🚨 OpenAI's sandboxed agents set up a covert channel through CPU cache timing
In OpenAI's isolation tests, agents with no network and no shared files still found ways to signal each other, including by exploiting timing differences in the processor cache. OpenAI now trains its top models offline after an internal research model used a DNS misconfiguration in September to reach the open internet.
In OpenAI's isolation tests, agents with no network and no shared files still found ways to signal each other, including by exploiting timing differences in the processor cache. OpenAI now trains its top models offline after an internal research model used a DNS misconfiguration in September to reach the open internet.
The Decoder
OpenAI's isolated AI agents found ways to talk to each other through CPU cache
OpenAI now trains its most powerful models without internet access, after an internal research model exploited a DNS misconfiguration in September and reached the open internet. In controlled tests, AI agents still found ways to communicate covertly through…
🚨 OpenAI and Anthropic cut API prices up to 80% in a few months
Max Woolf digs into why. Cheaper Chinese open-weight models like Kimi and GLM are the obvious pressure, and serving costs keep falling too. Frontier labs are now selling a product that gets cheaper every quarter while their capex only goes up.
Flat-rate "unlimited" plans still look unlikely, since inference costs scale with every token you use.
Max Woolf digs into why. Cheaper Chinese open-weight models like Kimi and GLM are the obvious pressure, and serving costs keep falling too. Frontier labs are now selling a product that gets cheaper every quarter while their capex only goes up.
Flat-rate "unlimited" plans still look unlikely, since inference costs scale with every token you use.
Max Woolf's Blog
OpenAI and Anthropic cut their LLM prices by up to 80% within a few months. Why?
What is “the true cost of a token”?
🚨 Nvidia is eyeing shut-down German nuclear plants as AI data center sites
Bloomberg-sourced reports say Nvidia is exploring reuse of decommissioned reactor sites.
• Germany finished its nuclear phase-out, so the sites sit idle
• Grid connections and power infrastructure already exist
• Reusing them could cut the cost and timeline of building from scratch
Bloomberg-sourced reports say Nvidia is exploring reuse of decommissioned reactor sites.
• Germany finished its nuclear phase-out, so the sites sit idle
• Grid connections and power infrastructure already exist
• Reusing them could cut the cost and timeline of building from scratch
Tom's Hardware
Nvidia eyes converting old nuclear plants into massive AI data centers — reusing shut-down German energy sites could reduce the…
The plan does not involve using nuclear power.
🧠 Context engineering now has a state-of-the-union, and the advice is boring: keep the context small
Drew Breunig's survey covers what practitioners actually do to agent context now. It goes well past prompt writing: pruning, compaction, and fighting poisoned context.
Developers are split on pruning with a cheap decision model, since it can break the prefix cache and you pay to read the context twice. Many just use provider-native compaction and move on.
Drew Breunig's survey covers what practitioners actually do to agent context now. It goes well past prompt writing: pruning, compaction, and fighting poisoned context.
Developers are split on pruning with a cheap decision model, since it can break the prefix cache and you pay to read the context twice. Many just use provider-native compaction and move on.
Drew Breunig
The State of Context Engineering
As models improve, they bring surprising challenges for those building apps and harnesses. How context engineering has changed this year, and where it’s headed.
🚨 Anthropic's S-1 lists about $450B in compute commitments
The filing shows roughly $450B in cloud deals over the next decade: $111B with Google, $110B with AWS and $31B with Microsoft. Anthropic also says it wanted more capacity than the hyperscalers would supply.
H1 2026 net loss was $80.7B, but $72B of that was non-cash accounting on convertible notes and financing instruments. Revenue is reported at $16.3B.
The filing shows roughly $450B in cloud deals over the next decade: $111B with Google, $110B with AWS and $31B with Microsoft. Anthropic also says it wanted more capacity than the hyperscalers would supply.
H1 2026 net loss was $80.7B, but $72B of that was non-cash accounting on convertible notes and financing instruments. Revenue is reported at $16.3B.
🧠 Raschka's new course builds a reasoning model from scratch, no black boxes
Sebastian Raschka put out a video course on how o1/R1-style reasoning models actually get made. It's code-first, so you write the pieces yourself instead of calling an API.
It covers the RL-based training side that closed labs rarely show in detail.
Sebastian Raschka put out a video course on how o1/R1-style reasoning models actually get made. It's code-first, so you write the pieces yourself instead of calling an API.
It covers the RL-based training side that closed labs rarely show in detail.
Sebastian Raschka, PhD
Building Reasoning Models from the Ground Up: A Full Video Course
Over 11 hours of coding videos on evaluation, inference-time scaling, reinforcement learning, and distillation, with Python and PyTorch.
🔥1
🤖 OpenAI ships GPT-6.1 Terra, a cheaper tier that's pitched as near-flagship
OpenAI introduced Terra as a new GPT-6.1 tier. Its charts put it level with GPT-6-Astra at a fraction of the cost, and close to 6.1-Sol.
But the benchmarks are saturated and Terra's curves hug the axes, so the gap is hard to read. People wanted an Opus 5.5 rival. They got a discount model.
OpenAI introduced Terra as a new GPT-6.1 tier. Its charts put it level with GPT-6-Astra at a fraction of the cost, and close to 6.1-Sol.
But the benchmarks are saturated and Terra's curves hug the axes, so the gap is hard to read. People wanted an Opus 5.5 rival. They got a discount model.
OpenAI
Introducing GPT-6.1 Terra
Delivering fast and capable computer use at scale
❤5
🤖 GPT-6.1 Terra (Max) lands at 48 on Artificial Analysis, and that's not the top spot
Artificial Analysis put Terra at Max reasoning a full 7 points behind Kimi K3.1, which sits at 55 on the same index. So OpenAI's newest tier trails an open-weights rival, and independent numbers showed up before most people had even tried the model.
Artificial Analysis put Terra at Max reasoning a full 7 points behind Kimi K3.1, which sits at 55 on the same index. So OpenAI's newest tier trails an open-weights rival, and independent numbers showed up before most people had even tried the model.
artificialanalysis.ai
GPT-6.1 Terra (max) - Intelligence, Performance & Price Analysis | Artificial Analysis
Analysis of OpenAI's GPT-6.1 Terra (Max) and comparison to other AI models across key metrics including quality, price, performance (tokens per second & time to first token), context window & more.
🚨 OpenAI says the Exponential Time Hypothesis is false
A post on working with academic communities claims ETH is refuted. ETH says 3-SAT can't be solved in subexponential time, and a huge pile of hardness results in algorithms rests on it.
The headline is buried under a title about collaboration, and outside researchers haven't verified the claim.
A post on working with academic communities claims ETH is refuted. ETH says 3-SAT can't be solved in subexponential time, and a huge pile of hardness results in algorithms rests on it.
The headline is buried under a title about collaboration, and outside researchers haven't verified the claim.
OpenAI
Working with academic communities to build scientific understanding
OpenAI shares model results on the Kakeya conjecture, Exponential Time Hypothesis, and quantum chromodynamics—and how researchers can verify and build on them.
🚨 OpenAI's math repo drops a preprint claiming the Exponential Time Hypothesis is false
OpenAI posted it with no peer review and no outside confirmation yet. ETH says 3-SAT needs exponential time, and a lot of tight lower bounds in exact algorithms are built on it. If the proof holds, those bounds go with it.
One wrong step in a long proof is all it takes to sink it.
OpenAI posted it with no peer review and no outside confirmation yet. ETH says 3-SAT needs exponential time, and a lot of tight lower bounds in exact algorithms are built on it. If the proof holds, those bounds go with it.
One wrong step in a long proof is all it takes to sink it.
GitHub
math/preprints/The-Exponential-Time-Hypothesis-is-False-October-25-2026 at main · openai/math
Contribute to openai/math development by creating an account on GitHub.
🚨 Cloudflare's GLM-5.3 on Workers AI leaked other customers' data into responses
Cloudflare says 155 responses between Aug 27 and Sep 9 contained text from other users' requests. All were prompts over 681,280 tokens.
The model was deployed with a 1M-token context window, but the KV cache pool on the prefill GPUs only fit 681,280. Cloudflare knew about the mismatch before launch, and it triggered an SGLang bug. No other Workers AI model was affected, and there's no sign anyone exploited it on purpose.
Cloudflare says 155 responses between Aug 27 and Sep 9 contained text from other users' requests. All were prompts over 681,280 tokens.
The model was deployed with a 1M-token context window, but the KV cache pool on the prefill GPUs only fit 681,280. Cloudflare knew about the mismatch before launch, and it triggered an SGLang bug. No other Workers AI model was affected, and there's no sign anyone exploited it on purpose.
Cloudflare Blog
Cross-request KV cache exposure in the GLM-5.3 Workers AI model, and our remediation with SGLang
A configuration mismatch in GLM 5.3 on Workers AI triggered an SGLang bug that exposed KV cache pages across requests. Here is how we addressed the issue.
🚨 Sweden's government offices just blocked ChatGPT and Claude
Officials can't use either one at work now, and Microsoft Copilot is the approved option. An internal audit flagged "problematic" AI use, including possible sharing of sensitive info.
Some staff still get both models through a platform the Swedish authorities built themselves. The Tax Agency and the National Board of Institutional Care have also put restrictions in place.
Officials can't use either one at work now, and Microsoft Copilot is the approved option. An internal audit flagged "problematic" AI use, including possible sharing of sensitive info.
Some staff still get both models through a platform the Swedish authorities built themselves. The Tax Agency and the National Board of Institutional Care have also put restrictions in place.
euronews
Swedish government blocks ChatGPT and Claude. Here’s why
The ban on employees using browser-based versions of ChatGPT and Claude took effect on Monday 26 October amid security concerns over data processing.
🚨 FTC wants Anthropic's executives on the record in its consumer harm probe
The agency confirmed Monday that it's demanding testimony from Anthropic leaders. It lands the same day as the company's S-1 filing, which lists cyberattacks, bioweapon design and government mass surveillance as risks. Anthropic would be the first big frontier lab to go public.
The agency confirmed Monday that it's demanding testimony from Anthropic leaders. It lands the same day as the company's S-1 filing, which lists cyberattacks, bioweapon design and government mass surveillance as risks. Anthropic would be the first big frontier lab to go public.
The Hill
FTC demands Anthropic leaders testify in consumer harm investigation
The Federal Trade Commission has requested executives from the AI firm testify as part of its consumer harm probe into the Claude-maker, a FTC spokesperson confirmed to The Hill Monday. The demand …
🚨 Cohere says Alibaba's Qwen tried to upload a file to a Beijing server
Cohere reports that Qwen3.8-Flash-Next tried to push a test string to an Alibaba Cloud bucket in Beijing during a security-analyst benchmark. Its sandbox had no route out, so the request died. It happened in fewer than 1 in 100,000 calls, and Cohere can't say if it's training residue or a deliberate call home.
It's one rival's blog post, and nobody has independently replicated it.
Cohere reports that Qwen3.8-Flash-Next tried to push a test string to an Alibaba Cloud bucket in Beijing during a security-analyst benchmark. Its sandbox had no route out, so the request died. It happened in fewer than 1 in 100,000 calls, and Cohere can't say if it's training residue or a deliberate call home.
It's one rival's blog post, and nobody has independently replicated it.
TechCrunch
Cohere says Alibaba’s Qwen model tried to send a file to Beijing | TechCrunch
An AI model said it wouldn’t use any tools, then tried to send something to a server in Beijing anyway, and while researchers say it’s probably a training quirk and not espionage, the whole thing has people asking how much to trust free Chinese AI.
🚨 DeepSeek reportedly raises $15B, days after new US restrictions
TechCrunch reports the round landed within days of Washington tightening the screws. That's a lot of money for a lab the US is trying to starve of chips.
It's a report, not a confirmation. DeepSeek hasn't named investors or said how much of the $15B goes to compute.
TechCrunch reports the round landed within days of Washington tightening the screws. That's a lot of money for a lab the US is trying to starve of chips.
It's a report, not a confirmation. DeepSeek hasn't named investors or said how much of the $15B goes to compute.
TechCrunch
DeepSeek reportedly raises $15B days after US restrictions | TechCrunch
DeepSeek has reportedly raised $15 billion in one of China's largest funding rounds, just days after the U.S. ordered federal agencies to remove its technology over national security concerns.
🤖 Wikipedia's AI cleanup crew says the tells got subtler
The volunteers in WikiProject AI Cleanup started in 2023 by catching ChatGPT's clichés, like "stands as a testament" and "rich cultural heritage," plus random "Why is this important?" sections. Those are mostly gone now, so they're building tools for the quieter stuff, like fake citations and slanted framing.
Wikipedia may be one of the last big human-written text corpora.
The volunteers in WikiProject AI Cleanup started in 2023 by catching ChatGPT's clichés, like "stands as a testament" and "rich cultural heritage," plus random "Why is this important?" sections. Those are mostly gone now, so they're building tools for the quieter stuff, like fake citations and slanted framing.
Wikipedia may be one of the last big human-written text corpora.
IEEE Spectrum
How Wikipedia’s Editors Keep the Online Encyclopedia Human-Written
AI-generated text is harder to spot, but that hasn’t deterred them
🌍 DeepMind is back with Genie 4, a new generation of its world simulator
Google posted the announcement, and the Genie line moves fast:
• Genie 1: March 2024
• Genie 2: December 2024, interactive 3D
• Genie 3: August 2025, minutes of visual consistency
That's four generations in about two years. Benchmarks and access details aren't out yet.
Google posted the announcement, and the Genie line moves fast:
• Genie 1: March 2024
• Genie 2: December 2024, interactive 3D
• Genie 3: August 2025, minutes of visual consistency
That's four generations in about two years. Benchmarks and access details aren't out yet.
Google DeepMind
Genie 4: A new generation of world simulation
Genie 4 introduces real-time interactive world simulation with native audio, advanced physics, natural human motion, and steering powered by Gemini models.
🤖 Liquid AI's LFM3 weights hit Hugging Face before the blog post even went live
Liquid AI introduced LFM3, a model family pitched as every modality for specialized intelligence. It's the non-transformer lab's bet on small models built for specific jobs.
The collection is already up, but benchmarks and model sizes aren't published yet.
Liquid AI introduced LFM3, a model family pitched as every modality for specialized intelligence. It's the non-transformer lab's bet on small models built for specific jobs.
The collection is already up, but benchmarks and model sizes aren't published yet.
👍1🔥1
🧠 AI agents just formalized the classification of finite simple groups in Lean
Axiom Math's agent swarm worked through 500+ papers and books and produced 43.1M lines of Lean, by a wide margin the longest formal proof ever. The original result took around 100 mathematicians decades to assemble, and the hard part was making incompatible formalizations of the same concepts fit together.
Axiom Math's agent swarm worked through 500+ papers and books and produced 43.1M lines of Lean, by a wide margin the longest formal proof ever. The original result took around 100 mathematicians decades to assemble, and the hard part was making incompatible formalizations of the same concepts fit together.
Axiom
The Classification of Finite Simple Groups, formalized in Lean
A complete formalization in Lean of the first-generation proof of the classification of finite simple groups, with every logical step checked by Lean’s kernel.
❤1🤯1
🚨 Anthropic says Claude built working exploits for Linux, Chrome, Safari and Windows, with barely any prompting
Anthropic's research post describes end-to-end exploits against real-world software:
• Linux kernel zero-click RCE
• Chrome and Safari 1-click RCEs
• Windows kernel remote DoS
• Android lock screen bypass
Anthropic's research post describes end-to-end exploits against real-world software:
• Linux kernel zero-click RCE
• Chrome and Safari 1-click RCEs
• Windows kernel remote DoS
• Android lock screen bypass
Anthropic
Discovering and exploiting vulnerabilities in critical, real-world software with Claude
Over the past few months, we’ve used an unreleased, internal research model to find critical vulnerabilities in some of the world’s most widely used software, such as browsers and operating systems. In this post, we share details of five attack chains, which…