This Week in Tech 131

What we explore this week:

Polymarket flagged the news: Anthropic has accused Alibaba of using nearly 25,000 fraudulent accounts to systematically query Claude and reverse-engineer its capabilities. The attack vector isn't hacking model weights, it's asking the model enough targeted questions to reconstruct what it knows. Anthropic filed a lawsuit, and Wired confirmed the scale of the alleged operation.

Frontier model behavior is now a strategic asset worth committing large-scale fraud to obtain, which tells you exactly how wide the capability gap feels from the outside.

The week Anthropic was defending Claude from alleged foreign espionage was also the week they shipped some of their most ambitious products yet. Claude AI's account announced Claude Sonnet 5 — autonomous planning, browser use, terminal access — capabilities that required flagship-tier models just months ago, now running in a mid-tier model. Separately, they launched Claude Science, a dedicated research app with 60-plus scientific databases, traceable artifacts, and on-demand environment management — not a chatbot wrapper, a genuine lab partner.

On the social side, Anthropic joined RAISE US as a founding partner, a nonprofit coalition focused on employer-led AI retraining before displacement peaks. The combination of defend, ship, and invest in the aftermath is a company that knows it's building something that will break things and is trying to hold all three responsibilities at once. The pressure on competitors running cheaper Chinese alternatives just got harder to justify.

Testing Catalog surfaced the announcement: OpenAI has designed its first custom AI chip in partnership with Broadcom, claiming a new state-of-the-art in performance per watt, with deployment at gigawatt scale planned across multiple generations. The Verge confirmed the chip is manufactured at TSMC. OpenAI used its own models to accelerate development, a nice piece of recursion. Bloomberg framed it as a direct challenge to Nvidia. This is the Apple and Google playbook: control your compute, control your costs, control your roadmap. Nvidia's moat just got narrower.

Every missed onboarding step, delayed offboard, or broken provisioning handoff has a root cause: HR and IT aren't aligned. This guide gives both teams a shared framework for the full employee lifecycle.

OpenAI announced a three-tier model family under the GPT-5.6 banner: Sol for frontier tasks, Terra for everyday balance, Luna for cheap high-volume work. This is explicit market segmentation where AI is priced and packaged like cloud compute tiers. You will soon choose your model the way you pick a data plan, and that framing will reshape how enterprises think about AI budgets entirely.

Anatoli Kopadze shared a key detail from Spotify's Chief Architect: adding loops to their AI agent workflows pushed task success rates from 20-30% to 80%, across 20 million lines of code. Seventy-three percent of Spotify's codebase is now AI-assisted. Most engineering orgs are still treating reliability as a model problem. Spotify treated it as a workflow problem and solved it.

Boris Cherny, who works on Claude Code at Anthropic, posted that traditional roles like engineer, designer, and PM are dissolving and what's emerging isn't a new job title but a set of archetypes, starting with the Prototyper. Hybrid roles are already reshaping tech teams. The practical question for your career: which archetype are you, and does your team have all five?

NotebookLM announced 60-second vertical video overviews that break down complex source material for mobile. Instead of fighting short attention spans, they're surfing them. Rolling out to Google AI Ultra and Pro subscribers now, with free users coming soon. If the audio overview format already hooked millions, video could make NotebookLM a genuine daily habit.

Reid Hannaford shared a workflow that chains Midjourney for style frames, Blender for motion blocking with basic shapes, and Seedance to render it into photorealistic video. Anyone who can move a cube can now direct a convincing action sequence. Hollywood pre-vis just became a solo weekend skill.

CMS Invests laid out the numbers: X Money is offering 6% APY, up to $10M in FDIC insurance through partner banks, unlimited 3% cash back, and a metal Visa card. The Verge confirmed the product details, and TechCrunch confirmed the Visa partnership. This is a 600-million-user social platform trying to eat traditional banking's lunch. The moat banks have relied on for decades is access and inertia. X is betting that if you already spend hours daily in the app, moving your money there is a small step. Bloomberg notes the regulatory hurdles are real, but the product terms are hard to argue with on paper.

A Google product lead shared portfolio tracking and a new Android app, and Google's official account confirmed the full global rollout out of beta, with an iOS app coming. The screenshot-to-portfolio import experience of uploading a screenshot from Robinhood or Yahoo Finance and Google parses it automatically is the kind of frictionless feature that actually moves people off entrenched products. 9to5Google has the full breakdown. Google is finally treating Finance like a product it intends to win with.

Two posts covered this: Meta AI's official account and lead researcher Jean-Rémi King. Meta published Brain2Qwerty in Nature: a non-invasive brain-to-text decoder that turns raw brain signals into typed sentences in real time using MEG and EEG headsets, no surgery required. The Nature paper covers v1; v2 is already publicly released and claims the highest-performing end-to-end pipeline for real-time sentence decoding from non-invasive recordings.

Prior BCI breakthroughs required implanted electrodes. If Meta can do this with a headset, the gap between Neuralink-style implants and consumer brain interfaces just compressed dramatically. The words you think could reach a screen before your hands do — sooner than the sci-fi version of that sentence suggests.

Aleph Neuro announced the highest-resolution 3D brain images ever captured from outside the skull. The biggest bottleneck in brain-computer interfaces has always been signal quality without implants. If Aleph can match or approach implant-grade fidelity non-invasively, the commercial path for neural interfaces gets dramatically shorter. Between Aleph and Meta's Brain2Qwerty, non-invasive neurotech is having a legitimately consequential week.

Bryan Johnson posted that he's renamed Blueprint to Immortals, staking a public claim that biological death can be defeated in our lifetime. Whether you find it visionary or delusional, the data behind Blueprint is more rigorous than most wellness companies produce. The rebrand separates serious longevity science from supplement marketing and forces the field to take a clearer position.

Longevity escape velocity is either the most important concept of the century or the most expensive cope. Johnson is forcing that argument into the open.

Polymarket flagged Secretary Hegseth's appointment of Marc Andreessen to the Pentagon's new Defense Policy Board. The New York Times has tracked Andreessen's political alignment in detail. The people funding AI startups are now formally advising on national security strategy. The feedback loop between Silicon Valley and the Pentagon is no longer informal. What gets built next, and who it gets built for, will be shaped by that room.

Stock Market Newz reported the scale: VW is preparing to eliminate roughly 100,000 jobs — about 15% of its global workforce — and close plants in Hanover and Zwickau, among others. Reuters confirmed the restructuring scope, and the FT tracked the union pushback. This is the most radical overhaul in VW's 89-year history, and it's a direct consequence of miscalculating the pace and economics of the EV transition.

When Europe's largest automaker needs emergency surgery, it signals that the old manufacturing model is collapsing faster than the policy roadmaps assumed. The workers absorbing this weren't the ones who made the strategic bets.

Space and Tech highlighted Shield AI's X-BAT, a runway-free, jet-powered aircraft running Hivemind AI which is the same software already proven in contested combat environments. It flies to 50,000 feet, has a 2,000 nautical mile range, and operates without GPS, comms, or a human in the loop. The military application is obvious.

The longer arc is what autonomous aviation looks like when this hardware and software matures into civilian contexts.

The story of this week isn't any single product. It's the shape of a war being fought on three fronts simultaneously. Anthropic is allegedly being probed by 25,000 fake Alibaba-linked accounts trying to extract what Claude knows, while Anthropic itself ships Sonnet 5 and Claude Science and joins a workforce retraining coalition. That combination, defend, advance, and prepare for the social damage, is a rare posture for a tech company. Most just ship and apologize later.

The Chinese AI story cuts deeper than the lawsuit. Companies switching to Chinese models because they're cheaper isn't just a procurement decision; it's a signal about where the perceived frontier is moving. The fact that Anthropic responded with a mid-tier model that now does what only flagships could do six months ago suggests they understand the pressure. Capability at lower cost is the only argument that beats "this one is free." Claude Sonnet 5 is that argument, made in code.

X Money is the week's most underrated story. A 6% APY with $10M FDIC coverage and unlimited 3% cash back isn't a fintech experiment; it's a direct assault on the terms every traditional bank has spent decades treating as untouchable. Banks survived the first wave of fintech because apps don't have 600 million daily users with social graph lock-in. X does. The question isn't whether X Money's terms are real. It's whether regulators let them stay real long enough to matter.

And then there's the brain. Meta published non-invasive thought-to-text in Nature. Aleph captured the highest-resolution non-invasive brain images ever. In the same week. The scenario in Her where the interface between human and machine becomes frictionless and intimate is no longer a design fiction, it's a research milestone with a publication date. The decade that follows those two papers will be stranger than most people are prepared for.

HubSpot switched to Mintlify and saw 3x faster builds with 50% fewer eng resources. Beautiful, AI-native documentation that scales with your product — no custom infrastructure required.

Simplify Your Docs Today

ChatGPT is a superpower if you know how to use it correctly.

Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.

Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.

Download the free guide