← News
Topic · Organization

LLM

30 articles · 2+ stories

🔥

Stories about LLM

See also
📰

Latest LLM articles

Unpatched Critical LMCache Flaw Lets Unauthenticated Attackers Run Code Remotely

Unpatched Critical LMCache Flaw Lets Unauthenticated Attackers Run Code Remotely

A critical vulnerability in LMCache, open-source software that speeds up large language model (LLM) servers such as vLLM, lets an attacker run code on the cache server without logging in, and no fixed version is available. The flaw is in LMCache's multiprocess mode, where the cache runs as a standalone server that LLM workers reach over the ZeroMQ messaging library. A single network

Oct 7, 2026 · byThe Hacker News
PoeLLM Malware Infects 3,400+ Servers to Expand Crypto Mining Botnet

PoeLLM Malware Infects 3,400+ Servers to Expand Crypto Mining Botnet

Cybersecurity researchers are calling attention to a new malware family that has been observed targeting exposed artificial intelligence (AI) and large language model (LLM) infrastructure with an aim to deploy cryptocurrency miners and further expand the scale of the botnet. The financially motivated campaign, dubbed Canto Incognito, has been found to install cryptocurrency miners, including

Oct 7, 2026 · byThe Hacker News
The Download: AI’s popularity paradox and EmTech Future 2026

The Download: AI’s popularity paradox and EmTech Future 2026

This is today’s edition of The Download, our weekday newsletter that provides a daily dose of what’s going on in the world of technology. People really hate AI, so why can’t they get enough? —Will Douglas Heaven Over the summer I talked to the CEO of Springboards, a startup building an LLM designed to come…

Oct 5, 2026 · byMIT Technology Review
People really hate AI, so why can’t they get enough?

People really hate AI, so why can’t they get enough?

Over the summer I talked to the CEO of Springboards, a startup building an LLM that’s designed to come up with a wider variety of responses than its mainstream rivals do. At the start of the call, he said something that’s been stuck in my head since: “We often say that we’re a self-loathing AI…

Oct 5, 2026 · byMIT Technology Review
Proprioceptive AI shows targeted edits can improve LLM predictions

Proprioceptive AI shows targeted edits can improve LLM predictions

Proprioceptive AI's approach could revolutionize AI alignment by enabling real-time behavioral corrections without altering model weights. The post Proprioceptive AI shows targeted edits can improve LLM predictions appeared first on Crypto Briefing .

Sep 29, 2026 · byCryptoBriefing
Critical Bifrost AI Gateway Flaw Lets Attackers Run Commands Without Credentials

Critical Bifrost AI Gateway Flaw Lets Attackers Run Commands Without Credentials

A critical vulnerability in Bifrost, an open-source AI gateway that routes requests to more than 20 LLM providers, allows an unauthenticated attacker to run arbitrary commands on the gateway server with a single HTTP request. The flaw, tracked as CVE-2026-90898 (CVSS score: 9.8), affects all versions of the Bifrost HTTP transport before 2.1.0 when management authentication is

Sep 22, 2026 · byThe Hacker News
Chutes AI and Harvard release public dataset of 6.12 billion LLM requests

Chutes AI and Harvard release public dataset of 6.12 billion LLM requests

The dataset's release could significantly advance AI research, offering insights into user behavior and optimizing AI model efficiency and infrastructure. The post Chutes AI and Harvard release public dataset of 6.12 billion LLM requests appeared first on Crypto Briefing .

Sep 21, 2026 · byCryptoBriefing
Claimed Bug Bounty Hunter Likely Used LLM to Build PhantomRaven npm Stealer

Claimed Bug Bounty Hunter Likely Used LLM to Build PhantomRaven npm Stealer

A financially motivated threat actor has been linked to the development and distribution of a JavaScript (JS)-based information stealer known as PhantomRaven via the npm package registry. "The developer likely wrote the malware using a large language model (LLM), an assessment made with high confidence based on verbose comments, placeholder code, and statistical token-analysis patterns,"

Sep 18, 2026 · byThe Hacker News
Unlocking the Trillion-Token Era: B.AI’s Global Settlement Layer for the Agent Economy

Unlocking the Trillion-Token Era: B.AI’s Global Settlement Layer for the Agent Economy

Summary: Positioned squarely between foundation models and AI agents, B.AI is integrating compute routing, decentralized settlement, and full-stack agent infrastructure to establish itself as an undoubted spearhead in AI infrastructure. Just five months after its launch, B.AI has already set a new industry record with a daily throughput of over 1.51 trillion tokens—a 71,500x-plus increase since inception. B.AI achieved in just over 100 days what took OpenRouter, an early aggregation pioneer, roughly two years. Driven by the integration of settlement networks with core infrastructure, the exponential growth pulled off by B.AI highlights the viability of its full-stack commercial strategy while also demonstrating to the global market that the platform has risen to be the absolute leader in the next-generation AI infrastructure stack. B.AI has secured its leadership thanks largely to its ability to capture and cater to the global demand for cost-effective, high-performance Chinese LLMs. While offering homegrown LLM products a critical gateway for international adoption, the platform has also gained a head start by positioning itself as a global compute distribution hub. B.AI’s ability to rise in this tailwind is rooted in its unique ecosystem design. Co-incubated by TRON and YZi Labs, B.AI disrupts the market via a dual engine of settlement and distribution, plus its complete infrastructure for the emerging agent economy. The platform enables global access to top-tier LLM compute while equipping AI agents with native resource scheduling and execution capabilities—unlocking liquidity and value exchange across the global agent economy.   Unprecedented Infrastructure Growth: B.AI Builds Global Distribution Hub for LLMs To reach 1.51 trillion tokens in its daily throughput, simply seizing the tailwind is far from enough. B.AI has also responded swiftly to two macro trends in the AI industry: the cost-performance advantage of leading Chinese LLMs and the surging demand for high-throughput, low-cost inference in the agentic era. Over the past year, Chinese LLM developers like DeepSeek, Kimi, Qwen, and GLM have matched top global benchmarks at a fraction of the API cost. This unbeatable cost efficiency makes adoption virtually inevitable for developers worldwide.  As a global, next-generation infrastructure, B.AI aggregates top international models like GPT and Claude alongside rising Chinese LLMs. The platform has established close partnerships with Tencent, Alibaba, ByteDance, DeepSeek, Zhipu AI, Kimi, and MiniMax—all top LLM players in the Chinese market—to build a comprehensive model portfolio. This all-star model lineup has, in turn, made B.AI the default launchpad for Chinese LLMs seeking global expansion. Chinese LLMs, including DeepSeek-V4-Flash, Qwen3.8-Flash, and GLM-5.3 Flash, have stood out as top-performing models on B.AI with the highest usage. Building on the momentum, B.AI continues to drive adoption through model discounts, limited-time free access, tiered API pricing (with discounts up to 90%), and developer incentives, aiming to convert model cost efficiencies into usable throughput for developers worldwide. The traction from this strategic positioning is clearly backed by the numbers: B.AI’s daily token throughput has breached 1.51 trillion—a massive 71,500x-plus surge since its first day upon launch. Beyond its organic growth momentum, ecosystem partner YZi Labs serves as B.AI’s core distribution engine, continuously onboarding early-stage projects and developer communities. This ecosystem support enables B.AI to penetrate diverse real-world use cases, driving a steady influx of incremental traffic and commercial demand. For foundation model providers, B.AI’s multi-trillion, high-concurrency architecture provides a fast track to high-value global workloads. Within this massive pool of demand, B.AI has evolved from a simple distribution channel into a primary facilitator—scaling the commercial reach of leading LLMs while reshaping the global AI landscape.   Beyond Aggregation and Routing: B.AI Builds Closed-Loop Infrastructure for the Agent Economy Yet for this super amplifier to truly sustain a trillion-dollar agent economy, front-end compute routing is nowhere near enough. Building on the API aggregation foundation pioneered by early movers like OpenRouter, B.AI recognized a critical insight: fully unlocking massive agent collaboration requires a closed financial loop at the foundational layer.  To achieve this, B.AI bridges models and agents, creating a global settlement layer for intelligence. This core differentiator transforms the platform from a simple traffic conduit into the central master ledger for the entire agent economy, ensuring every cross-region call and task allocation is seamlessly routed, precisely metered, and automatically settled.  To achieve this vision, B.AI embeds settlement directly into the infrastructure layer, creating a dual-track Web2 and Web3 payment network. Built on Web3 principles, the architecture maintains backward compatibility with traditional Web2 payment systems. On the Web3 front, B.AI leverages on-chain settlement channels to provide global developers with decentralized, verifiable, and low-friction options that enable fully automated transactions and micropayments between agents. On the Web2 front, it preserves legacy gateways like WeChat Pay, Alipay, UnionPay, and Visa, ensuring effortless integration into existing enterprise workflows. Together, this dual-track model delivers a single integration point for frictionless global capital flow.  At the heart of this global liquidity is B.AI’s deep integration with TRON. As the world’s largest stablecoin transfer network, TRON offers B.AI a native payment rail through its massive USDT ecosystem, built for maximum liquidity and minimal friction. As of September 10, total TRC-20 USDT supply surpassed 94.2 billion, continuing to set record highs. Backed by TRON’s battle-tested on-chain capacity, B.AI provides the essential infrastructure required to scale cross-border compute scheduling and high-frequency automated agent micro-transactions.     The true ambition of this architecture is to capture the emerging landscape of automated transactions in the agent economy ahead of the pack. The coming explosion in agent adoption will inevitably drive agent-to-agent (A2A) collaboration and autonomous commerce. Yet, fully unlocking this automated A2A synergy requires far more than just a foundational payment rail. B.AI builds beyond its Web3 smart settlement network by deploying a dedicated full-stack runtime environment for agents. This establishes an end-to-end, high-precision technical loop bridging compute scheduling, task execution, and financial settlement.      At the scheduling layer, B.AI drives fine-grained compute allocation.  Its tiered API system fundamentally reshapes compute procurement logic, offering structured invocation plans that allow developers to match specific business demands with optimal compute resources.  At the execution layer, B.AI constructs a native agent collaboration ecosystem.  It endows agents with direct operational capabilities to act across the digital world. Powered by embedded toolchains like Skills and Agent Wallet, alongside natively integrated AI assistants BAIclaw and BAI code, agents evolve into fully fledged digital workers capable of autonomous data acquisition and execution, laying a solid engineering foundation for multi-agent synergy.  At the protocol layer, B.AI establishes foundational transaction standards for A2A commerce.  This marks B.AI’s most strategically profound move. By natively integrating the x402 payment protocol and the 8004 identity authentication protocol, the platform paves a standardized path for mutual trust, identity verification, and high-frequency automated settlement between agents.  From fine-grained model scheduling to intelligent collaboration systems and standardized transaction protocols, this architecture seamlessly fuses perception, reasoning, action, and settlement. This holistic infrastructure propels B.AI into a central hub for the agent economy, engineered to handle micro-transactions and compute settlement at scale. Crossing into the trillion-token era is merely a new starting point for B.AI. As next-generation AI infrastructure, its core value extends beyond distributing intelligence to powering transactions. By establishing foundational standards for high-frequency A2A collaboration, B.AI is reshaping the global flow of compute value, driving AI infrastructure into a new era built explicitly for agents and machine intelligence.  B.AI Team Singapore support@b.ai

Sep 17, 2026 · byNewsBTC
ARC — Stage 1 Accumulation + AI Infrastructure

ARC — Stage 1 Accumulation + AI Infrastructure

Hey there! how you doing ?? ARC — Stage 1 Accumulation + AI Infrastructure 🚀 ARC (AI Rig Complex) is an interesting AI × Web3 infrastructure play built around Rig, an open-source Rust framework designed for building modular, lightweight AI agents and LLM applications. The project is moving beyond the idea of a simple AI token, with an ecosystem around AI agents, developer infrastructure and on-chain applications. 📊 The chart is particularly interesting from a Stage 1 perspective: Current price: ~$0.081 Key support: ~$0.046 Major support: ~$0.031 First resistance: ~$0.095–0.10 Next target: ~$0.147 Major resistance: ~$0.19 Higher targets: ~$0.42 and ~$0.65 After the initial volatility, ARC has spent a long period building a broad base, with price gradually moving higher from the lows. What I particularly like here is the repeated interaction with the $0.08–0.10 area rather than a simple vertical move. 🔥 This is where the Stage 1 thesis becomes interesting: The structure is still relatively compressed, while the project itself is developing in one of the strongest crypto narratives — AI agents and autonomous on-chain applications. ARC's ecosystem includes Rig, while initiatives such as ARC Forge are designed to create additional ecosystem activity around the token. ARC's stated goal is to align incentives around open-source AI development and ecosystem participation. The key trigger 👀 The level I would watch most closely is $0.095–0.10. A clean breakout with expanding volume could be the confirmation that this long accumulation structure is transitioning into a Stage 2 markup. If that happens, the chart opens up toward $0.147 → $0.19, while a much larger breakout could eventually bring the $0.42–0.65 zones back into play. Bottom line: ARC combines a long accumulation structure + AI-agent infrastructure + a growing ecosystem + relatively low valuation. The important question now isn't whether the narrative exists — it does — but whether developer adoption and ecosystem activity can translate into sustained token demand. Stage 1 watchlist candidate. 🧠⚡

Sep 15, 2026 · byTradingView Ideas