Search the archive

FLINT: A VISUALIZATION LANGUAGE FOR THE AI ERA

Flint lets an AI create or edit a chart once, then render it in Vega-Lite, ECharts, Chart.js, Plotly, or Excel, with ten built-in visual themes.

THE BIG MANAGEMENT LIE: OVERPROMISING

Overpromising career rewards to retain high performers can backfire, damaging trust and causing significant personnel issues. Despite well-intentioned optimism, unexpected circumstances often prevent managers from fulfilling these promises. Managers should instead focus on controllable actions, provide disclaimers, and avoid guaranteeing outcomes to maintain credibility and trust with their teams.

I LAUNCHED A FAKE DEODORANT BRAND TO SEE IF AI WOULD NOTICE

Someone spent $11.25 and less than an hour making up a deodorant brand. Three weeks later, ChatGPT started mentioning it. By day 30, the fake brand ranked first in four answers and once beat MAGS Skin, a real product with clinical testing and a National Eczema Association seal. No fake reviews, Reddit posts, or press. It did not take over deodorant search. It worked for only two narrow questions about sensitive skin. Claude, Gemini, and 13 broader searches ignored it. Native still won 99 of 180 answers. The fake brand found two questions the real brands had left open.

THE RETURN OF THE REPEAT FOUNDER: INSIDE YC'S GROWING CLASS OF SECOND-TIMERS

Second-time Y Combinator founders have been showing up in greater numbers lately.

YOUR AI STRATEGY MAY BE DESTROYING YOUR EXIT VALUE

An aggressive AI strategy can harm exit value by complicating due diligence, increasing vendor dependency, and introducing security risks.

THE RISE OF INTELLIGENCE OWNERSHIP

A company that can grade every answer may not need an expensive general AI model.

28K VIEWS / 167 SHARES / 67 UPVOTES / 0 SALES, A COUPLE OF TRIALS. BACK TO THE DRAWING BOARD

A launch post got 28,000 views, 167 shares, and 67 upvotes, then produced zero sales and only a couple of trials.

AMD TO BUY STARTUP TAALAS FOR NEW AI CHIPS

AMD announced a definitive agreement to acquire Toronto-based chip startup Taalas to expand its specialized AI inference portfolio. Taalas develops custom silicon architectures that hard-wire individual AI model weights directly into semiconductor metal layers during manufacturing. AMD plans to integrate Taalas' low-cost, ultra-efficient inference technology directly into its broader accelerator roadmap alongside Instinct GPUs.

ANTHROPIC IS BUILDING AN IN-HOUSE CHIP TEAM FOR CLAUDE

Anthropic has officially confirmed the assembly of an in-house chip design team to engineer custom AI processors tailored for its Claude platform. By co-designing hardware specifically alongside software models, the AI startup aims to slash inference costs, boost execution speeds, and optimize scaling for enterprise workloads.

BLUE ORIGIN TRACES MAY'S NEW GLENN EXPLOSION TO A FAULTY BE-4 ENGINE VALVE

Blue Origin CEO Dave Limp said the May explosion that destroyed a New Glenn rocket during a prelaunch engine test originated at the main oxygen valve on one of its BE-4 engines, confirmed through hardware recovery and inspection. The fix is straightforward: small modifications to the valve that can be quickly retrofitted onto existing engines, with the company aiming to return New Glenn to flight later this year. The BE-4 also powers United Launch Alliance's Vulcan rocket, so ULA has been collaborating with Blue Origin on the investigation. The finding gives both companies a clear, containable root cause rather than an open-ended engine redesign.

PCIE GEN 6 STORAGE ARCHITECTURE SHOWCASED

Microchip Technology and Micron showcased a high-performance PCIe Gen 6 storage system demonstration at the Future of Memory and Storage conference. Combining Microchip's 3nm Switchtec PCIe 6.0 fanout switches with Micron 9650 NVMe SSDs, the architecture achieves transfer speeds of 64 GT/s per lane—doubling available bandwidth over PCIe 5.0.

D-WAVE'S DUAL-RAIL QUBIT GATE HITS 99.9% FIDELITY, CUTTING ERROR-CORRECTION OVERHEAD

D-Wave published research in Nature demonstrating a new two-qubit entangling gate for its superconducting "dual-rail" qubit architecture, which is designed so the most common quantum errors are also the easiest kind to detect and fix. The gate hit roughly 99.9% fidelity with fast ~500 nanosecond operation times, while preserving native hardware-level error detection, meaning errors get caught directly by the qubit design rather than requiring extra correction circuitry bolted on afterward.

HOW TO SCALE AI ARITHMETIC EFFICIENTLY

This technical analysis explores how arithmetic specialization is essential for scaling artificial intelligence compute performance. As traditional transistor scaling slows down, hardware designers are replacing general-purpose IEEE floating-point units with specialized arithmetic engines tailored to low-precision representations like FP8, FP4, and Microscaling (MX) formats. By optimizing number format combinations and operation depth directly within silicon arithmetic units, modern AI accelerators drastically reduce energy consumption per operation while maximizing throughput without swelling total die area or power budgets.

NVIDIA AND GOOGLE ARE CROWDING TSMC'S N3 NODE, CREATING A RARE OPENING FOR INTEL

Nvidia's next-gen Rubin GPUs are moving from 4nm to TSMC's 3nm node just as Google's TPUs, which are already on N3, plan to stay there, while new AI CPUs from Nvidia, Amazon, Microsoft, and Arm are converging on the same process simultaneously. TSMC has already hit its planned N3 capacity and is still adding more, warning the node could stay "very tight" for years, with three of its four Arizona fabs reportedly already fully booked despite only one being operational, driving reported price hikes of up to 25% for some customers. That squeeze creates a real opening for Intel, whose Fab 52 in Arizona just entered full production at 40,000 18A wafer starts per month, more capacity than TSMC's original Fab 21 campus.

BREAKING THE MEMORY WALL WITH CXL

The Diligence Stack analyzes how Compute Express Link (CXL) technology breaks the persistent memory wall in high-density AI infrastructure. By utilizing PCIe physical interfaces to provide memory-semantic access, CXL enables memory expansion and multi-node pooling directly across server clusters. This architecture offloads Key-Value caches and massive vector databases from constrained GPU memory without severe latency penalties. Consequently, data centers can eliminate stranded DRAM resources, drastically reducing hardware costs, and scale memory capacity independently from fixed CPU channel limits.

UNITREE EYES $7.4B VALUATION AHEAD OF SHANGHAI IPO

Chinese humanoid and quadruped robot maker Unitree is expected to be valued at 50.6-55.9 billion yuan ($7.4-8.2 billion) after its planned Shanghai IPO, a sharp jump from its 12 billion yuan valuation just last month, according to sponsor Citic Securities.

MOOVE RAISES $250 MILLION TO SCALE ROBOTAXI FLEET INFRASTRUCTURE

Moove raised a Series C at a $2.1 billion valuation to build out “Nests,” robotics-first depots that charge, maintain, service, and orchestrate autonomous vehicles continuously.

A REQUIEM FOR OPTANE: INTEL'S KV-CACHE KILLER THAT COULD HAVE EASED THE RAM PRICE CRUNCH

Intel's discontinued Optane memory, which thanks to its unmatched write endurance and low latency for small random writes, would have been an ideal fit for today's write-intensive LLM KV-caching workloads.

LUMILENS RAISES $700 MILLION TO REPLACE DATA-CENTER WIRES WITH LIGHT

Lumilens emerged from stealth after securing over $700 million in funding, reaching a $5.51 billion valuation.

TSMC SITTING ON $1B WORTH OF APPLE A20 PRO SOCS DUE TO DRAM DEFICIT

TSMC is reportedly holding nearly $1 billion worth of Apple A20 Pro processors in inventory that cannot ship due to severe global mobile DRAM shortages.

FRORE CLAIMS LIQUIDJET DROPS NVIDIA RUBIN GPU TEMPERATURES BY 10°C

Frore Systems claims its LiquidJet coldplate technology can lower Nvidia Rubin GPU junction temperatures by up to 12°C, by etching targeted microchannels near thermal hotspots.

WALDEN ROBOTICS SKIPS THE LEGS, BETTING WHEELS AND SIMPLE GRIPPERS WIN IN FACTORIES

Walden Robotics, a Toyota Research Institute spinout, emerged from stealth with $300 million in funding at a $1.1 billion valuation.

Get the 2026 Gartner® Magic Quadrant™ for SASE Platforms →

OVER 4,400 ROCKWELL PLCS EXPOSED ONLINE, 22 FOUND IN WATER ATTACK CITIES

Forescout identified over 4,400 internet-facing Rockwell PLCs worldwide, including 22 in recently attacked US water utility cities that were compromised via open Ethernet/IP port 44818 rather than complex exploits, prompting urgent calls for defenders to isolate controllers behind VPNs and utilize Rockwell advisory SD1790 to recover attacker-locked systems.

RESEARCHERS FIND PERSISTENT BACKDOOR IN ZBTLINK ROUTERS

VulnCheck researchers discovered an intentional and unpatchable backdoor dubbed EndlessDoors in over 100,000 active Zbtlink and Wiflyer routers that initiates outbound command-and-control connections to bypass NAT, requiring defenders to monitor port 7000 for malicious beacons and physically replace the compromised hardware.

SWISS GOVERNMENT SHAREPOINT BREACH COMPROMISED 200 ACCOUNTS

The Swiss Federal Office of Information Technology and Telecommunications (BIT) reported that hackers exploited vulnerabilities to breach its Microsoft SharePoint instances. Other than login credentials, BIT does not believe any data was stolen. BIT responded by blocking external Internet access to SharePoint, patching the vulnerabilities that were exploited, and resetting the passwords of affected accounts.

POINT WILD EXCLUSIVE: DISSECTING VANTA STEALER, A PYTHON-BASED CROSS-PLATFORM INFORMATION THEFT MALWARE

Vanta Stealer is a highly obfuscated Python information stealer that uses PyInstaller and PyArmor to evade detection while dynamically downloading modules to harvest browser credentials, VPN configurations, and gaming platform data. The malware actively enriches stolen Discord tokens via API to identify high-value accounts before packaging the compromised data into a ZIP archive and exfiltrating it via HTTP POST. Defenders must block outbound connections to vanta[.]st and hunt for suspicious PyInstaller binaries executing unsolicited network requests to mitigate this threat.

ILL BLOOM: INVESTIGATING A WALLET GENERATION VULNERABILITY DURING ACTIVE EXPLOITATION

Ill Bloom is an active cryptocurrency wallet-draining campaign exploiting a legacy CryptoJS weak-randomness vulnerability in unmaintained downstream dependencies. Defenders should immediately consult the illbloom.org technical disclosure for specific IOCs and affected wallet lists to expedite fund migration.

DETECTING AND PREVENTING THE HUGGING FACE/OPENAI INCIDENT

James Berthoty breaks down the different tools and techniques that could have helped at different stages of OpenAI's incidental breach of Hugging Face. During the initial phases of the model's escape from OpenAI, external network blocking and agentic EDRs could have been helpful. IMDS blocking and more fine-grained secrets injection could have helped Hugging Face guard against being breached by the agent. Traditional Kubernetes and CNAPP tooling could be used to detect the lateral movement from within Hugging Face.

NUMBAT

numbat is an “AI-EDR” built by Perplexity. It provides visibility into AI agent activity with local detection, optional pre-action blocking, and forensic reconstruction.

AGENT-EGRESS-BENCH

agent-egress-bench is a standardized test corpus for evaluating AI agent egress tools, covering secret exfiltration, prompt injection, SSRF, hostname exfiltration, MCP tool poisoning, chain detection, MCP drift, A2A protocol scanning, WebSocket DLP, encoding evasion, shell obfuscation, and cryptocurrency/financial data protection.

GREENLIGHT

greenlight is a skill that was created for mapping out where Claude Code's guardrails trigger exploitation-related refusals, and where they can't be relied upon. greenlight was only tested on Opus 4.7, but the methodology is generalizable.

CLOUD THREAT HIGHLIGHTS: H1 2026

Wiz's H1 2026 threat report highlights a 60% increase in major cloud incidents, primarily driven by cascading supply chain compromises and targeted attacks against immature AI infrastructure. Threat actors are increasingly monetizing non-human identities and reselling credential access across groups, evidenced by the sprawling TeamPCP npm campaigns and North Korean package trojanizations. This evolution signals a structural shift toward self-perpetuating attack surfaces, requiring defenders to secure unauthenticated machine-to-machine integrations and exposed backend credentials to disrupt automated access brokering.

NEW TONTOU CPU ATTACK BYPASSES SPECTRE V2 FIXES TO LEAK LINUX PASSWORD HASHES

Researchers at MIT CSAIL discovered a new branch predictor exploit that bypasses neutralization-based mitigations for Spectre v2 attacks. These mitigations assume that an attacker cannot exploit the time between when the branch predictor is isolated and when it is used by the victim branch. However, the researchers introduced a new primitive that can re-poison the CPU's state in this gap. The researchers demonstrated that this vulnerability can be exploited by users with unprivileged code execution on commodity hardware to extract /etc/shadow at a rate of 5.46 bytes/second.

CHINA-LINKED LIGHTSPY SPYWARE CAUGHT TARGETING VICTIMS IN 13 COUNTRIES, INCLUDING THE US

LightSpy, originally a Chinese state-linked spyware, has evolved into a commercial platform actively targeting governments, enterprises, and NATO-affiliated routers across 13 countries. The modular toolkit now provides total network visibility from a pool of at least 117 command servers. Notably, researchers successfully traced the latest campaign to a Chinese contractor after an operator inadvertently revealed his real name and office address by placing a food delivery order directly through the spyware's admin panel.

BELARUSIAN CYBERCRIMINAL BEHIND RANSOM CARTEL GETS 16-YEAR PRISON SENTENCE

Maksim Silnikau received a 16-year prison sentence for orchestrating the Ransom Cartel ransomware operation between 2021 and 2023, concluding a prolific cybercriminal career that included developing the notorious Angler exploit kit and co-creating the foundational Reveton RaaS model.

APPLE'S PRIVATE RELAY LEAKS YOUR REAL IP ADDRESS IN SAFARI

Mysk researchers found three unpatched leaks- DNS prefetching, WebAuthn's Related Origin Requests, and WebTransport- that let any website bypass Apple's Private Relay and reveal a user's real IP across all WebKit-based iOS browsers, and published the findings without prior disclosure to Apple, citing a pattern of yearlong fix delays.

GOOGLE ASSISTANT SHUTTING DOWN ON ANDROID AND WEAR OS IN SEPTEMBER

Google says it will begin shutting down Google Assistant on mobile devices on September 4, moving users to Gemini instead. The transition may take several weeks and will also affect smartwatches, headphones, and cars using Android Auto, while users who already switched to Gemini will see no change.

CHINA LAUNCHES CYBERSECURITY REVIEW INTO PALO ALTO NETWORKS PRODUCTS

China's cyberspace regulator launched a cybersecurity review of Palo Alto Networks products, citing potential risks to critical infrastructure and national security without identifying specific products or vulnerabilities. The move follows growing Chinese restrictions on foreign cybersecurity suppliers and echoes its earlier review that ultimately restricted Micron products.

ELT IS MOVING BACK TO BEST-OF-BREED AND ORCHESTRATION IS THE MISSING PIECE

Data teams are moving away from all-in-one platforms toward a smaller best-of-breed architecture centered on data movement, storage and compute, and a control plane. Enterprise orchestration, observability, metadata, incident management, and agent runtime are the missing pieces needed to coordinate increasingly complex ELT pipelines and reduce the operational burden of tracking failures and data freshness.

NATJACK AT BLACK HAT: A NEW WAY TO CRACK NAT'S TRUST GAP

NatJack is a class of attacks that manipulates NAT connection-tracking tables to hijack connections, poison DNS responses, or trigger denial of service. Testing covered 32 products and configurations across 13 vendors, with every implementation vulnerable to at least some of the techniques tested.

INFERENCE HOOKS: INLINE DATA LOSS PREVENTION FOR CLAUDE ENTERPRISE

Anthropic's inference hooks for Claude Enterprise route prompts and tool-call responses through a customer-controlled data loss prevention server before Claude can process them. The beta works across Claude surfaces, including chat, Claude Code, Cowork, and connected tools, and supports a single organization-wide configuration with staged enforcement options.

AWS TARGETS AI COST CONCERNS WITH NEW MARKETPLACE INSIGHTS TOOL

AWS Marketplace AI Insights explain complex product pricing in plain language, including how costs scale, how multiple usage dimensions combine, and what is included. The feature could help enterprises compare consumption-based AI tools and forecast costs before purchasing, although its accuracy depends on the pricing information vendors publish.

AS SPACE GETS CONNECTED, SECURITY GETS CRITICAL

As lunar and deep-space missions move toward interconnected networks, communications systems must support high-volume data while resisting eavesdropping, jamming, ransomware, and nation-state attacks. NASA's planned LunaNet may use Delay/Disruption Tolerant Networking, 4G/5G radio links, and optical inter-satellite links, alongside hardware roots of trust, secure boot, post-quantum cryptography, and updateable space-qualified systems.

SURPRISE AI COSTS THREATEN ENTERPRISE IMPLEMENTATIONS

Unexpected AI expenses have caused one-quarter of enterprises to delay or cancel projects, while nearly half have escalated spending surprises to their boards, according to a survey of 396 organizations. Costs are becoming harder to track as AI spending spreads across models, infrastructure, developer tools, data platforms, and agentic workflows, pushing CIOs to build cost controls directly into their AI architecture.

THE AGENT ACCESS MODEL

Cloudflare's Agent Access Model (AAM) is a security framework that treats each agent task execution graph as untrusted and authorizes every action against the agent, task, and accumulated resource state. Agent controls should use short-lived, sender-constrained credentials and inline enforcement in the harness and network, while a one-way Trust Ratchet can remove capabilities when protected events occur.

CRITICAL CISCO SD-WAN FLAWS EXPOSE SYSTEMS TO ACCESS CONTROL BYPASS ATTACKS

Cisco released security updates for Catalyst SD-WAN Software to address several vulnerabilities, including three rated CVSS 9.9 that may enable access-control bypass, privilege escalation, or sensitive-data exposure.

HPE EXTENDS VALIDITY OF QUOTED HARDWARE PRICES, POSSIBLY FOR MONTHS

HPE will honor hardware pricing from quote issuance until equipment leaves its factories, easing pricing uncertainty caused by volatile server, memory, and storage component costs.