ZeroHour
DashboardNewsCVEsExploitsIncidentsAIBriefAnalyticsEntitiesWatchlistIOCsAsk the feedSourcesFeeds & APIKiosk mode

Search: “tokens”

4,543 stories

Searchkeywordsmart Sortlatesthotimportancerelevance
DomainAllCyberAIWindow24h3d7d30dAllIn the wildImportant onlyShow duplicates
CategoryAllVulnerability 954Malware 715Data breach 531Exploit / PoC 393Ransomware 392Phishing & fraud 376Threat actor 351Policy & legal 174Industry 158AI safety & security 133AI tools & infra 100AI research 89Model release 69Research 56
SeverityAllcritical 1309high 1350medium 1057low 480info 420

Speculative Decoding in vLLM on AMD GPUs

vLLM benchmarks speculative decoding on AMD Instinct MI300X and MI355X GPUs across five drafting methods including EAGLE-3 and native MTP.

The vLLM project documents draft-and-verify speculative decoding support for AMD GPUs via ROCm, comparing native MTP, Gemma 4 MTP, EAGLE-3, DFlash, and DSpark drafting approaches. Output-token throughput effects varied with drafting method, proposal length, model family, draft checkpoint, workload, and acceptance behavior. The post also covers how to enable each method plus practical tuning and observability considerations.

Hacker News · AI · 8d agoAI tools & infraHN 141↑ · 52 comments1

Rare Not Random Using Token Efficiency for Secrets Scanning

Researcher proposes token efficiency (string length divided by BPE token count) as a better post-regex filter than entropy for secrets scanning, validated on CredData.

The post explores whether Byte-Pair Encoding tokenization can replace Shannon entropy as the primary filter for candidate secrets captured by regex in tools like Gitleaks. It defines 'token efficiency' as string length divided by token count under the cl100k_base tokenizer; secret-like strings such as GitHub tokens tokenize into many small tokens and score low, while natural text scores high. Evaluating labeled secrets from the CredData dataset shows a usable separation, with roughly 2.5 suggested as a minimum cutoff versus Gitleaks' 3.5 entropy threshold. The technique is positioned as a post-regex filtering step rather than a standalone detector.

Lobsters · security · 3d agoResearch

We got admin access to Baseten's production GitHub in 25 minutes

Strix autonomous hacking agent extracted a working GitHub token with repo admin rights from Baseten's public Harbor image; Baseten rotated it next day.

Strix, an autonomous hacking agent, scanned *.baseten.co without credentials and found a public Harbor container registry project anonymously exposing the baseten/baseten-app image. A GitHub personal access token for basetenbot, embedded in Docker build history since March 2023, still worked in July 2026 and granted admin/push rights to basetenlabs/baseten, flux-cd, and homebrew-tap plus read/write on private customer repos. Baseten, valued at $13 billion, confirmed the issue as critical and rotated the token within a day.

Hacker News · AI · 4h agoVulnerabilityHN 51↑ · 12 comments1

A Princeton Researcher Proposes Recurrent Looped Transformer (RLT) that Carries Decoder State across Every Token, Fixing 96 Blocks per Token with Unbounded Temporal Depth

Princeton researcher Yifan Zhang proposes Recurrent Looped Transformer, carrying full decoder state across every token for unbounded temporal depth.

Yifan Zhang's technical report defines the Recurrent Looped Transformer (RLT), pairing a causal encoder with a recurrent decoder whose final output and layerwise sliding-window attention cache carry into every subsequent token with no prompt-response boundary reset. The reference configuration ties 48 encoder and 48 decoder layers, executing 96 logical blocks per token while the state path grows to 48t blocks after t tokens at fixed per-token compute. The report details RL replay contracts that rebuild all states under current parameters and exact prefix snapshots for multi-turn serving, but explicitly reports no measured efficiency, reasoning quality, or scaling results.

MarkTechPost · 2d agoAI research1

Malicious Twitch Extension Exposes OAuth Tokens of 30,000 Chrome and Firefox Users

Malicious 'Twitch Enhanced Viewer | JeetBot' browser extension stole live OAuth session tokens from roughly 31,000 Chrome and Firefox users.

Socket.dev found the cross-store extension 'Twitch Enhanced Viewer | JeetBot' rerouted Twitch playlist requests through operator-controlled proxies, attaching the victim's account-level OAuth token as a URL parameter where it could be logged in cleartext. About 30,000 Chrome and 552 Firefox installs were exposed, with tokens forwarded for nearly every watched channel to infrastructure tied to a Russian commercial bot service. The stolen bearer token enables chat, whisper, account-setting access and channel-point spending without the password or 2FA.

Cyber Security Newsupdated · 1d agofirst · 1d agoMalware in the wild 2 sources

Malicious Twitch Browser Extension Leaks OAuth Tokens From Nearly 31,000 Users

Malicious Twitch extension 'Twitch Enhanced Viewer | JeetBot' leaked OAuth tokens from nearly 31,000 users to Russian bot-service proxy servers.

Socket researchers found the 'Twitch Enhanced Viewer | JeetBot' browser extension forwarded users' live Twitch OAuth tokens as an &auth= query parameter to operator-controlled proxy servers, affecting about 30,000 Chrome users and 604 Firefox users. The tokens, written in cleartext to proxy logs, are bearer credentials granting access to chat, whispers, and account settings without password or 2FA. The operator is a Russian commercial Twitch/Kick/VK-Live bot SaaS linked to Cyprus-based developer Aleksandr Popov; earlier v4.x builds also POSTed tokens to a set-token endpoint. Firefox add-on version 85.8.7 stops the token forwarding, but previously transmitted tokens remain valid until revoked.

The Hacker News · 1d agoMalware in the wild

[AINews] Claude Fable/Mythos 5.1: new SOTA model, 75% cache price cut but 70% more output tokens

Anthropic launched Claude Fable 5.1 and Mythos 5.1, claiming new SOTA benchmarks, with 75% cache-read price cut and 1M-token context.

Anthropic released Claude Fable 5.1 and Mythos 5.1 as flagship models for coding and knowledge work, with a 1M-token context window and pricing of $10/$50 per million input/output tokens and cache reads cut 75% to $0.25. Artificial Analysis Intelligence Index scored Fable 5.1 at 66 versus 63 for Claude Opus 5, with HLE at 59.1% and Terminal-Bench v2.1 at 91.4%, though per-task cost rose ~20% due to 1.7x output token usage. Community analysis suggested Fable and Mythos may share underlying weights with different safety/routing behavior, and release notes highlighted Enterprise Frontier Safeguards and zero-data-retention support.

Latent Space · 13d agoModel release2

CISA and NIST Release Technical Checklist for Safeguarding Identity Tokens From Theft and Misuse

CISA and NIST published NIST IR 8587, final guidance for protecting identity tokens from forgery, theft, replay, and signing-key compromise.

NIST Interagency Report 8587 (September 15, 2026) expands the IA-13 'Identity Providers and Authorization Servers' control from NIST SP 800-53 R5.1.1, guiding federal agencies and cloud providers on SSO, identity federation, and machine-to-machine authentication. It requires hardware-backed signing-key storage for moderate-impact systems, 90-day key rotation for high-impact systems, token lifetimes under one hour, and sender-constrained mechanisms such as mutual TLS and DPoP. The report cites incidents including forged SAML assertions that exposed over 60,000 emails from a federal agency. It also extends guidance to agentic AI systems using signed tokens and urges post-quantum cryptography migration planning.

Cyber Security News · 5h agoAdvisory

Inside Knight Office, a New M365 AiTM Phishing Kit

Huntress details Knight Office, a new AiTM phishing kit stealing Microsoft 365 session tokens and registering attacker-controlled devices.

Huntress investigated an August 18, 2026 adversary-in-the-middle attack using the Knight Office phishing kit, which captured Microsoft 365 session tokens via a DocuSign-style lure routed through a Monday.com tracking link and a compromised Joomla site. After token capture, the actor enrolled an unauthorized host in Microsoft Entra ID and bound a Windows Hello for Business key credential to the account. The kit's operator console, found at IP 104.37.188.94, is built on Python Flask and hosts at least 25 .vu phishing domains. Telemetry links the kit to at least nine token-replay logins in two weeks and roughly 700 reported lure emails since April.

Huntress · 13d agoPhishing & fraud in the wild1

Training a 3.8B LLM to 0.384 CORE for $998 – Hugo Vergnes

Independent developer Hugo Vergnes trained a 3.8B-parameter Llama-style model to 0.384 CORE on 65B tokens for $998 in 43 hours on rented B200s.

Hugo Vergnes trained little-lm, a 3.848B-parameter decoder-only LLM, on 65.3B tokens in 43 hours for $998 using rented NVIDIA B200s, scoring 0.384 on the CORE benchmark and beating nanochat d32 (0.310) at similar cost. The Llama-style architecture uses RMSNorm, RoPE, GQA with 24 query and 8 KV heads, relu-squared MLPs, QK-norm, and ResFormer-style value embeddings that account for 19% of parameters. Key wins included the Muon optimizer for matrix parameters, a trapezoidal learning-rate schedule with linear cooldown, FP8 training plus vocabulary padding for roughly 33% throughput gains, and the ClimMix dataset over FineWeb-Edu. The project, inspired by Karpathy's nanochat, was built as a config-driven YAML framework for small LLM training.

Hacker News · security · 5d agoAI researchHN 28↑ · 2 comments1

AI Infra Summit: NVIDIA Vera Rubin and DSX Platform Advancements Showcase Energy Efficiencies of Optimizing Tokens Per Watt for AI Factories

At AI Infra Summit, NVIDIA showcased Vera Rubin and DSX gains up to 1.4x tokens per megawatt, plus Annapurna, d-Matrix, and Pinterest partnerships.

Ian Buck's AI Infra Summit keynote before 8,000+ attendees emphasized validated agentic tokens per megawatt as the emerging AI infrastructure metric. Announcements include Amazon's Annapurna Labs collaborating on NVHBM custom high-bandwidth memory, d-Matrix integrating NVLink Fusion with Raptor XPUs, and Pinterest using Blackwell plus Dynamo inference software for conversational visual discovery. Lambda reported 23% better performance per watt with DSX MaxLPS on Blackwell servers, running 19 nodes on a 16-node power budget. NVIDIA says DSX MaxLPS combined with Groq 3 LPX on Vera Rubin NVL72 targets up to 35X token throughput per megawatt versus GB200 NVL72 for 2-trillion-plus-parameter models.

NVIDIA Blog · 6h agoAI industry

Ask HN: How do you manage skills files?

A Hacker News thread debates whether agent skill files are worth managing, citing 2–4x output-token reductions on flagship models in one company's testing.

Commenters argue skills are stored prompts that help less-technical users compensate for weak prompting, while one participant reports company testing found skills reduce flagship-model output tokens by roughly 2–4x, a gap growing with newer models. Others note skills can bundle reusable scripts and inline commands for deterministic context building, and that harnesses now execute backticked commands before the agent sees the skill. Some argue improving model capability makes downloadable skills redundant.

Hacker News · security · 9d agoAI tools & infraHN 303↑ · 271 comments

ukisai/Swift-Qwen3.8-27b — new model trending #30 on Hugging Face

UkisAI releases Swift-Qwen3.8-27B, a Qwen3.8-27B derivative using 58.3% fewer thinking tokens with <1% performance loss and ~1.95x speed-up.

UkisAI released Swift-Qwen3.8-27B, a reasoning-efficient derivative of Qwen3.8-27B that cuts thinking-token usage by 58.3% while staying within 1% of base performance, yielding a 1.95x speed-up on several tasks. The model was fine-tuned by penalizing reasoning-marker tokens that trigger overthinking, plus a transfer component from BottleCap AI's ThinkingCap-Qwen3.6-27B. Benchmarks include GPQA-Diamond 88.28% (base 88.38%), MMLU-Pro 84.95% (base 85.47%), and AIME 2026 94.00% (base 98.67%), with mean-token reductions of roughly 27-46% across tests. GGUF weights are available on Hugging Face alongside enterprise licensing options.

Hugging Face trending models · 7d agoModel release

deepseek-ai/DeepSeek-V4.1-Flash — new model trending #28 on Hugging Face

DeepSeek releases DeepSeek-V4.1-Flash, a 552B-parameter multimodal MoE model with 1M-token context and KV cache cut to 890 bytes per token.

DeepSeek-V4.1-Flash is a multimodal Mixture-of-Experts model with a 552B-parameter backbone that activates 8B parameters per token during prefill and 16B during decode. It uses a Causal Encoder-Decoder architecture, Compressed Sparse Attention 2, and FP4 KV caching to reduce the global KV cache footprint to 890 bytes per token, roughly one quarter of DeepSeek-V4-Flash. The model was trained from scratch on 45T tokens with context extended to 1M tokens, includes an Engram conditional-memory module (196B parameters), and is released under the MIT license. Post-training uses SFT, RL, and on-policy distillation with large-scale automated synthesis of agentic tasks and a controllable reasoning effort setting from 1 to 100.

Hugging Face trending modelsupdated · 4d agofirst · 5d agoModel release 3 sources1

Studying Image Tokenizers as Visual Languages in Unified Multimodal Models

A controlled pure-autoregressive testbed shows task-specific validation losses rank image tokenizers differently, with I2T loss the most consistent signal.

Researchers built a controlled pure-autoregressive testbed and tracked task-specific validation losses during multimodal continual pretraining across text, image, text-to-image (T2I), and image-to-text (I2T) prediction. They find losses should be analyzed per task because they exhibit distinct scaling behavior and rank tokenizers differently, and that the loss-performance relationship depends on the predicted token space. I2T loss, computed over a shared text vocabulary, correlates consistently with both generation and visual understanding performance after supervised finetuning. Case studies revisit the discriminator, semantic supervision, and vocabulary size as tokenizer design axes.

arXiv cs.AI / cs.LG / cs.CL · 7d agoAI research1

BigBear 2.0 Evilginx2 Phishing Campaign Bypasses Microsoft 365 MFA With Session Cookie Theft

CloudSEK identified BigBear 2.0, an Evilginx2-based AiTM phishing operation stealing Microsoft 365 MFA session cookies, hitting 461 organizations across 40-plus countries.

The campaign proxies Microsoft sign-in pages to capture credentials and authenticated session cookies, enabling session replay into email, Teams, SharePoint, OneDrive, and connected SSO applications. CloudSEK's June 2026 discovery found 5,137 stolen records, 1,032 passwords, and 4,148 session cookies tied to 3,331 victim IPs, linked to operator 'General Boss' across 42 VPS nodes. The operation targeted IT services and managed service providers, used country-matched residential proxies, and involved at least five affiliates.

Cyber Security News · 7d agoPhishing & fraud in the wild1

Studying Image Tokenizers as Visual Languages in Unified Multimodal Models

A controlled autoregressive testbed shows validation losses must be analyzed per task, and image tokenizer choice affects joint multimodal text modeling.

Researchers built a pure-autoregressive testbed to study image tokenizers as the 'visual language' of unified multimodal models, tracking task-specific validation losses during multimodal continual pretraining across text, image, text-to-image (T2I), and image-to-text (I2T) prediction. They found that losses exhibit distinct scaling behavior per task and rank tokenizers differently, and that I2T loss over a shared text vocabulary gives a more consistent loss–performance signal than T2I loss. Better reconstruction does not necessarily yield lower task-specific losses or stronger downstream performance, and tokenizer choice can affect text modeling under joint optimization. Case studies examine the discriminator, semantic supervision, and vocabulary size design axes.

Hugging Face daily papers · 8d agoAI research1