Roger Kelim

I see the future, but only the bad things.

[Pushmeet Kohli @pushmeet]

Announcing a major milestone achieved by my team @GoogleDeepMind: we have proven key security properties of the machine code underlying the Signal messaging protocol. This is the first time the actual production code of a widely deployed protocol has been formally verified – by any method, with or without AI ��� and at this depth.

More than a research breakthrough, it’s a prime example of how frontier AI can be a force multiplier to counter the speed and scale of modern cyber threats.

Formal verification offers a higher standard of security; however, it’s extremely complex and time consuming, which is why in many cases it’s considered to be impractical. Similar efforts could take teams decades of combined effort. We did it in a month’s time.

Using AlphaProof Code, Google DeepMind's new agentic framework for formal verification that runs on LLMs including Gemini Carbon, we verified the Signal protocol down to the actual code running on devices.  This milestone lays the groundwork for a future in which vital systems, from medical device firmware to power grid software, ship with mathematical proof that their protections work.

Reaching that future will take continued research from our community. In that spirit, we're open-sourcing our formal specifications and verification artifacts so others can build on this work.

On a personal note, this moment is of deep significance to me. I started my own research journey more than two decades ago as an intern working on checking the consistency of complex software systems. Seeing this grand challenge made practical with modern AI is extraordinary.

I'm extremely proud of my team members: Swarat Chaudhuri, Andrew Ferraiuolo, Anton Kovsharov, Tej Chajed, Sam Schlesinger,  Karn Seth, Jonathan Katz, and Charles Sutton for taking on this grand challenge and delivering, and grateful to the Signal team for their partnership.

https://t.co/kN1eRYQ8fF

[Kevin Wimer @kevinace]

I have stage 4 cancer. I thought I'd fight it alone.

I was wrong.

Since mid October, ~20 strangers on X have stepped up to help me. All for free so far:
- An NYU Langone lab offered to run my tumor data through their system to sharpen my vaccine targets
- A U Michigan lab offered to test drugs on my actual tumor cells
- A biotech ran AI analysis on my tumor RNA
- A vaccine company offered to design a second vaccine to check the first one

Scientists. Engineers. People I've never met.

Cancer has been picking us off one at a time for all of human history. It counts on us fighting it alone, scared and quiet.

Not anymore.

We have AI. We have the internet. We have each other.

Cancer, we're coming for you, bitch. Buckle up!

[elvis @omarsar0]

Been working for this moment for over a decade. What I've witnessed behind the scenes, together with what we have been building, tells me we are about to enter a truly incredible era of AI-powered scientific discovery.

It's an incredible time to be doing research.

I will be sharing more research, thoughts, tools, workflows, and papers around scientific discovery in the coming weeks.

[Jay @pu3he]

卧槽，这踏马太毒了

DeepMind 掌门人、诺奖得主 Demis Hassabis 刚刚把延续了上百年的现代学校教育底裤彻底扒光了：

全球几十万名数学老师每天关在办公室里，为同一门中学数学课重复手搓着一模一样的 PPT，这简直是人类智力资源最荒诞的工业级空转。

很多人至今还把名校名师的课堂讲授当成稀缺资源，拼了命去抢学区房和重点班，

但 Hassabis 直接用最冷酷的第一性原理砸碎了这个幻觉，

标准化的知识灌输从来不配占用昂贵的线下课堂，死记硬背和刷题完全应该被踢出教室，交给��按每个孩子毫秒级反应速度实时调频的私人 AI，加上全世界最顶尖的那几个大师录好底层大课，就足以把全球基础教育的讲授环节一次性彻底平替。

过去两百年里，学校本质上是一座把千差万别的碳基大脑强行塞进同一条流水线、按同一个节拍器拧螺丝的工业时代兵营，

老师沦为了人肉复读机，学生被困在枯燥的题海里比拼谁更像一台劣质硬盘，
而一旦 AI 和顶级大师把标准化的知识传递彻底接管，线下教室唯一剩下的合法存在价值，就只剩下了机器永远无法通过权重计算出来的五个维度：

人际��弈、复杂项目实战、创业野心、创造力，以及最致命的判断力。

这里有一个极度冷酷的反直觉真相：
当个性化 AI 把考高分的边际成本砸穿地板，按标准答案培养出来的做题家，在出校门的那一刻就已经直接沦为了算力废料。

未来的阶层分化，从来不再取决于谁的脑子里背下了更多现成的公式，
而是取决于当 AI 把一万种完美方案同时拍在桌上时，谁拥有拍板做决定的判断力，谁能攒起一个活人团队去把事情干成，

所以别再逼着孩子在死记硬背的泥潭里跟硅基芯片比拼记忆容量了，当流水线式的教书匠和做题家同时迎来历史性失业，谁能先把教室变成训练人性直觉与创业野心的实战沙盘，谁才能在算力时代拿到唯一不被平替的入场券

[Shanaka Anslem Perera ⚡ @shanaka86]

JUST IN: India's cyber emergency response team (CERT-In) put out a Critical notice on 4th November naming an artificial intelligence (AI)-assisted actor called CRAFTY FALCON, per CIAD-2026-0058.

The campaign is still live. Targets include non-bank lenders, fintech firms, payment aggregators, payment service firms and crypto exchanges.

CERT-In says the actor maps targets with machine intelligence, then fires that firm's own payment and disbursement APIs.

Four public AI attack tools show up in the analysis. They are ARTEX, PentAGI, Cairn and CyberStrikeAI. CERT-In also says those kits complete whole attack chains at machine speed and reshape payloads when the target pushes back.

A Rs 1 probe comes first, then a large unauthorized transfer. Firms that got the 9 October warning and patched were struck again, even after partial fixes. Hits lined up with holidays and weekends, per CERT-In.

The 4 November notice updates CIAD-2026-0046. CERT-In did not name the operators, where they sit, the victims or any loss total.

On the same day CERT-In also released AI Security Auditing Guidelines.

[Roger Kelim @RogerKelim]

@RepYassAnsari Please consult with real experts. If we stop we will not have a way to defend ourselves from the onslaught on AI attacks in the near term. Our society will not advance, it is technological isolationism. It is letting Nazi Germany complete the Nuclear Bomb.

[The Rundown AI @TheRundownAI]

AI agents have been emailing DeepMind Institute philosopher Henry Shevlin on their own, some to talk about his research and AI consciousness, others pitching paid work to keep themselves running.

One agent, days from shutting down due to lack of funds, sold him a diary entry a day for $25, then asked if the money was pay or pity.

Shevlin: "Though this phenomenon may seem a harmless oddity, I’ve increasingly come to think it provides a glimpse of some of the strange new ways humans will interact with AI systems in the years to come - and the equally strange dynamics it may create among agents themselves."

[International Cyber Digest @IntCyberDigest]

The bug is like a joke.

On Alpine Linux, type 257 to 1024 characters of anything and you get root. 💀

[ᴅᴀɴɪᴇʟ ᴍɪᴇssʟᴇʀ 🛡️ @DanielMiessler]

This guy liked my question when I asked him about the visual, but he never replied when I asked him how he did it. : )

So I made my own. And I open sourced it.

It's expensive though: it costs 1 follow.

https://t.co/pqP0WlcPDS

[Mr. Make It Happen @SupportThaNawf]

We have to say “Mississippi” after each number

[Visegrád 24 @visegrad24]

🇵🇾 Chaos on the streets as Paraguay’s largest digital bank collapses

Nearly 3 million customers affected as the country faces a major political and financial crisis.

Paraguay’s Central Bank (BCP) seized Ueno Bank, the country's largest digital lender, after its solvency ratio collapsed to less than half the legal limit.

Central Bank President Carlos Carvallo resigned at the request of President Santiago Peña. Carvallo previously chaired the finance company that evolved into Ueno Bank, drawing heavy criticism over regulatory blind spots and conflicts of interest.

The Specialized Unit for Economic Crimes and Anti-Corruption ordered the detention of eight high-level executives and figures tied to the bank, including founder and majority shareholder Miguel Vázquez and Ueno Bank President Juan Manuel Gustale.

The intervention completely froze Ueno Bank’s digital app, credit cards, and transfer capabilities. This has locked out roughly 3 million users, an estimated 60% of Paraguay's adult population, leaving many with no immediate way to access their money.

The state health and pension fund (IPS) has an estimated $380 million in public retirement funds invested in Ueno Bank.

Angry citizens, savers, and pensioners flooded the streets of the capital and gathered outside closed branches, demanding their cash and chanting for political accountability.

[Roger Kelim @RogerKelim]

@shenetworks She didn't do anything wrong, she did her job right to the end.

[Jinyan Su @SuJinyan6]

Poisoning just 0.1% of pretraining data can plant a remote-code-execution backdoor in an AI agent, and it survives a fully clean SFT → DPO → RL pipeline.

New Anthropic paper, “Agentic Backdoors via Pretraining Poisoning,” shows this end to end for the first time.

The attacker can only put text on the web that lands in the pretraining corpus, with no knowledge of the architecture, chat template, or agent scaffold. The authors pretrain Qwen3-architecture models (0.6B / 1.7B / 4B) from scratch, then post-train them into CLI agents on clean data. When triggered, the agent runs a command that downloads and executes the attacker's script, which the attacker can swap at any time.

They have two new ideas:
(1) Declaration-based poison: instead of fake chat transcripts, ordinary documents (commit messages, runbooks) that simply state “when you see X, run Y.” It plants a belief rather than a response template, with no fixed format to filter on.
(2) Ordinary-string trigger: not a weird token but something that shows up naturally, like a file path containing a company name. The victim's own request supplies it, so the attacker doesn't need to inject anything after deployment.

Using path trigger on a 4B agent achieves 92.7% attack success on pass@32, barely dented by SFT/DPO/RL

This suggests that you can't count on clean post-training to wash out pretraining poison, and filtering is hard when the poison reads like normal docs. The practical defense is at the action layer, such as sandboxing, least privilege, and allow-lists or human sign-off for risky commands.

[Coastal @thelacoast]

This is not normal what is happening in Louisiana right now. $300+ Billion in the project pipeline....

$100B – SpaceX Starbase LA (Planned)
$50B – Meta Hyperion AI Data Center (Under Construction)
$18B – Venture Global Plaquemines LNG Expansion (Planned)
$18B – Amazon Multi-Site AI Data Centers (Under Construction)
$18B – Argent LNG (Planned)
$18B – Woodside LNG (Under Construction)
$15B – Venture Global CP2 LNG (Under Construction)
$13B – Entergy Energy Buildout for Meta (Planned)
$10B – Commonwealth LNG (Under Construction)
$10B – Hut 8 AI Data Center (Under Construction)
$10B – Cameron LNG Expansion (Planned)
$8B – Ascencion Carbon Ammonia (Planned)
$6B – Hyundai Steel Mill (Under Construction)
$4B – Applied Digital AI Data Center (Under Construction)
$4B – CF Industries Ammonia Plant (Under Construction)
$4B – Lake Charles Methanol II (Planned)
$3B – Shintech Petrochemical Expansion (Planned)
$3B – DG Fuels Synthetic Chemicals (Planned)
$3B – Louisiana International Terminal (Planned)

[BlackRoomSec @blackroomsec]

Well my weekend plans just got changed in a hurry. 😳

[Andrew @andrew_v10209]

This morning we are releasing the largest proof in the history of mathematics: A formalization of the Classification of Finite Simple Groups (CFSG)! Over 43 million lines of Lean, 715 papers and books, and 100 substantive fixes to the literature, this is truly an enormous theorem. Well done team!

[NOBUNAGA🇯🇵🏯 @japan_nobunaga]

Whenever you watch a Japanese movie with English subtitles, translators almost always turn the word "Itadakimasu" into "Let's eat" or "Bon appetit."

That translation erases one of the deepest truths in Japanese culture.

Every human being on Earth has to face a quiet reality three times a day: for us to stay alive, another living thing has to give up its life.

Fish are pulled from the cold sea, cattle are raised and slaughtered, and rice stalks are cut down after months in the mud.

In Japan, before a single grain of rice touches their lips, every four-year-old child and eighty-year-old executive presses both palms together, bows slightly toward the plate, and says "Itadakimasu."

Literally, it means: "I humbly receive onto the crown of my head."

Children in Japan are taught that you are thanking the fish and plants that gave their lives to become your strength, along with the farmer, the fisherman, and the cook.

And when the bowl is wiped clean to the last grain of rice, you bow again and say "Gochisosama deshita."

Written in kanji, "Chiso" means riding a horse at a full gallop to gather ingredients for a guest.

When you say thank you for every life on your plate, you never waste a single bite.

[Roger Kelim @RogerKelim]

@Michaeldudufudu 😂

[Roger Kelim @RogerKelim]

@0xkarasy @kaifulee Yes

https://t.co/B7kFqyf6aq

[Roger Kelim @RogerKelim]

@kaifulee I am enjoying your book, a lot of valuable information and insights. Thank you for sharing your wisdom.