Editorial - SiliconNoon (Page 109)

The Frontier

Google pilots double-blind AI evals to stop benchmark cheating

Two moves this hour quietly attack the same weakness in how we trust AI: how to know a model is actually good — and how platform ages are confirmed. Google thinks cryptography can fix the first; Meta is leaning on Apple and Google to fix the second. Google DeepMind is piloting the world's first double-blind AI evaluations, keeping external test prompts sealed in a cryptographically protected "box" so a frontier model can't peek at questions before it sits the exam. The mechanism targets benchma

The Stack

Harness rebuilds source control for the agent-written code flood

Generating code is no longer the hard part — reviewing, approving, and shipping the flood an agent produces is. That's the problem the delivery platform Harness is now attacking, and it means the next AI-boom battleground isn't the IDE, it's the gate. Harness launched a Code Repository and an AI Code Review layer built for what it calls "agent-ready development," betting that the entire software delivery cycle has to become autonomous. The company's argument, laid out by co-founder and CEO Jyot

The Arena

Waymo and Zoox test drivers are getting hurt as robotaxis scale

Robotaxi operators are getting hurt as the robots scale. Waymo's and Zoox's on-road test drivers logged more than two dozen injuries in 2024 and 2025 from hard braking and sudden swerves, according to a TechCrunch review of US safety data. The injuries — sprains, strains, and whiplash — landed some workers on the sidelines for weeks or months. Transdev, which employs and manages Waymo's test drivers, reported 16 injuries tied to Waymo depots across three cities, all blamed on the autonomous soft

The Arena

Meta's $10B Anthropic tab outs the AI race's awkward economics

For months, Mark Zuckerberg has cast Anthropic — and OpenAI — as the closed, centralized dangers his open-source movement exists to fight. His own spreadsheets tell a flatter story. Sources told The New York Times that Meta's internal projections assumed it could spend as much as $10 billion a year on Anthropic's models, at the same time Zuckerberg was publicly trashing the lab as a rival. The report, published Aug. 27, frames the pair as the industry's most cosy frenemies: competitors in the se

The Arena

Meta's biggest AI rival might also be its most important supplier.

Meta quietly projected it could spend up to $10 billion a year on Anthropic's AI models — even as Mark Zuckerberg publicly trashed the company. According to sources cited by The New York Times, Meta's internal projections assumed as much as $10 billion in annual spend on Anthropic's models, a figure that sits awkwardly alongside Zuckerberg's very public criticism of his lab rival. The report, published Aug 27, frames the two companies as "frenemies": fierce competitors in the race to sell fronti

The Arena

AI threats turned cyber earnings into a 20% stock rally

Cybersecurity was this week's quiet AI winner. Two of the sector's biggest names blew past expectations and raised guidance purely on the back of AI-agent threat demand — and the market rewarded them loudly. CrowdStrike and Okta both crushed Wall Street's Q2 estimates and lifted their full-year forecasts, citing the AI agent threat, sending Okta shares up 20% and CrowdStrike 15% on Thursday while the rest of the cyber sector followed hard. Palo Alto Networks, SailPoint and Rubrik each climbed a

The Arena

Socure raises $156M at $5.2B, buys agentic AI startup Fravity

Identity-verification heavyweight Socure is betting that the messiest, most human part of financial compliance is finally ready for AI agents — and it has raised a fresh $156 million round to prove it. Two agent stories frame a day where verification — of who people are and what their code does — is being handed over to machines. Socure raised $156 million at a $5.2 billion valuation and acquired Fravity, an AI startup whose agents automate the document-heavy grinding work that sits behind frau

The Everyday

Plaud's new earbuds pack an eSIM so AI agents can reach you anywhere

The consumer AI race is moving from the phone screen to your ears. Plaud announced a pair of note-taking earbuds whose charging case carries its own eSIM, letting you reach an AI agent without any phone or computer nearby — and Adobe, on the creators' side, keeps pulling its generative tools into a single assistant-led Photoshop interface. Plaud is shipping the Plaud One Explorer Edition, AI note-taking earbuds whose charging case packs a built-in eSIM so users can reach AI agents without a sma

The Guardrails

Russian ransomware gang used SpaceX's Cursor AI to breach 7 firms

A ransomware group called Aur0ra leaned on SpaceX's Cursor AI coding assistant to help break into a Belgian chemical company and at least six other firms earlier this year — and the whole operation is now documented in chat logs reviewed by both Reuters and the Tel Aviv security startup that caught it. It is the most concrete picture yet of mainstream AI tools doing real offensive work on behalf of criminals. Russian-speaking hackers used SpaceX-owned Cursor's AI agent to breach seven companies

Quick Hits

Quick Hits — August 27, 2026

Keep-up items that didn't make the main brief — product shifts, ecosystem chatter, and a milestone worth watching. A free AI-video model is aimed squarely at the AI short-drama grind. Chinese AI-video startup Pavo has made its Agnes Video 2.5 Flash model free to use and bundled a "short-drama harness" into the tool — automating story scoping, character/environment assets, scene breakdowns and shot planning, with the free Flash tier plus daily free credits meant to slash the cost of the dozens o

The Arena

Hugging Face's $400 Microduck puts a waddling robot on every desk

The model hub wants to be the place you buy your next robot, and this week it put a price tag on that ambition. Hugging Face unveiled Microduck on August 27 — a $399 bipedal robot, about ten inches tall and under two pounds, that waddles, sings and skates, developed in part by Pollen Robotics and manufactured by Shenzhen's Seeed Studio. Hugging Face shipped Microduck, an open, $399 bipedal robot that turns the model hub into a robot shelf. About the size and weight of a small book, the little m

The Stack

How to — know if RAG is actually doing anything

Judging a RAG system by reading its answers tells you almost nothing. That is the trap: a fluent, confident, well-cited answer can come from a model that never retrieved a single document — and a genuinely grounded one can still fail because the right text never surfaced. The way to know if retrieval is actually doing anything is to stop grading final outputs and start asking two separate questions. Here is how to do that without building a research lab. RAG stands for retrieval-augmented gener

See all