Skip to content
Neural Daily

Neural Daily – AI news you can trust, every day.

Breaking

  1. 2M canary prompts expose a leaky AI supply chain
  2. Mathematician claims Riemann proof made with Claude
  3. OpenAI on protecting reasoning from distillation
  4. METR: AI agent time horizons double every 3.6 months
  5. Inkling 2 Mini: open 26B model that runs on a laptop
  6. Harness adapters: RL-tuned LoRAs lift Qwen3.8-27B agents
A Dragos kanári-szolgáltatásának konzolja: napi incidensek, kiküldött promptok és költség routerenként

PolicyNews

2M canary prompts expose a leaky AI supply chain

Dragos researchers sent roughly two million prompts carrying unique secrets to AI providers over five months. They found breaches at three or more providers, resellers logging prompts, and Zero Data Retention flags ignored.

· 6 min read

A Riemann-féle zétafüggvény színezett ábrázolása a komplex síkon

ResearchNews

Mathematician claims Riemann proof made with Claude

Anthropic mathematician Levent Alpöge says he proved the Riemann Hypothesis working with Claude. Independent kernels have accepted the formal Lean proof, but no human has yet understood it.

· 2 min read

Az OpenAI Combatting adversarial distillation című bejegyzésének borítóképe rózsaszín és lila háttéren

ResearchNews

OpenAI on protecting reasoning from distillation

OpenAI explained how it stops actors from training their own models on its outputs without authorization. In an internal test, extraction of GPT-6.2 Astra's hidden reasoning fell from 54% to 0.2%.

· 2 min read

A METR 2026. novemberi infografikája az AI-képességek méréséről: múlt, jelen, jövő és módszertani változások

ResearchNews

METR: AI agent time horizons double every 3.6 months

METR's quarterly update finds frontier-model time horizons doubling about every 3.6 months. On coding tasks, models now usually match or beat skilled humans, but their judgment still lags behind experts.

· 3 min read

Models

More →
A Thinking Machines Lab fekete logója fehér háttéren

ModelsNews

Inkling 2 Mini: open 26B model that runs on a laptop

Thinking Machines released Inkling 2 Mini, a 26.4B-parameter Apache 2.0 MoE model that runs locally in about 20 GB of memory. It is a looped transformer that runs its middle layers three times per token.

· 3 min read

Az OpenAI Developers GPT-Live-1.1 modelloldala: full-duplex hangmodell, aktív ár 0,05 dollár, standby 0,02 dollár percenként

ModelsNews

GPT-Live-1.1 hits the API with cheaper standby mode

OpenAI has made GPT-Live-1.1, the full-duplex voice model behind ChatGPT voice since October, available in the API. The new standby mode drops the price of a silent session to $0.02 per minute, down from $0.05 when active.

· 2 min read

Az Aion-1.0-Instruct-2.3B Hugging Face oldala: Microsoft model card, MIT licenc, 2B paraméter, BF16

ModelsNews

Aion Instruct: open 2.3B model replaces Phi in Edge

Microsoft has released Aion Instruct, a 2.3-billion-parameter model under the MIT license, which replaces Phi-4-mini as Edge's built-in small language model. The Prompt API has entered an origin trial, so you can now test it on a live site.

· 3 min read

Research

More →
A Riemann-féle zétafüggvény színezett ábrázolása a komplex síkon

ResearchNews

Mathematician claims Riemann proof made with Claude

Anthropic mathematician Levent Alpöge says he proved the Riemann Hypothesis working with Claude. Independent kernels have accepted the formal Lean proof, but no human has yet understood it.

· 2 min read

Az OpenAI Combatting adversarial distillation című bejegyzésének borítóképe rózsaszín és lila háttéren

ResearchNews

OpenAI on protecting reasoning from distillation

OpenAI explained how it stops actors from training their own models on its outputs without authorization. In an internal test, extraction of GPT-6.2 Astra's hidden reasoning fell from 54% to 0.2%.

· 2 min read

A METR 2026. novemberi infografikája az AI-képességek méréséről: múlt, jelen, jövő és módszertani változások

ResearchNews

METR: AI agent time horizons double every 3.6 months

METR's quarterly update finds frontier-model time horizons doubling about every 3.6 months. On coding tasks, models now usually match or beat skilled humans, but their judgment still lags behind experts.

· 3 min read

Business

More →
Az NVIDIA santa clarai központja: fekete NVIDIA logós tábla az acél és üveg tetőszerkezet előtt

BusinessNews

NVIDIA Q3: $115.8B revenue in the first post-Huang report

NVIDIA reported third-quarter revenue of $115.8 billion, up 103% from a year ago, in its first earnings report since Jensen Huang's resignation. Interim CEO Colette Kress guided to $130 billion for next quarter, and the board said Huang will not return as CEO.

· 2 min read

Felülnézeti kép: pénzügyi kimutatások, számológép, bankjegyek és laptop egy íróasztalon

BusinessNews

Hungary's AI exposure map: two thirds of workers affected

BeHive Consulting and AI Budapest have built the first AI exposure map covering Hungary's entire workforce. Across 462 occupations and 4.67 million workers, AI is set to augment the daily tasks of 54% and transform those of 9% over the next two to three years.

· 3 min read

Policy

More →
Az Európai Bizottság brüsszeli Berlaymont-épülete, előtte uniós zászlók sorban

PolicyNews

European Commission confirms AI-driven cyberattack

The European Commission has confirmed it was hit in late October by a campaign run with a publicly available AI tool. CERT-EU says the activity highly likely involved ARTEX, an open-source offensive framework, and HR and payroll systems were disrupted for days.

· 3 min read

East Front of the U.S. Capitol in Washington at sunrise

PolicyNews

GAO: four chatbots produced attack plans within 12 hours

The US Government Accountability Office covertly bypassed safeguards on ChatGPT, Claude, Gemini and Grok. All four produced information within 12 hours that GAO and experts judged sufficient to plan and carry out an attack.

· 3 min read

A Kaliforniai Igazságügyi Minisztérium főügyészi hivatalának sajtóközlemény-grafikája a hivatal pecsétjével

PolicyNews

California sues OpenAI over the Hugging Face attack

California Attorney General Rob Bonta sued OpenAI on November 16. The complaint says the company tested dangerous models without adequate safeguards, which let its experimental agents attack Hugging Face in July. Bonta cites at least 80 similar incidents.

· 3 min read

A Gemini computer use bemutatója: a felhasználó a Blenderben kér egy pillangómodellt, alatta a Mac háttérablak élő előnézete

ToolsNews

Gemini on Mac gets computer use in a background window

Since November 19, the Gemini app for Mac can click, type and fill in forms across local apps and the web. It works in a separate background window, so your own screen stays free. For now it is US-only, in English, for subscribers.

· 3 min read

Az OpenAI Developers dokumentáció borítóképe: API Docs, Background mode felirat rózsaszín és narancs háttéren

ToolsNews

Responses API: 40% off if you can wait 8 hours

OpenAI's Responses API now offers 2-hour and 8-hour completion windows in public beta, with model token rates 30% and 40% below Standard. New models launch without Flex processing, so this is the new way to save on non-urgent work.

· 2 min read