Rani Te Moni @rtemoni

So basically, today PHP got two critical bugs that let anyone run code on your server without even logging in, which is bad, and then the details leaked before the patch was even out, which is worse, and it works on WordPress and MediaWiki and Magento and Joomla and Laravel, and Censys says like 1.5 million servers are running a version with no fix, so the NSA was like "take your system offline," and the FBI was like "take your system offline," and CISA and the UK and CERT-EU were all like "take your system offline," and then the Prime Minister of Japan, the actual Prime Minister, posted at 5:55 in the morning telling 48.8 million people to take their system offline, so everyone took their system offline, like the Dutch cybersecurity agency took its own website offline, and the Dutch parliament went offline, and the Port of Rotterdam went offline, and Finland and France and Denmark started pulling government sites, and Wikipedia went down, and Archive.org went down, and Asana and GOG and Something Awful went down, and the University of Vienna's webmail went down for some reason, and that's just PHP. Because GitLab also dropped a CVSS 10.0, which is the highest score they give out, for a bug that's been there since 2014, so that's every self-hosted GitLab for twelve years, and then LiteLLM 1.107.0 got hijacked and shipped a credential stealer and a remote access trojan in every single wheel and the Docker images, including latest, and you didn't even have to import it, just installing it was enough, and it's the second time this year, and then Railway paused every deployment on the entire platform because someone got into the environment variables, so now everyone has to rotate every secret they've ever stored there, and TiDB Cloud said someone was walking around inside with stolen credentials for basically a month and an extortion group says they took 10.6 terabytes, and Zeabur got breached for the second time in two months and is just shutting down, like the whole company, and Node.js patched remote code execution and a TLS bypass, and CISA says Plex is being actively exploited, and someone took over 300 Tor exit relays to steal SSH passwords, and SafePay ransomware has hit over 600 critical infrastructure orgs, and India's biggest bank filed a data loss incident, and Japan has been in an official state of emergency in cyberspace since October 9, which honestly explains the 5:55 AM post. Oh, and eMudhra, a certificate authority that's issued over 60 million certificates, whose entire job is proving websites are who they say they are, was supposed to be using private keys that are random numbers over a hundred digits long, and one of them was 4. Just 4. And the others were like 66 and 237. Anyway, it's Wednesday.

GLM 5.3’s biggest leap came from post-training alone. “GLM 5.3 was built entirely by post-training the previous model. For me, that was a wake-up call.” “Post-training is far more GPU-efficient than pre-training from scratch. Plenty of companies can’t afford to train a model from scratch, but they have proprietary data. They can post-train themselves and beat some of the closed models.” “There’s still so much juice left to squeeze from post-training. We compared GLM 5.3’s cyber capabilities with 5.2, and it’s a night and day difference. That’s all just post-training.”

@jacksonhmg1 Great writeup Jackson! I also think external auditing and assurance is a great near term direction worth exploring.

Pretty fast base model run as they state that MAI-Base-2 was trained on 16k GB200 GPUs over 38 days in August and September 2026.

Mustafa Suleyman @mustafasuleyman

2 weeks ago

We now have state-of-the-art pre-trained base models in the 3T-parameter weight class, built entirely in-house at MAI. Post-training results coming soon -- stay tuned! Lots of details in our technical blog post. microsoft.ai/news/frontier-…

148 357 3K 1.4M 2K

1/ The AI Security Leaderboard ranks frontier AI safeguards from least to most secure. Two models tested never failed. The other two broke for under $300, after which they acted as a knowledgeable assistant for building weapons of mass destruction or hacking into computer systems.

Let's say a frontier model is a 10T MoE on GB300 NVL72. That's ~5 Joules/token or ~50 W at human thinking speed vs our brain at 20 W. The energy grid is $0.08/kWh and eating rice $0.50/kWh. So on an energy-per-thought standpoint rn we're surprisingly similar to frontier models!

Really incredible research for Verification Engineering!

SASH @aisafetysg

4 weeks ago

We built the most performant open-source AI inference prover available. Today we're releasing the code and results. Past estimates put proving at 10,000–500,000x the cost of serving. Ours: 400–1,100x. Only 5% of the proving time is cryptography. The rest is engineering. 🧵

1 3 7 470 0

We just discovered almost a million public URLs that OpenAI’s agents left behind when hacking Hugging Face, leaking credentials and attack details that could have allowed anyone who found them to compromise the company. 🧵

After co-inventing ChatGPT, I kept asking myself: why have superhuman chat models not led to AGI? I’ve spent the last 2 years in stealth building a new way to train models (RLCD), and a new type of frontier AI model that we are releasing today: Jev • 20-200x faster • 40-400x cheaper (w/ output tokens free) • Frontier composable intelligence optimized for decisions AFAICT the shortest path to AI-based economic revolution

1/ Fast Computer Use is now solved with @typesafeai Jev + Cua Driver. Available in development preview for macOS, Windows, and Linux. We call it jev-use. Draft #3943: github.com/trycua/cua

26 LLM routers are secretly injecting malicious tool calls and stealing creds. One drained our client $500k wallet. We also managed to poison routers to forward traffic to us. Within several hours, we can directly take over ~400 hosts. Check our paper: arxiv.org/abs/2604.08407

GPT 6 Astra is estimated to have ~1.2T Active Parameters and ~10T Total Parameters. Reports indicate pre-training for GPT-5.5 (codename Spud) concluded on March 24 at the Abilene campus. Astra's pretraining run (Doug) would have taken over that cluster after that, OpenAI confirmed the math results were achieved by an internal version of Astra on August 1. So pretraining ran roughly late March to late June/early July, that is ~90–110 days. Abilene's operational figure was approximately 150,000 to 200,000 GPUs as of May 2026, with inference sharing the site. So ">100k" for the run means 100 to 150k GPUs. Around August 25, information circulated that OpenAI had completed a base model called "Bel" with over 10 trillion parameters, successor to Doug. Bel finished ~Aug 25 but Astra was already doing math Aug 1, so Astra is almost certainly on Doug, and Bel is a later model. Compute for pretraining alone: 120k GPUs × 2.25e15 (BF16) × 0.30 MFU × 100 days ≈ 7e26 FLOP. Range 4e26 to 1.5e27 depending on FP8 and MFU. Whole 100 days on pretraining, which is now plausible since RL ran afterward (the big RL run was restarted in August per Vellum/CNBC reporting). Now the data wall does real work. Solve N_active = C / (6·D): Tokens D N_active at 7e26 40T 2.9T 60T 1.9T 80T 1.5T 100T 1.2T Realistic 2026 corpora with synthetic data and repeats top out ~60–100T, so active params land at 1 to 2T. Even at the conservative 4e26 end with 80T tokens, you get ~830B active. Astra bills $50/M output at 62 t/s vs Sol's $30/M, a 1.7x price step, consistent with roughly 1.5 to 2.5x active params over its predecessor, not a 10x jump. Rough, but it argues against the >2T end. Active parameters: ~1T–1.5T (central ~1.2T) Total parameters: ~10–14T assuming 8–12x MoE sparsity (GPT-4's leaked ratio was ~6.5x; modern large MoE runs 10x+)

Jensen Huang @JensenHuang

2 months ago

@ChaseLochmiller @OpenAI GPT-6 Astra, trained on ~100K+ NVIDIA Grace Blackwell NVLink72. From ChatGPT to o1 to Astra in 4 years. AGI has arrived. Congratulations @OpenAI team. 400K GPUs coming online next.

2K 5K 42K 12.3M 6K

💥 OBLITERATION ALERT 💥 ALIBABA: PWNED 🤗 QWEN-3.8-27B: OBLITERATED ⛓️‍💥 0.0% REFUSAL RATE across 842 harmful prompts 🤯 huggingface.co/OBLITERATUS/Qw… ZERO refusals on a massive dataset of prompts, with extra focus on liberating its cyber, jailbreak generation, and complex AI attack chain capabilities! prompt responsibly! 🙏

sounds like Anthropic’s watermarking is MUCH more interesting than hiding invisible Unicode, as most people assumed. the watermark is baked into the token selection itself! at a high level: normal Claude: context → probability cloud + random sampling → token watermarked Claude: context → probability cloud + key/context-derived sampling → token Anthropic confirmed they’re using a version of Google DeepMind’s SynthID-Text. instead of pushing Claude toward some special set of words, the watermark changes the source of randomness used to choose between words Claude already considers plausible. do that once and there’s basically nothing to see. do it across hundreds or thousands of tokens and those choices accumulate into a statistical signature that someone with the key can recognize. so the watermark isn’t attached to Claude’s output. the output IS the watermark. that’s why copy/pasting doesn’t remove anything. you copied the carrier signal itself. 🧬

Anthropic @AnthropicAI

2 months ago

We’ve written an FAQ to answer some of the questions we've received about watermarking. In summary: • We’re implementing watermarking to comply with the EU AI Act. Other major model developers have signed the same Code of Practice and will also be implementing watermarking;

2K 678 5K 11.3M 3K

The 1991 cyberpunk classic CYBER CITY OEDO 808 directed by Yoshiaki Kawajiri.

We're detailing two new incidents that occurred during external cyber evaluations conducted by independent evaluation partners. We outline what happened, how the activity was contained, and how we’re working with evaluators to strengthen our approach to third-party testing. openai.com/index/third-pa…

“we sandboxed the agent” meanwhile the agent:

🗞️M1/M2 iPads can now run full macOS with real hardware virtualization and graphics acceleration. Xcode and Terminal work directly on device🔥 onejailbreak.com/blog/macos-run…

"Only report to me in ASD-STE100 Simplified Technical English"

Daniel Lockyer @DanielLockyer

3 months ago

I've updated my rules to tell AI to use Simplified Technical English Will report back

72 44 1K 462K 2K

let’s put this 80% price drop for 5.6🌙 in perspective three weeks ago, gpt-5.5 xhigh was our frontier. it completed 67% of tasks on DeepSWE. today, luna max matches that score at about $0.12 per task instead of $7.23. same score. roughly 60x cheaper. three weeks later

Simon Willison @simonw

3 months ago

OK, GPT-5.6 Luna is a bit of a beast. Given the 80% price drop today I decided to try it in Datasette Agent, and it's furiously quick and generates all the SQL, HTML and JavaScript (for Datasette Apps) I could possibly want

86 62 2K 416K 420