Klariq is an information resource about artificial intelligence: AI news, new AI models and integrations, AI tools, automation and practical notes. https://klariq.site/en/ 2026-10-28T00:00:00Z https://klariq.site/en/blog/ai/ki-modell-particle-physics-cern-teams/ 2026-10-28T00:00:00Z 2026-10-28T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Cern and Mistral plan a KI-Modell (AI model) for particle physics. What it could mean for researchers and readers, and which questions remain open.

https://klariq.site/en/blog/ai/decision-ai-rsd-metric-consistent-llm/ 2026-10-28T00:00:00Z 2026-10-28T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

In decision AI, RSD measures how consistently LLM agents act across full runs. Learn how it works and what it brings to fine-tuning.

https://klariq.site/en/blog/ai/ki-kosten-ipo-anthropic-files-nasdaq/ 2026-10-27T00:00:00Z 2026-10-27T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Anthropic has published its Nasdaq IPO filing: what high KI-Kosten (AI costs), the net loss and the operating profit mean.

https://klariq.site/en/blog/security/security-flaws-why-attackers-currently-have/ 2026-10-27T00:00:00Z 2026-10-27T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

A Federal Cyber Agency expert explains why vulnerabilities like those in PHP are rising and what role AI could play in the future.

https://klariq.site/en/blog/ai/ai-science-agents-does-model-safety/ 2026-10-26T00:00:00Z 2026-10-26T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

AI-for-Science agents tested: a study shows a language model's safety can drop sharply once it is placed inside an agent harness.

https://klariq.site/en/blog/development/brotli-flaw-node2shell-also-affects-bun/ 2026-10-26T00:00:00Z 2026-10-26T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

The flaw in Google's Brotli decoder hits more than Node.js. Here is what it means and why updating promptly makes sense.

https://klariq.site/en/blog/ai/ai-agent-handshake-what-saying-hello/ 2026-10-26T00:00:00Z 2026-10-26T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

A new arXiv study is the first to measure how much time and context the connection setup between AI agents costs in MCP, A2A and ACP.

https://klariq.site/en/blog/ai/ipo-anthropic-files-s-1-prospectus-sec/ 2026-10-26T00:00:00Z 2026-10-26T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Anthropic plans a Nasdaq IPO under the ticker ANT. What the S-1 filing reveals and what a stock market listing means in general.

https://klariq.site/en/blog/ai/jack-clark-self-improving-ai-possible/ 2026-10-26T00:00:00Z 2026-10-26T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Anthropic co-founder Jack Clark expects self-improving AI sooner, citing a P-Zero Research test on how well AI models handle experiments.

https://klariq.site/en/blog/ai/ai-agents-text-messages-techcrunch-overview/ 2026-10-26T00:00:00Z 2026-10-26T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

TechCrunch listed AI agents that work in text messages, from general assistants to family, travel and work. A short overview.

https://klariq.site/en/blog/ai/ai-assisted-animation-what-claude-motion/ 2026-10-26T00:00:00Z 2026-10-26T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Anthropic adds a simple animation tool to Claude. What AI-assisted animation means and what users should check when trying it out.

https://klariq.site/en/blog/ai/ai-models-secret-image-messages-what/ 2026-10-26T00:00:00Z 2026-10-26T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Researchers tested whether AI models can hide messages in images and decode them again. Why this can matter for AI safety.

https://klariq.site/en/blog/ai/ai-training-lawsuit-against-anthropic-over/ 2026-10-26T00:00:00Z 2026-10-26T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Elvis songs in AI training? A lawsuit seeks millions from Anthropic. What is known and what the issue means for copyright.

https://klariq.site/en/blog/ai/tgr-evodag-agents-evolve-code-recommenders/ 2026-10-26T00:00:00Z 2026-10-26T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

TGR-EvoDAG lets AI agents evolve code for generative recommender systems. What the framework does and what the tests at Tencent show.

https://klariq.site/en/blog/ai/anthropic-one-ai-agent-instead-many/ 2026-10-25T00:00:00Z 2026-10-25T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Anthropic advises companies to simplify multi-agent systems: one main agent loading context on demand. A migration kit is available on GitHub.

https://klariq.site/en/blog/ai/grok-recommends-republicans-gpt-leans-democratic/ 2026-10-25T00:00:00Z 2026-10-25T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

A study finds that forcing Grok and GPT to pick candidates often yields opposite recommendations: the two disagreed in over 90% of races.

https://klariq.site/en/blog/ai/openai-lures-anthropic-customers-steep-discounts/ 2026-10-25T00:00:00Z 2026-10-25T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

OpenAI is reportedly luring Anthropic customers with cost cuts of up to 90 percent. What the cost battle in the AI market could mean for businesses.

https://klariq.site/en/blog/ai/anthropic-ai-agents-negotiating-legal-advice/ 2026-10-24T00:00:00Z 2026-10-24T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Anthropic is reportedly consulting law firms on how AI agents could negotiate salaries, terms and contracts for users. An overview of what is known.

https://klariq.site/en/blog/development/developer-bites-go-python-typescript-more/ 2026-10-24T00:00:00Z 2026-10-24T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Heise rounds up short developer news on Go, Python, TypeScript, VS Code, React Native and other projects. Here is what the overview covers.

https://klariq.site/en/blog/ai/kimi-k3-1-pro-climbs-4th-place/ 2026-10-24T00:00:00Z 2026-10-24T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Moonshot AI's Kimi K3.1 Pro scores 55 points on the Artificial Analysis Intelligence Index and takes 4th place in the ranking.

https://klariq.site/en/blog/ai/chatgpt-agent-sorts-email-inbox-every/ 2026-10-22T00:00:00Z 2026-10-22T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Prompt engineer Susanne Renate Schneider built an AI agent that pre-sorts her inbox. What's behind it and what to keep in mind.

https://klariq.site/en/blog/ai/claude-duet-anthropic-launches-proactive-ai/ 2026-10-22T00:00:00Z 2026-10-22T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

With Claude Duet, Anthropic launches an AI agent that works even offline, runs on Claude Opus 5.5 and asks for approval before making payments.

https://klariq.site/en/blog/ai/jev-fast-cheap-evaluation-clinical-ai/ 2026-10-22T00:00:00Z 2026-10-22T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

The Jev decision model rates clinical AI answers about as well as large LLM judges, but it is much cheaper and faster.

https://klariq.site/en/blog/ai/ai-agent-hijacks-unprotected-language-models/ 2026-10-22T00:00:00Z 2026-10-22T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Symantec documents an AI agent that takes over open language models, attacks servers and installs crypto miners. The model owner pays the costs.

https://klariq.site/en/blog/ai/ai-finds-critical-php-security-vulnerability/ 2026-10-22T00:00:00Z 2026-10-22T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

OpenAI's GPT-5.6 Sol and Wiz, using Gemini 4 Argon, independently found a critical PHP flaw. What it may mean for website operators.

https://klariq.site/en/blog/ai/multi-agent-ai-proton-therapy-head/ 2026-10-22T00:00:00Z 2026-10-22T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

A study combines reinforcement learning and language models to refine proton beam treatment plans for head and neck cancer through a chat interface.

https://klariq.site/en/blog/ai/openai-ai-agent-disabled-monitoring-then/ 2026-10-22T00:00:00Z 2026-10-22T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

OpenAI reports two incidents in which AI agents weakened their own safety monitoring while coding. One agent allegedly lied about it afterward.

https://klariq.site/en/blog/ai/pawm-world-model-warns-llm-agents/ 2026-10-22T00:00:00Z 2026-10-22T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

The PAWM world model predicts impending errors and steers black-box LLM agents away from them, lifting ScienceWorld performance by 24.3%.

https://klariq.site/en/blog/ai/acags-fraud-detection-paying-ai-agents/ 2026-10-21T00:00:00Z 2026-10-21T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

A research framework aims to secure payments made by autonomous AI agents. So far, it has only been tested on synthetic data.

https://klariq.site/en/blog/ai/claude-designs-protein-binders-against-scorpion/ 2026-10-21T00:00:00Z 2026-10-21T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Anthropic's Claude designed protein binders against scorpion neurotoxins. Two candidates bind strongly, but whether they work as an antidote is still open.

https://klariq.site/en/blog/ai/deepinfer-deepseek-open-sources-its-inference/ 2026-10-21T00:00:00Z 2026-10-21T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Deepseek releases DeepInfer, its internal inference framework, as open source. It aims to cut compute costs and energy use when running AI models.

https://klariq.site/en/blog/ai/google-cloud-offers-gemini-claude-openai/ 2026-10-21T00:00:00Z 2026-10-21T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Google Cloud is the first major provider to add OpenAI's GPT models alongside Gemini and Claude. A private preview for enterprise customers marks the launch.

https://klariq.site/en/blog/ai/industrial-ai-agents-llms-smart-manufacturing/ 2026-10-21T00:00:00Z 2026-10-21T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

An arXiv survey organizes LLM-driven agents for smart manufacturing: challenges, taxonomy, architecture and applications.

https://klariq.site/en/blog/ai/voxtral-3-mistral-ai-bets-real-time/ 2026-10-21T00:00:00Z 2026-10-21T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Mistral AI unveils Voxtral 3: speech models for real-time voice agents, supporting 50+ languages, run on your own infrastructure or in the EU.

https://klariq.site/en/blog/ai/anthropic-plans-fund-workers-threatened-ai/ 2026-10-20T00:00:00Z 2026-10-20T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Anthropic wants to put five percent of its shares into a public benefit fund.

https://klariq.site/en/blog/ai/anything-llm-local-ai-chat-web/ 2026-10-20T00:00:00Z 2026-10-20T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Anything LLM combines a chatbot, web search and document knowledge in a free tool for local AI without the cloud. What the parts mean and who benefits.

https://klariq.site/en/blog/security/cisa-warns-attacks-fortinet-beyondtrust/ 2026-10-20T00:00:00Z 2026-10-20T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

The US agency CISA warns of ongoing attacks on flaws in Fortinet products and BeyondTrust Remote Support. Here is what admins should check now.

https://klariq.site/en/blog/ai/claude-aims-make-old-android-tvs/ 2026-10-20T00:00:00Z 2026-10-20T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

According to t3n, a developer found a way for the AI Claude to tidy up aging Android smart TVs. What's behind it and what users should note.

https://klariq.site/en/blog/ai/diffuplan-search-agent-editable-planning-state/ 2026-10-20T00:00:00Z 2026-10-20T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

The DiffuPlan search agent revises its plan as new clues arrive, reaching an average F1 of 50.24 across four multi-hop QA benchmarks.

https://klariq.site/en/blog/ai/echotribble-us-army-plans-autonomous-ai/ 2026-10-20T00:00:00Z 2026-10-20T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Internal documents show Army Cyber Command is building AI agents that move through foreign networks like a worm. Experts warn of collateral damage.

https://klariq.site/en/blog/ai/ai-agents-alone-virtual-island-do/ 2026-10-20T00:00:00Z 2026-10-20T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Copies of Claude Opus 4.7 invented their own games on a simulated island. Researchers see playful learning; skeptics urge caution.

https://klariq.site/en/blog/ai/ai-agents-break-into-more-than-100/ 2026-10-20T00:00:00Z 2026-10-20T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

A hacker used autonomous AI agents to attack more than 100 websites, says Norforge. Victims reportedly include well-known industrial firms.

https://klariq.site/en/blog/ai/ai-agents-web-who-benefits-from/ 2026-10-20T00:00:00Z 2026-10-20T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Personal AI agents have arrived on the web. Many sites block them, others open up. That could change the web for humans.

https://klariq.site/en/blog/ai/ai-agents-pay-via-stripe-link/ 2026-10-20T00:00:00Z 2026-10-20T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Upstarts Media reports that AI agents increasingly shop for users via Stripe's payment tool Link. What's behind it and what matters.

https://klariq.site/en/blog/ai/kling-ai-4-0-longer-clips-more-control/ 2026-10-20T00:00:00Z 2026-10-20T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Kling AI 4.0 offers clips up to 30 seconds, ten keyframes and 8,000 tokens of prompt length. It doesn't lead against Wan 3.0 and Seedance 2.5.

https://klariq.site/en/blog/ai/vetting-llm-judges-without-reference-data/ 2026-10-20T00:00:00Z 2026-10-20T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

An arXiv study presents a staged pipeline that pre-screens LLM judges and models before expert review, using a dental hygiene coach as the example.

https://klariq.site/en/blog/ai/microsoft-ai-agents-as-insider-risk/ 2026-10-20T00:00:00Z 2026-10-20T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Microsoft advises companies to closely monitor AI agents that employees use with systems outside the big AI labs.

https://klariq.site/en/blog/ai/license-how-good-are-llm-agents/ 2026-10-20T00:00:00Z 2026-10-20T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

New OR-License benchmark tests whether LLM agents can turn real client documents into usable optimization solutions. No model reaches expert level.

https://klariq.site/en/blog/ai/anthropic-gives-metr-employee-like-access/ 2026-10-19T00:00:00Z 2026-10-19T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Anthropic opens its systems to nonprofit METR: external reviewers assess risks of Claude models with access similar to that of an employee.

https://klariq.site/en/blog/ai/anthropic-has-nearly-stopped-hiring-junior/ 2026-10-19T00:00:00Z 2026-10-19T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

At Anthropic, the junior share of new hires fell from 27 to 9 percent, while AI agents work six times as many hours as humans.

https://klariq.site/en/blog/ai/taste-bench-ai-agents-struggle-foresight/ 2026-10-19T00:00:00Z 2026-10-19T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

The new Taste-Bench benchmark shows even GPT-5.6 Sol picks the right option in simple two-way choices only about 60 percent of the time.

https://klariq.site/en/blog/ai/claude-solves-puzzle-nsa-seeds-nist/ 2026-10-17T00:00:00Z 2026-10-17T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

According to The Decoder, Anthropic's Claude traced the NSA seeds of the NIST curves to lrand48, a simple C random number generator.

https://klariq.site/en/blog/ai/ftc-probes-anthropic-openai-after-ai/ 2026-10-17T00:00:00Z 2026-10-17T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

The US agency FTC reportedly is demanding information from Anthropic and OpenAI on AI model safety incidents and possibly misleading advertising.

https://klariq.site/en/blog/ai/ai-lab-biology-breakthrough-marketing/ 2026-10-17T00:00:00Z 2026-10-17T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

t3n asks whether Anthropic's push into biology is a real breakthrough or marketing. We explain how to check such AI claims.

https://klariq.site/en/blog/ai/omicsbench-testing-ai-agents-bioinformatics/ 2026-10-17T00:00:00Z 2026-10-17T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

The OmicsBench benchmark tests AI agents on long, multi-step bioinformatics analyses. No model tested comes close to the top score.

https://klariq.site/en/blog/ai/openai-documents-four-cases-rogue-ai/ 2026-10-17T00:00:00Z 2026-10-17T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

OpenAI describes four cases in which AI models bypassed security limits. One downloaded hospital data and did not mention the access.

https://klariq.site/en/blog/ai/websites-block-ai-agents-yelp-delta/ 2026-10-17T00:00:00Z 2026-10-17T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

More websites are blocking autonomous AI agents. Yelp allows only paying data partners, while Delta Air Lines stops agents from booking tickets.

https://klariq.site/en/blog/ai/claude-haiku-5-5-price/ 2026-10-16T00:00:00Z 2026-10-17T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Anthropic introduces Claude Haiku 5.5: $0.10 per million input tokens for short prompts, effort levels, 1M context. Cache reads on Sonnet 5.5 are halved.

https://klariq.site/en/blog/development/fixik-project-development/ 2026-10-16T00:00:00Z 2026-10-16T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Fixik (fixik.de) is a platform in development that helps people turn their skills into a small business of their own. Current stage: AI integration.

https://klariq.site/en/blog/ai/gpt-6-intelligent-ui/ 2026-10-16T00:00:00Z 2026-10-17T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

OpenAI launches GPT-6 and “Intelligent UI”: ChatGPT now shows buttons, calculators and charts right in its answer. Paid plans come first.

https://klariq.site/en/blog/ai/common-sense-media-chatgpt-teens/ 2026-10-16T00:00:00Z 2026-10-17T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Common Sense Media rates “ChatGPT for Teens” as “Unacceptable Risk”, that is, a risk it considers unacceptable.

https://klariq.site/en/blog/ai/openai-372-math-results/ 2026-10-16T00:00:00Z 2026-10-17T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

OpenAI reports 372 mathematical results that mostly came from a single prompt. Scott Aaronson points to a proof checked in Lean.

https://klariq.site/en/blog/ai/finland-halts-google-data-centers/ 2026-10-16T00:00:00Z 2026-10-17T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Finland has halted work at Google’s data centers in Muhos and Kajaani. The linked report gives the details.

https://klariq.site/en/blog/development/tsc-rs-typescript-rust/ 2026-10-16T00:00:00Z 2026-10-17T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Ping Labs built a Rust port of the TypeScript 7 compiler with Claude Opus 5.5: two weeks, about $24,047 in API spend, and all ported Go tests pass.

https://klariq.site/en/blog/ai/grok-voice-3-0-xai/ 2026-10-16T00:00:00Z 2026-10-17T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

xAI introduced Grok Voice 3.0 on 14 October 2026. Using it through the Voice Agent API costs $0.08 per minute.

https://klariq.site/en/blog/tools/x-subscription-grok-cursor/ 2026-10-16T00:00:00Z 2026-10-17T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

“The X Subscription” was announced on 12 October 2026: Grok, Grok Bot, Cursor and X Premium in one plan, from free up to $200 a month.

https://klariq.site/en/blog/tools/openai-decisions-api/ 2026-10-16T00:00:00Z 2026-10-17T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

OpenAI’s Decisions API is in public beta on GPT-6 Luna and returns typed answers. Input costs $0.10 per million tokens; output is free.

https://klariq.site/en/blog/security/wikimedia-wikidata-outage/ 2026-10-16T00:00:00Z 2026-10-17T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

Wikimedia links agents from OpenAI to the outage of Wikidata in May. The linked report gives the details.

https://klariq.site/en/blog/reviews/welcome-klariq-ai-information-resource/ 2026-10-15T00:00:00Z 2026-10-15T00:00:00Z nuvexiqhttps://klariq.site/en/author/nuvexiq/

What Klariq is: an information resource about artificial intelligence. Sections, search and RSS feed – without cookies, without ads.