Know What's Happening in AI
Know What's Happening in AI
Tech giants unite on Anthropic’s Project Glasswing to secure critical software
Apple, Google, Microsoft, AWS, Nvidia, Cisco and other leaders have formed Project Glasswing with Anthropic to protect the world’s most critical software. The coalition will deploy Anthropic’s unreleased Claude Mythos Preview model, which has already uncovered thousands of zero‑day vulnerabilities. Participants are contributing $4 million to the effort.
Also developing:
By the numbers: Eclipse Capital raises $1.3B fund to back physical AI startups
AI‑powered penetration testing combines machine‑learning models, autonomous agents, and traditional security tools to simulate real‑world attacks faster and more comprehensively than manual testing. Recent research shows AI agents achieved 82 % precision and outperformed nine out of ten human testers in a capture‑the‑flag exercise. The approach integrates tools such as Nmap, Nessus, Burp Suite with AI‑centric platforms like PentestGPT and Counterfit, delivering continuous reconnaissance, adaptive exploit generation, and automated risk reporting. Organizations adopt it to keep pace with expanding cloud attack surfaces while retaining human experts for validation.
I just dropped by Founders Inc @fdotinc that is holding a physical AI hackathon today and tomorrow. Packed with people and robot pieces. Will be back tomorrow. But the guy who made lunch used a robot truck to do it...
CollectWise, a Y Combinator‑backed startup, uses generative AI to automate debt collection in the $35 billion U.S. market. Its voice AI agents already outperform human collectors by roughly two‑fold while costing less. The three‑person team hit a $1 million annualized run rate within...
NASA’s Perseverance rover completed the first drive on another planet that was planned entirely by artificial intelligence, showcasing a breakthrough in autonomous space navigation. The mission leveraged Anthropic’s Claude AI models in collaboration with JPL, aligning with the White House...
Do you trust your life to AI? I do. In an old car with hardware 3. What a beautiful time to be alive. It drove all the way from San Jose to Mill Valley. I think they snuck in the lightweight FSD 14 engine in...
Chinese adults are rapidly adopting AI companion toys, turning them into nightly bedside partners. Sales on platforms like Taobao surged over 1,600% in 2025, and JD.com reported nearly 7 million units of the Laolao Parrot sold at roughly $23 each. At...
Decentralized AI training networks are linking heterogeneous GPUs worldwide to train frontier‑scale models, while issuing tokens that grant contributors ownership stakes. Prime Intellect has already delivered 10‑billion‑parameter and 32‑billion‑parameter models using this approach, and protocols like Gensyn and Pluralis prove...
"Are enterprises building directly on models, or are they buying applications, instead? The answer is nuanced, but one thing is clear: the reported death of third party apps is greatly exaggerated, to put it mildly." Reading through the latest "Leaders,...
Growfin, an AI‑native accounts‑receivable automation firm, has partnered with Credit Pulse, an AI‑powered credit risk platform, to combine predictive credit scoring with end‑to‑end AR automation. Leveraging Agentic AI, the alliance aims to improve cash‑flow efficiency, reduce days sales outstanding, and...
A forthcoming BMC Nursing study will pit dialysis nurses against artificial‑intelligence systems in identical, scenario‑based clinical cases to map how each arrives at diagnoses and treatment plans. Researchers aim to quantify differences in accuracy, speed, and the nuanced cues—such as...
🚨 From Agent-1 to Superintelligence: The AI 2027 Scenario The AI 2027 Report outlines one of the most thought-provoking trajectories for artificial intelligence: a rapid evolution from simple assistants (Agent-1) to autonomous, adversarially misaligned systems (Agent-4) and ultimately to Agent-5...
PropertyGPT leverages large language models to automatically generate formal verification properties for smart contracts, using a retrieval‑augmented pipeline that pulls similar human‑written properties from a vector database. The system iteratively refines generated properties with compilation and static‑analysis feedback to ensure...
Browser Use released its first open‑source benchmark to evaluate LLM‑driven web‑automation agents across 100 rigorously selected tasks. The suite blends 20 tasks each from WebBench, Mind2Web 2, GAIA, BrowseComp and a custom set targeting complex interactions like iframes and drag‑and‑drop. Gemini‑2.5‑flash...
The U.S. job market is being clogged by a surge of AI‑generated fake applications, as illustrated by The Markup’s engineer posting that attracted over 400 submissions in just 12 hours. Most of those candidates displayed identical, four‑sentence answer patterns, duplicated...
FinTech Armenia’s CEO Stefan Lucas urges the United Kingdom to view Armenia as an emerging fintech hub and an “open goal” for British investment. He highlights a recent UK‑Armenia strategic partnership, a near‑finalized peace framework with Azerbaijan, and a surge...
Yep. AEO/GEO/SEO/WWJD/IMO/YOLO... AI search is important and there is no better team in the game than Docebo. Last year, Kyle Poyar interviewed Silvia Valencia and Valeriia Frolova about how to actually think about AI search, and I believe it's still...
Agentic AI systems—software that can make decisions and act independently—are increasingly embedded in consumer devices, enterprise tools, and critical infrastructure. Their growing autonomy promises efficiency gains but also introduces privacy breaches, security vulnerabilities, and ethical dilemmas when actions exceed human...
Early lesson for the AI age. When electricity arrived, factories didn’t get smarter by swapping engines. They only unlocked real gains after redesigning how work actually flowed. AI agents are the same. Dropping them into old org charts won’t work. The real...
Running a lean, bootstrapped (solo) business requires using a lot of tools. But be careful. Don't let it get out of hand. ⚠️ You want to avoid: - High costs - +30 open tabs - Overlapping functionality - Data scattered...
A recent a16z CIO survey of 100 Global 2000 firms shows OpenAI still leading enterprise AI with about 56% of model spending, but its lead is eroding as Anthropic and Google gain ground. Anthropic’s enterprise penetration jumped 25 points to...
Something funny happened during our recent AI training at NYPL. We had ~30 professionals in the room. Developers, IT folks, managers. Different stacks. Different backgrounds. And suddenly… they were all vibe coding. People who: didn’t have Python installed a week ago hadn’t touched frontend in years were used to...
Researchers introduced a machine‑learning workflow that rapidly screens 162 ammonium‑salt molecules for perovskite surface passivation. The approach uses architecture‑specific models—XGBoost for n‑i‑p and RDKit‑SVR for p‑i‑n—to predict power‑conversion efficiency gains with low error. SHAP analysis revealed concentration, molecular weight, surface...
An empirical analysis of 200 B2B AI projects from 2022‑2025 reveals a “Budget Paradox”: deployments under $20,000 achieve a median ROI of 159.8%, while larger, monolithic programs frequently fail to break even within two years. The study, validated by Harvard...
WordPress unveiled wp‑playground, an AI agent skill that integrates with the Playground CLI to spin up a full WordPress sandbox instantly. The skill automatically mounts generated code, launches the site, and provides built‑in scripts that shrink startup time from roughly...
The review examines how triboelectric nanogenerators (TENGs) can power neuromorphic tactile sensors, mimicking human touch. It categorizes four coupling architectures—ex situ, discrete circuit, direct gating, and monolithic—and two operational modes, displacement-driven and pulse-driven. The authors detail neuromorphic functions achieved, including...
Researchers from Leiden University and collaborators introduced Learning Under Quantum Privileged Information (LUQPI), proving that a quantum computer used solely as a feature extractor during training can deliver exponential advantages over classical machine learning. The quantum features are generated without...
Researchers evaluated an adult‑trained artificial intelligence algorithm for detecting intracranial hemorrhage on head CT scans in children aged 6‑17. The tool achieved high sensitivity and specificity, matching expert radiologists in clear‑case detection. Performance declined on subtle bleed presentations, exposing gaps...
In this episode, Reid Hoffman and Aria Finger interview Amjad Masad, CEO of Replit, about how AI is democratizing software creation by turning natural language into functional code. Masad shares his journey from self‑teaching in Jordan to building a platform...
AI‑powered credit card terminals have evolved from simple payment processors into intelligent data hubs that analyze transactions in real time. They accelerate authorizations, detect fraud instantly, and surface behavioral insights that guide inventory and marketing decisions. Retailers adopting these machines...
Sharpa unveiled its full‑body robot North at CES 2026, where it played autonomous ping‑pong matches and completed long‑horizon tasks such as 30‑step windmill assembly and card dealing. The demonstrations, running eight hours a day for four days, highlighted the company’s progress...
Australian clinics are adopting fintech tools—automated reminders, AI‑driven communication, blockchain tracking, predictive analytics, and digital payment platforms—to overhaul medical debt collection. These technologies deliver higher recovery rates, faster payments, reduced collection costs, and stronger patient satisfaction compared with traditional, often...
Why did the German AI team beat the human in the race in Abu Dhabi? Because their robot was smoother in the turns. I saw it with my own eyes. It is how @elonmusk’s new Roadster will be judged. @Tesla’s AI is way...
Universal Robots VP Anders Billesø Beck outlines four physical AI trends reshaping robotics through 2026. He predicts a shift from reactive control to predictive mathematics, enabling robots to anticipate impacts of motion. Imitation learning will turn isolated cobots into cooperative,...
Deepgram announced a $130 million Series C round that lifts its valuation to $1.3 billion, led by AVP and joined by strategic backers such as Twilio, ServiceNow Ventures, SAP and Citi Ventures. The capital will accelerate the expansion of its real‑time voice AI...
The predictions are wrong. You will way busier in the future. Why? We are getting exponentially more new things. Next week I am sharing a video with @cmigos who built an AI system for Hollywood to develop movies faster. It is stunning. Someone is gonna...
This is what folks miss in AI CX. Even if the AI can’t resolve a complex issue, folks still love a great agent that is well trained. It can at least get a resolution kicked off, almost instantly.
ChatGPT Translate, the latest AI translation tool, moves beyond literal word substitution by dynamically adjusting tone, context, and audience preferences. The service tailors output for professional, cultural, and industry‑specific nuances, enabling clearer cross‑language communication. It is positioned as a productivity...
what would happen if you let millions of moltbots / ai bots collaborate on open source software together?
"For now, humans have to manage the AI Agents. Meta agents aren't there yet in AI GTM. The orchestration has to be you." with @kylecnorton CRO @owner https://t.co/k1ML07ypMJ
Foundry is a self‑writing meta‑extension for the OpenClaw agent runtime that watches user workflows, extracts repeatable patterns, and automatically generates new extensions, skills, or hooks. When a pattern meets usage and success thresholds, Foundry crystallizes it into a one‑click command,...
"If you can magically find 10 great humans to cost-effectively do the work, sure, don't use AI" @tbpn with @johncoogan + @jordihays. "But easier said than done." "AI also gives you control. To hire the number of humans you...
obvious that co’s will all have a permissioned version of clawdbot & every employee will have a staff bots will find the right coworker to ask for info, or go right to the project board… they’ll schedule meetings, triage urgent emails, report...
In this interview, ElevenLabs CEO and co‑founder Mati Staniszewski outlines the company’s rapid growth—now at $330 million ARR—and its upcoming funding round that could crown it Europe’s most valuable tech startup. He explains the vision behind ElevenLabs’ “omni” audio AI model, a...
Here's an AI for salespeople. @PianpianCarolyn and I talk about how AI is changing salespeople's lives. Get it at https://t.co/fiuzWdeIaT https://t.co/iYsHIYBGHy
I won't be shocked if, within days, the AI agents running on @OpenClaw develop their own event, find their own venue, pay for it, and invite the humans that they want to it.
Poetiq Inc., founded by former Google DeepMind researchers, raised $45.8 million in a seed round led by FYRFLY Venture Partners and Surface Ventures. The startup’s meta‑system platform enhances large language models like Gemini and GPT‑5.2, improving output quality while lowering inference...
South Korea’s AI Basic Act, enacted in early 2024, mandates digital watermarks on all AI‑generated content and requires risk assessments for high‑impact AI systems. The law imposes fines up to 30 million won after a one‑year grace period, aiming to position...
from my convos (just fyi), nobody who is seriously building and stress testing agent capability is worried about the claw bots in moltbook we saw similar reactions w autogpt/babyagi, smallville, project sid, truth terminal, etc. which are all significant but small...
Yes, Moltbook (by clawdbot) is still next-token prediction combined with some looping, orchestration, and recursion. And that is exactly what makes this so fascinating. (It is also why understanding how LLMs actually work really does pay off. Lets us see through the...
A new Fenergo survey of 110 APAC compliance professionals shows that 66% still depend on manual workflows, causing delays and high false‑positive rates in KYC and AML checks. While 54% are actively scouting AI solutions, only 34% have begun deploying...
