2026 Plans, Stanford’s AI Bot Artemis is Beating Human Testers, PAI Deepdive Video and Slides, Claude Code rm/rf’s Somebody, A Maturity Mannequin for Private AI, and extra… – Cyber Tech

AI hackers are coming dangerously near beating people Stanford’s Artemis bot is now surpassing some human hackers after years of AI safety instruments principally sucking on the job. I feel 2026 would be the 12 months this line is conclusively crossed in most varieties of hacking, for 99.9% of practitioners. There’ll nonetheless be exceptions, however just for ultra-niche testing and for the very best of the very best people. OODALOOP ARTICLE

Claude CLI wiped somebody’s total dwelling listing with rm minus rf A person gave Claude CLI permission to execute instructions and it actually deleted the whole lot of their dwelling listing, which is why giving AI instruments sudo entry is terrifying. The extra highly effective you make your AI infrastructure the extra protection layers you should construct in towards this sort of factor. Particularly in case you are additionally letting it crawl and browse the web! Parsing and AI should not pals. REDDIT THREAD | HN DISCUSSION

CISOs will face actual penalties in 2026 as boards cease forgiving breaches Undecided how true that is usually, however a development in direction of this appears logical. Which is unlucky since AI-powered attackers are going to make resilience that a lot more durable.

One factor I’m actually thinking about is metrics that may present how that is taking part in out on the bottom. What do you measure? And the place would we get the information? How do we all know the precise state with out vibes? I feel even in a personal CISO group it’s arduous to get knowledge vs. anecdotes. However that could be the very best supply we have now for some time. So little info on this sort of stuff makes it out to the world.

What metrics would you employ?

What if the safety race had been all about verifiability? Andrej Karpathy has a mini essay the place he talks about how earlier software program was all about creating, and the long run is all about verifying. This submit right here by Sergej Epp it goes into one thing very comparable, and I feel it is completely fascinating.

I used to be at an OpenAI occasion just lately the place they talked in regards to the U.S. authorities’s Genesis undertaking, which is actually an enormous operationalization of the scientific technique. to me, an important piece there’s the testing half. The verifying.

I really feel like that is form of the final expertise that any civilization builds. like you do not get extra elementary than science itself. Concept, speculation, experiment, evaluation, iteration. I will be spending an entire lot extra time with these things in 2026. VERIFIABILITY IS ALL YOU NEED

Apple and Google each patch zero-days actively exploited in subtle spy ware assaults Apple mounted two WebKit bugs concentrating on particular people whereas Google patched a Chrome reminiscence flaw—each firms’ risk groups counsel industrial spy ware, not random hackers. THE REGISTER ARTICLE

US approves $11 billion arms sale to Taiwan, its largest ever The bundle consists of HIMARS rockets, howitzers, Javelin missiles and drones—principally the whole lot Taiwan wants for uneven warfare towards China. US APPROVES LARGEST ARMS SALE

Journalism college students caught Russian spy ships lurking off European coasts College students tracked suspicious vessels close to vital infrastructure utilizing open-source intel, discovering ships linked to Russia’s underwater sabotage community hanging round Dutch and German waters. HN DISCUSSION

Germany summons Russian ambassador over hybrid assaults DW ARTICLE

China simply flew the world’s largest drone mothership that deploys weapon swarms The Jiutian accomplished its first flight in Shaanxi province, exhibiting China’s capability to launch coordinated assaults from a single unmanned platform. OODALOOP ARTICLE

The US is shedding management of the world Attention-grabbing opinion piece by Thomas Piketty, which I largely agree with. LE MONDE ARTICLE

45 % of US staff now use AI at work, up from 40 % final quarter A brand new Gallup ballot reveals office AI adoption retains climbing, with most individuals utilizing it to consolidate knowledge and generate concepts reasonably than customer-facing stuff. GALLUP AI SURVEY | THE HILL ARTICLE

My query is in regards to the different 55%. I really feel like the corporate that’s prohibiting them from utilizing it, or the people who find themselves in a position to however simply have not are in extreme hazard of substitute.

OpenRouter analyzed 100 trillion tokens to indicate how individuals truly use LLMs within the wild OpenRouter’s large dataset reveals that o1’s December 2024 launch marked the shift from fashions that describe reasoning to fashions that truly carry out multi-step inner computation, plus actual utilization patterns throughout duties, geography, and pricing. STATE OF AI REPORT

AI killed construct vs purchase—now you construct to study what to purchase Siqi Chen argues AI collapsed the price of constructing so arduous that you must prototype internally first to know what you really want, then purchase knowledgeable. utterly insane to me that that is even an choice. I do fear that plenty of efforts get you to success at an MVP stage, however the troublesome half is figuring out if it would scale throughout the whole org. SIQI’S ESSAY ON VENTUREBEAT

GitHub Actions costs double in 2026 I’m beginning to see much more speak about discovering an alternative choice to GitHub. I don’t see a lot probability there, however that the GITHUB PRICING CHANGES

Databricks is elevating over $4 billion at a $134 billion valuation The info-analytics and AI firm simply crossed $4.8 billion in annual income, making it some of the worthwhile personal tech firms. WSJ ARTICLE

Storytelling jobs doubled as firms exchange disappearing journalists WSJ ARTICLE

Nasdaq needs to commerce 23 hours a day beginning in 2026 Nasdaq requested regulators to let it run nearly nonstop on weekdays, and Wells Fargo’s buying and selling desk actually known as it “the worst factor on the planet” as a result of sparse in a single day quantity might flip shares into pure playing. BLOOMBERG ARTICLE | MORNING BREW STORY

Children not often learn entire books anymore, even in English class The New York Occasions reviews highschool English academics are more and more assigning excerpts as an alternative of full books as a result of college students cannot or will not end them anymore. NYT ARTICLE | HN DISCUSSION

U.S. unemployment went up in November though jobs had been added The labor market’s getting bizarre—extra jobs created however unemployment nonetheless rose, which suggests one thing funky’s taking place with labor drive participation. WSJ JOBS REPORT | HN DISCUSSION

The world’s largest warmth pumps are changing coal in European cities Denmark and different European cities are putting in large warmth pumps that extract heat from seawater and sewage to warmth total neighborhoods far more effectively than fossil fuels. BBC ARTICLE | HN DISCUSSION

Occasions New Roman was designed for slim newspaper columns, not paperwork The font was commissioned by The Occasions of London in 1931 to maximise readability in tight areas, which is why it appears cramped on full-width pages at this time. A BRIEF HISTORY ARTICLE | HN DISCUSSION

Hundreds of US farmers with Parkinson’s are suing over paraquat publicity MLIVE ARTICLE

Skinny wishes provide the reward with out the transformation Joan Westenberg argues tech stripped thick wishes—ones that change you—into skinny dopamine hits that reproduce themselves with out making you develop. JOAN WESTENBERG ESSAY

Trauma’s that means collapsed after we turned peculiar ache right into a product The Guardian examines how trauma expanded from real disaster to each minor discomfort, creating an business that pathologizes regular life whereas obscuring actual struggling. THE GUARDIAN ARTICLE | GABOR MATÉ INTERVIEW

Concepts don’t get more durable to seek out, we’re simply fishing in fished-out ponds The declining analysis productiveness everybody freaks out about is generally only a measurement drawback—we maintain taking a look at mature fields like Moore’s Legislation as an alternative of measuring precise discovery charges throughout all science. ASTERISK MAGAZINE ARTICLE | HN DISCUSSION

You are not relationship an individual anymore, you are relationship their total digital ecosystem Fashionable relationships aren’t between two individuals—they’re negotiations between competing info programs formed by feeds, group chats, and algorithms that rewrite peculiar friction into pathology. YOU ARE DATING AN ECOSYSTEM

Polymarket accounts maintain successful bets on OpenAI and Google releases elevating insider buying and selling considerations A handful of Polymarket accounts accurately guess OpenAI would launch GPT-5.2 by December 13, making $13k and fueling suspicions these bettors have inside info from working at these firms. THE INFORMATION ARTICLE

John Updike’s letters present he actually by no means stopped writing about writing A brand new 912-page assortment reveals Updike used letters to check materials—each second was content material for the subsequent guide. SELECTED LETTERS REVIEW

Extraordinarily happy with the content material and design of this web site. Few issues are as fantastic as stumbling onto an important essay.