworthy answers.
Figma has unveiled some new design and coding product updates at its annual Config conference that aim to help creatives "push their ideas further" and automate tedious tasks with AI. Part of this is a reimagined canvas that's now optimized for full-stack development, bringing teams, AI agents, tools, and materials together in one place.
Reimagined canvas optimized for full-stack development, integrating teams, AI agents, tools, and materials.
Coding layers allow code editing directly within the Figma Design canvas.
A practical guide to adding memory to AI agents, covering short-term and long-term memory concepts, trace analysis, and how LangSmith's tools enable a complete memory loop for agent improvement across runs.
Memory enables agents to remember user preferences and corrections, reducing repeated instruction.
Short-term memory handles current tasks; long-term memory persists facts, preferences, and skills.
Pro- and anti-AI groups spent $24m on a congressional contest in New York, but it’s unclear to what end. The race targeted state assembly member Alex Bores, who sponsored an AI safety bill.
The Democratic primary for New York’s 12th congressional district saw over $24m in spending from tech-backed groups.
Pro-AI PACs spent $8m to oppose candidate Alex Bores, while pro-regulation groups spent $16m to support him.
SparQ Pulse is an open-source Developer Experience suite for GitHub-native teams, featuring project management, async standups, blockers tracking, team presence, action items, chat, documents, time tracking, and optional AI. It self-hosts via Docker with SQLite/PostgreSQL and uses Python, Flask, HTMX. The project has 12 stars on GitHub and is licensed under AGPL-3.0.
SparQ Pulse is the first fully open-source DevEx suite for GitHub-native teams, with modules Pulse (available), Metrics and Knowledge (coming soon).
Key features: GitHub sync, async standups with audio recording/transcription, blockers board, presence, action items (3 urgency tiers), chat/DMs, documents, people management, time & attendance, and optional AI assistant (OpenAI/Anthropic).
New York Assemblyman Alex Bores narrowly lost the Democratic primary to Micah Lasher, marking a temporary truce in a $27 million proxy war between Anthropic and OpenAI. Bores, known for his AI safety bill, was outspent but outperformed other high-profile candidates, with AI industry PACs spending $27.41 million combined.
Bores lost with 35% to Lasher's 39.1%, trailing behind other candidates like Jack Schlossberg (10.8%) and George Conway (7.1%).
AI-related super PACs spent $27.41 million in total, with $19.26 million supporting Bores and $8.15 million opposing him.
German researchers have found that medical AI models are vulnerable to membership inference attacks, potentially exposing patients' data, especially those from underrepresented groups. The study calls for better privacy standards and differential privacy techniques.
Discriminative AI models in healthcare can be tricked into revealing training data membership.
Attacks can succeed with near-perfect accuracy, particularly for individuals in underrepresented groups.
Facebook announced it's reimagining its Creator Studio tool as a standalone AI companion app to help creators grow their audiences on the social network, competing with TikTok and YouTube.
Meta launches AI companion app for Facebook creators
Redesigns Creator Studio tool into a standalone app
A developer noticed uneven AI recommendation traffic across his products and built a free tool that scans your website, simulates buyer queries, checks if AI recommends your product, compares with competitors, and provides an optimization checklist to increase recommendation rates.
Product traffic from AI chatbots varied significantly, prompting investigation.
A free no-signup tool scans websites and evaluates AI recommendation status.
Google Research reveals a counterintuitive phenomenon: even for simple factual questions, prompting LLMs to generate reasoning chains improves answer accuracy. Two mechanisms are identified: computational buffer (extra tokens provide additional computation) and factual priming (generating related facts facilitates retrieval).
Reasoning helps models recall simple facts that are otherwise unreachable, even without step-by-step reasoning.
Mechanism 1: Computational buffer — generating meaningless reasoning tokens also provides extra computation, improving recall.
NVIDIA NeMo AutoModel builds on HuggingFace Transformers v5, adding Expert Parallelism, DeepEP fused all-to-all dispatch, and TransformerEngine kernels to achieve 3.4-3.7x higher training throughput and 29-32% less GPU memory for fine-tuning MoE models, with no API changes.
NeMo AutoModel subclasses AutoModelForCausalLM, requiring only one import line change for performance gains.
On a 550B model, Expert Parallelism enables full fine-tuning across 16 nodes of H100s, where Transformers v5 runs out of memory.
The LlamaParse Platform community node (v5 and v6) is now an officially verified n8n community node. It exposes five LlamaCloud resources (Parse, Classify, Split, Extract, Retrieve) that can be used as tools in n8n AI Agents. v5 rewrote the foundation with direct HTTP calls and configurable API base URL. v6 consolidated multiple nodes into one and added index actions. The post presents three example workflows: retrievers as agent tools, a classify-extract-verify pipeline, and evaluating parsed outputs across different parsing modes.
LlamaParse Platform node provides five resources: Parse, Classify, Split, Extract, and Retrieve, all usable as AI Agent tools.
v5 replaced the SDK with direct HTTP calls, migrated Extract to V2, and made the API base URL configurable.