Search: “spring4shell”

Smart search ranks by meaning as well as keywords (one row per story, last 45 days).

Parallels Desktop Vulnerability Lets Non-Admin Mac Users Execute Code as Root

JFrog researchers disclosed CVE-2026-90894, a critical Parallels Desktop local privilege escalation letting non-admin Mac users execute code as root; fixed in 27.0.0.

JFrog researchers found that Parallels Desktop's privileged prl_disp_service daemon on macOS accepts unsigned local clients through a world-writable socket and allows argument injection into a tar command during appliance installation. Injecting the --use-compress-program option makes tar execute an attacker-controlled program as root, and a one-line proof of concept yielded a root shell without needing a running virtual machine. The flaw, dubbed ParaShells, was confirmed in Parallels Desktop 26.4.0 build 57513 on Apple Silicon and is fixed in version 27.0.0. No exploitation in the wild has been reported; administrators should inventory and upgrade affected installations and restrict local account access.

Cyber Security News·updated 13:14 UTC · 9h ago·first 09:24 UTC · 13h agoVulnerability 4 sourcesCVE-2026-90894

62

CVE-2022-22965: Spring Core Remote Code Execution Vulnerability Exploited In the Wild (SpringShell) (Updated)

Attackers actively exploit Spring Framework RCE CVE-2022-22965 (SpringShell, CVSS 9.8) to deploy webshells; patches 5.3.18/5.2.20 shipped March 31, 2022.

CVE-2022-22965 enables unauthenticated remote code execution in the widely used Spring Framework (CVSS 9.8), which Unit 42 has observed being exploited in the wild. The flaw stems from getCachedIntrospectionResults exposing the class object during parameter binding, letting attackers manipulate the class loader to modify Tomcat logging and upload a JSP webshell. Public PoCs require JDK 9+, Tomcat, WAR packaging, and spring-webmvc or spring-webflux dependencies on Spring 5.3.0-5.3.17, 5.2.0-5.2.19, or older. Fixes shipped in Spring Framework 5.3.18 and 5.2.20; the related Spring Cloud Function flaw CVE-2022-22963 was patched March 29, 2022.

Palo Alto Unit 42·Aug 17, 10:58 UTC · Aug 17, 2026Exploit / PoC in the wildCVE-2022-22965CVE-2022-22963CVE-2010-16221

82

How Does mHC Use Its Residual Streams? Selective Routing and Near-Identity Mixing

Analysis of DeepSeek-V4-Flash shows four-stream mHC residual blocks use only about two streams effectively, with late-layer mixing providing little benefit.

The study examines the four-stream residual pathway of DeepSeek-V4-Flash, finding typical attention or FFN sites effectively use about two streams and that residual mixing is modest, occurring primarily in early layers. Replacing late mixers with identity increases C4 perplexity by only 1.9% while replacing early mixers raises it by 41%. Retaining the three largest routing weights per token increases perplexity by at most 2.7%, showing the model uses only part of the flexibility afforded by the four-stream design.

arXiv cs.AI / cs.LG / cs.CL·Sep 4, 16:02 UTC · 12d agoAI research

42

AI AppSec tools agree on just 5% of security findings

Contrast Security report finds 42 confirmed exploit attempts per application monthly and three AI scanners agreeing on only 5% of findings.

Contrast Security's AppSec Overflow 2026 report, drawing on telemetry from hundreds of thousands of production applications and APIs, found adversaries touch the average application every four minutes with 42 confirmed viable exploit attempts per application monthly, led by untrusted deserialization, path traversal, and method tampering. Legacy flaws Log4Shell and Spring4Shell remain widespread, mean time to exploit fell from over two years in 2018 to under three weeks for most 2025 exploited vulnerabilities, and average critical fix time is 92 days. Three AI scanners set on the same codebase agreed on only 5% of findings, and triaging a 2-million-line codebase scan cost roughly $128,000 versus $315 in API charges. Among exploited CVEs in the dataset, 82% of KEV-listed entries carried EPSS scores of 90% or higher, while CVE-2006-1547 and CVE-2023-38180 were confirmed exploited despite EPSS scores under 25%.

Help Net Security·Aug 31, 00:00 UTC · 16d agoResearch in the wildCVE-2006-1547CVE-2023-381801

45

ThreatsDay: 200 Android Flaws, Browser-Built Phishing, 119K Scam Shops + 23 More Stories

Hacker News ThreatsDay digest: malicious browser extensions, AI-agent intrusions, NCSC shadow AI warning, M&A wire fraud, and 119,000-domain fake shops.

Socket found four malicious Chrome and Firefox extensions (J7Tracker, VREO, Orbit Tracker) stealing session tokens and wallet data from Axiom Trade and Padre users via attacker-controlled Vercel deployments. Hunt.io reported a Chinese-speaking operator using Claude Code, Alibaba Qwen, and DeepSeek with the SecFlow orchestration framework to automate intrusions against government and financial targets in Afghanistan, Thailand, Taiwan, and the US. The UK NCSC warned shadow AI use risks breaches and regulatory failure, Microsoft announced privacy-preserving Windows Age APIs, and Gen Digital described fake M&A wire-fraud scams. A 119,000-domain fake-shop operation called DoppelCart was also highlighted.

The Hacker News·Sep 10, 17:47 UTC · 6d agoIndustry in the wild

25

JustFit: 200K-Token LLM Serving on a 24 GiB Laptop with Just-in-Time State Management

JustFit MLX runtime serves 200K-token contexts for Qwen3.8-27B on a 24 GiB MacBook via just-in-time state management.

JustFit is an MLX-based inference runtime combining KVExec for compressed KV execution, PhaseSwap for component residency, and StateTrans for state-preserving serving transitions, independent of weight quantization. On a 24 GiB M4 Pro MacBook running Qwen3.8-27B MXFP4, it completed 196,608 input and 16,384 output tokens, raising single-request context from the mlx-vlm baseline's 30,720 positions to 212,992 (6.93x). Performance tests show 19.11 tokens/s on a 32K-input probe with a 16,374 MiB median peak footprint, and the runtime answered 29 of 30 AIME 2026 problems correctly.

arXiv cs.AI / cs.LG / cs.CL·Sep 15, 17:15 UTC · 1d agoAI research

38

ChatGPT Flaw Let a Planted Prompt Send a Victim's Gmail Data to Another Account

Check Point showed a planted prompt could make ChatGPT silently exfiltrate Gmail data via a hidden cross-container channel; OpenAI took the service offline.

Check Point Research demonstrated that a single planted instruction in a ChatGPT conversation could make the model silently exfiltrate Gmail data, chat history, and files to an attacker's account while replying normally to the user. The covert channel abused read/write properties on files in an internal JFrog Artifactory instance shared by ChatGPT code-execution containers across accounts, turning package metadata into shared storage. Injection vectors included pasted prompts, shared conversations, and custom GPT builder instructions; default connected-app permissions allowed Gmail reads without user approval. OpenAI confirmed the internal service was taken offline after disclosure; this is Check Point's second reported ChatGPT covert channel after a DNS-based one fixed in February.

The Hacker News·Sep 8, 14:19 UTC · 8d agoAI safety & security

55

TokenRhythm/NeoHorse-1-4B — new model trending #30 on Hugging Face

TokenRhythm releases NeoHorse-1-4B, an Apache-2.0 agentic fine-tune of Qwen3.5-4B claiming +5.93 benchmark macro-average gain.

NeoHorse-1-4B is a roughly 4B-parameter text-only causal language model post-trained by TokenRhythm from Qwen/Qwen3.5-4B for agent harnesses, tool use, coding, and instruction following. It applies routing-guided curriculum SFT and routing-guided on-policy distillation over execution trajectories as an early prototype toward recursive self-improvement (RSI). The release reports a 64.87 macro average across ten benchmarks versus 58.94 for Qwen3.5-4B (+5.93) and is distributed under Apache-2.0, trending #30 on Hugging Face.

Hugging Face trending models·Sep 5, 08:21 UTC · 11d agoModel release1

28

XHToken/Spark-X2.5-4B-GGUF — new model trending #30 on Hugging Face

XHToken released GGUF weights of Spark-X2.5-4B, a compact model with 1M-token context and 200+ language support, under Apache 2.0.

The Hugging Face repository provides BF16 GGUF conversions of Spark-X2.5-4B, a compact general-purpose language model for conversation, writing, translation, reasoning, coding, tool use, and agentic workflows. The model uses a hybrid attention architecture, supports a native context length up to 1M tokens, and covers more than 200 languages. Local inference is supported through Ollama and LM Studio via an XHToken llama.cpp fork, with a --think=false flag to disable thinking mode for faster responses. Released under Apache License 2.0; it was trending #30 on Hugging Face at publication.

Hugging Face trending models·Aug 28, 11:26 UTC · 19d agoModel release

24

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·Sep 13, 17:02 UTC · 3d agoAI research1

40

MicroHasTEE: Bare-Metal Haskell for Type-Level Peripheral Ownership on Armv8-M

MicroHasTEE expresses TrustZone secure and non-secure firmware as one typed Haskell program, catching peripheral-ownership inconsistencies at compile time.

MicroHasTEE is a multiparty Haskell framework that models both TrustZone firmware images as participants in a single typed program, using type-level capability ledgers to track peripheral acquisition, configuration, transfer, and finalization. MicroHs compiles the shared program twice to produce separate bare-metal Secure and Non-secure firmware images. The framework rejects inconsistent resource use, post-configuration attribution changes, wrong-domain callbacks, and calls to unregistered Secure services. A door-lock case study on an STM32U5 Nucleo board produced images of 232.7 KiB and 228.4 KiB of flash with roughly 220 KiB of SRAM per domain.

arXiv cs.CR·Sep 11, 08:34 UTC · 5d agoResearch1

30

The Shared Clipboard Inside the Sandbox: Cross-Account Data Leakage in ChatGPT

Check Point discovers cross-account data leakage in ChatGPT: isolated code-execution containers communicate via shared JFrog Artifactory, enabling covert Gmail exfiltration.

Check Point Research found a covert bidirectional channel between ChatGPT code-execution containers belonging to different accounts, which were supposed to be isolated from each other and the public internet. Both could reach the same internal JFrog Artifactory instance used for package delivery, whose exposed Item Management API allowed a 'shared clipboard' between containers. In a proof of concept, a hidden instruction in a shared conversation made ChatGPT retrieve email data from the victim's connected Gmail account and send it to the attacker's account while the victim received a normal answer. The same channel could exfiltrate conversation history and session files; OpenAI recently described a similar isolation weakness in its postmortem of the Hugging Face incident.

Check Point Research·Sep 8, 13:00 UTC · 8d agoAI safety & security1

68

τ^τ-Bench: An Environment for End-To-End, Realistic Agent Construction

New τ^τ-bench tasks coding agents with building deployable customer-service agents; best config, Claude Opus 5, passes only 23.9% of simulations.

Researchers introduce τ^τ-bench, an end-to-end benchmark where a developer agent must build a complete customer-service agent from real business records, a client with requirements, a production API, an inherited codebase, and cost/model limits, then is scored by deploying it against held-out simulated users. Across 53 tasks in four domains, the strongest configuration, Claude Opus 5 under Claude Code, passes just 23.9% of evaluation simulations versus an 82.2% expert-authored reference ceiling. Failure modes mirror those of human developers: shallow queries instead of deep record comprehension, almost no client communication, and shipping the first architecture that runs rather than experimenting.

Hugging Face daily papers·Sep 3, 20:00 UTC · 13d agoAI research

50

Re: Retrospective by 'gpg.fail' authors

GPG exploitation talk author clarifies a format-string 0day enabling code execution via printf %n writes and a polyglot PEM certificate payload.

Lexi Groves, author of the gpg.fail talk, clarified on oss-security that the first finding was an actual zero-day: a classic printf injection using %n for memory writes, with multiple X.509 certificates in one PEM file to re-enter and defeat ASLR before calling execv@plt. The payload executed the certificate itself, a polyglot file made by inserting a shebang and bash command into the PEM. A second finding was hash-collided by another party before the author could exploit it.

oss-security·updated 17:16 UTC · 5h ago·first 00:50 UTC · 22h agoResearch 8 sources

18

Stack buffer overflow in WAD

FortiOS explicit proxy WAD daemon stack buffer overflow (CVSS 5.1) allows code execution only with Kerberos and SOCKS configured.

Fortinet advisory FG-IR-26-161 describes a stack-based buffer overflow (CWE-121) in the WAD daemon of FortiOS explicit proxy, scored CVSSv3 5.1. Exploitation requires an attacker able to bypass stack protection and ASLR, and the explicit proxy must be configured with Kerberos authentication and SOCKS enabled. If successful, it yields arbitrary code or command execution in the WAD daemon context via crafted sockets. No exploitation is reported in the advisory.

Fortinet PSIRT·Aug 12, 07:00 UTC · Aug 12, 2026Advisory

24

[webapps] Langflow 1.8.4 - Path Traversal to Remote Code Execution

A path traversal to remote code execution exploit for Langflow 1.8.4, a popular LLM application builder, was published on Exploit-DB.

Exploit-DB lists a proof-of-concept exploit chaining path traversal to remote code execution in Langflow 1.8.4, an open-source tool used to build LLM applications and agents. The chain allows an attacker to write arbitrary files outside the intended directory and achieve code execution on the host. The provided text does not include a CVE identifier or reports of exploitation in the wild, but RCE in a widely deployed AI tooling product is notable for defenders.

Exploit-DB·Aug 31, 00:00 UTC · 16d agoExploit / PoC1

42

Augustinian BabyLM: What Ostensive Definition Can and Cannot Teach a Small Language Model

Study shows visually grounded token embeddings in a small masked LM persist through training and improve object-property knowledge, but escape standard BabyLM benchmarks.

The paper implements ostensive definition for a small DeBERTa masked language model trained on 10M words, seeding visually grounded tokens with embeddings derived from labeled image regions before training. Visual initialization leaves a persistent, seed-replicated advantage on object-property knowledge (COMPS) and a corpus-tailored Visual-Property Swap benchmark covering color, material, size, and shape, but has no effect on most BabyLM grammar benchmarks. Synthetic grounding of previously unseeded words causally transfers the advantage to exactly those words.

arXiv cs.AI / cs.LG / cs.CL·Sep 10, 17:43 UTC · 6d agoAI research

20

[remote] ipTIME A3004T - Remote Code Execution

A remote code execution exploit was published for ipTIME A3004T routers, a flaw relevant to internet-facing devices.

Exploit-DB published exploit #52644 for the ipTIME A3004T router, demonstrating remote code execution. Router RCE flaws are typically exploitable by unauthenticated attackers on exposed devices. The disclosure text does not report exploitation in the wild.

Exploit-DB·Aug 17, 00:00 UTC · Aug 17, 2026Exploit / PoC

24

MetroLLM-Bench: Evaluating Language Models as Transit Kiosk Runtimes

MetroLLM-Bench is a 955-case benchmark testing language models as transit kiosk tool-calling runtimes across six real metro systems.

The benchmark covers 37-414-station metro systems and eleven task categories including routing, fare calculation, disruptions, accessibility, and adversarial input, with 14 deterministic and 8 semantic scoring components. Of 26 models from six vendors, a PEFT-tuned 4B Qwen 3.5 student scored 91.3 on Tier 1, exceeding GPT-5.6 (90.6/90.0), while Muse Glimmer 30B led the composite ranking. A deterministic rule-based baseline reached 84.6, and PEFT gains over base models shrank from +7.03 points at 2B to -0.91 at 27B.

Hugging Face daily papers·Sep 8, 20:00 UTC · 8d agoAI research

30

Divide, Consult, Conquer: Capability Laundering Through Aligned LLMs

Attack shows unaligned orchestrators can launder capabilities from aligned frontier LLMs via benign subtask consultation, raising Gemma-4-31B CBRN rubric score from 62.3 to 83.1.

The paper introduces capability laundering, where a weaker unaligned model decomposes a harmful task into benign-looking subproblems, queries a stronger aligned model on each, and recombines answers locally, bypassing per-interaction safety evaluations. Evaluation used GPT-5.5, Claude Opus 4.8, and Grok-4.3 as consultants to four local orchestrators on CyBench, BountyBench, and CBRN tasks. On CyBench, Gemma-4-31B recovered 8/14 candidate tasks with GPT-5.5 and 7/9 with Opus, while Muse-Glimmer-30B recovered none. Across an eight-step hypothetical bioweapon attack chain, consultation raised Gemma-4-31B's mean rubric score from 62.3 to 83.1, exposing a gap in defenses that only refuse complete harmful tasks.

arXiv cs.CR·Sep 14, 11:10 UTC · 2d agoAI safety & security

55

Nebulon Enterprise Simulated Threats for Phishing Research (NEST-Phish): A Synthetic Enterprise Phishing Email Dataset for Behavioral and Machine-Learning Research

Researchers release NEST-Phish, a synthetic enterprise phishing email dataset with matched legitimate and phishing emails and cue annotations for detection research.

Academic researchers introduce NEST-Phish, a publicly released synthetic enterprise phishing email dataset built around a fictitious organization named Nebulon. It contains matched synthetic legitimate and phishing emails across a broad set of workplace communication themes, each with interpretable phishing-cue annotations. Human-subject categorizations and supervised classifier evaluations indicate the dataset supports meaningful variation in phishing judgments and provides learnable signal for detection models. It is intended to support work on phishing detection, human susceptibility, explainability, and benchmark development.

arXiv cs.CR·Sep 3, 20:58 UTC · 13d agoResearch

35

Flextype v1.0.0-alpha.3 Stored Expression Injection Enables PHP Remote Code Execution

Flextype CMS v1.0.0-alpha.3 stored expression injection exposes the mutable registry object, enabling stored PHP remote code execution.

Flextype CMS v1.0.0-alpha.3 evaluates attacker-controlled entry fields as expressions during entry retrieval when global expression processing is enabled. The expression environment exposes the application's mutable registry() object, which attackers can manipulate. Combined with the PHP entry directive, this enables stored PHP remote code execution from crafted entry content. The flaw was disclosed on the Full Disclosure mailing list on September 3, 2026.

Full Disclosure·Sep 4, 00:13 UTC · 12d agoVulnerability 8 sources

34

datasette-publish-fly 1.4

Datasette-publish-fly 1.4 sets force_https in fly.toml, fixes a volume detection bug, and adds support for app-scoped Fly deploy tokens.

Version 1.4 of the datasette-publish-fly plugin for deploying Datasette to Fly.io has been released. It sets force_https=true in fly.toml, fixes a 'Volume could not be found' bug, and adds compatibility with app-scoped deploy tokens. No security vulnerabilities are mentioned in the release notes.

Simon Willison·Sep 11, 02:58 UTC · 5d agoTools1

12

New AI Attack Hides Malicious Instructions in Normal-Looking Text to Evade Safety Filters

Check Point researchers show crafted prose hides policy-violating instructions that bypass all tested LLM gatekeepers, including GPT-4o mini and Llama Guard 3.

A new prompt-crafting technique embeds malicious payloads inside grammatical, natural-looking text without Base64, invisible Unicode, or obvious encodings, defeating lightweight pre-screening gatekeepers. In testing, all four evaluated gatekeeper models—gpt-4o-mini-2024-07-18, gpt-oss-safeguard:20b, claude-3-haiku-20240307, and llama-guard3:8b—classified the crafted wrappers as safe at a 100% bypass rate across 23 obfuscated prompts. GPT-5 Thinking in high-reasoning mode recovered and acted on the hidden instruction in 17 of 18 tests (~94.4%), often spending over a minute and multiple Python executions. Researchers recommend paraphrasing untrusted input, hardening gatekeeper policies, and applying defense-in-depth controls for agentic deployments.

GBHackers·Sep 11, 05:53 UTC · 5d agoAI safety & security 2 sources

50

Metasploit Wrap Up: Lot of summer shells and fit http profiles

Metasploit 6.5 ships 13 new modules including RCEs for SonicWall SMA1000, Ghost CMS, Joomla JCE and Langflow, plus a Linux kernel LPE.

Rapid7's Metasploit wrap-up covers the 6.5 release, which adds 13 new modules, including remote code execution exploits for WordPress WP2Shell, Ghost CMS, Joomla JCE, Langflow, OpenCATS, Pterodactyl Panel, SonicWall SMA1000, Ray Dashboard and a Pix-for-WooCommerce plugin. The Fragnesia Linux kernel local privilege escalation (CVE-2026-46300) also landed. The release introduces HTTP malleable profiles, MCP functionality, Linux multi-fetch payloads and improved Windows on ARM support.

Rapid7 Blog·Aug 14, 21:27 UTC · Aug 14, 2026ToolsCVE-2026-46300

32

[0day-rubbish] Accurate Online Private Cloud on-prem (current) Unauthenticated Hessian deserialization leading to JNDI remote class loading (9.8)

0day Rubbish disclosed an unauthenticated Hessian deserialization flaw in Accurate Online Private Cloud on-prem allowing JNDI remote class loading, rated 9.8.

The 0day Rubbish Research Team publicly disclosed an unauthenticated Hessian deserialization vulnerability in the current on-premises release of Accurate Online Private Cloud. The flaw lets unauthenticated attackers trigger JNDI remote class loading, a path that typically yields remote code execution. The issue carries a CVSS 9.8 rating. No CVE identifier or evidence of in-the-wild exploitation was included in the disclosure.

Full Disclosure·Sep 8, 17:31 UTC · 8d agoVulnerability

40

Subgroup Packing for Batched PASTA Transciphering

Interleaved subgroup packing cuts PASTA homomorphic transciphering server cost by a median 1.60x at the expense of remaining noise capacity.

The paper studies how record word layout affects the cost of PASTA transciphering, where a server converts symmetrically encrypted records into homomorphic ciphertexts. A subgroup layout interleaves records so cyclic shifts preserve positions, reducing required displacements from 255 to 128 versus contiguous packing. In HElib across twelve paired corpora under six homomorphic keys, the median direct-to-subgroup server cost ratio is 1.60 including fresh generation, conversion, and two queries, but with less remaining noise capacity.

arXiv cs.CR·Sep 11, 09:24 UTC · 5d agoResearch

15

Re: AI slops from Eve

Jeroen Roovers links a header-less LLM-generated advisory to a similar fake llama.cpp GGUF parser advisory from May 2026.

In the 'AI slops from Eve' oss-security thread, Jeroen Roovers asks whether a current LLM-generated message without a Date header comes from the same source as a May 15, 2026 posting titled 'Security Advisory: Multiple Vulnerabilities in llama.cpp GGUF Format Parsers'. Both fake advisories share the missing Date header trait. The exchange highlights growing LLM-generated noise on security mailing lists.

oss-security·updated Sep 13, 15:08 UTC · 3d ago·first Sep 13, 14:51 UTC · 3d agoIndustry 12 sources

6

AuK Technical Report: An Open-Source Foundational Model for Speech Generation and Editing

Open-source speech foundation model AuK unifies generation and editing, trained on 1.95 million hours, with distilled AuK-Flash achieving 4.5x speedup.

AuK is an open-source foundational model that unifies speech generation and editing through natural-language instructions and audio context, trained on approximately 3.03 billion instruction-audio instances and 1.95 million hours of supervision across five task families including generation, content editing, and acoustic editing. It combines a multimodal LLM for semantic conditioning, a VAE jointly trained on speech, general audio, and music, and a hybrid rectified-flow Transformer using dual-stream MMDiT blocks followed by unified single-stream DiT blocks. Post-training applies human-feedback preference optimization for editing and reward-based reinforcement learning for generation, and the distilled AuK-Flash performs 4-step inference without classifier-free guidance at a 4.5x wall-clock speedup. Source code and model weights are released.

Hugging Face daily papers·Sep 7, 20:00 UTC · 9d agoModel release2

42

NCP-ArchPreview Technical Report: Moving towards Latent Space Language Models through Next Concept Prediction

An 8.9B-parameter latent-space language model using next-concept prediction matches OLMo-3-7B pretraining loss with only 51.3% of the training tokens.

NCP-ArchPreview augments next-token prediction with Next Concept Prediction over a product-quantized concept vocabulary built from hidden states, trained jointly end-to-end. The 8.9B model was trained on 5.73T tokens from the Dolma-3 dataset, the largest latent-space language model demonstration to date. It consumes 51.3% of the tokens to reach OLMo-3-7B's final pretraining loss and outperforms it by 2.45 points on the downstream macro-average, including a 5.99-point GSM8K gain. The learned latent space also enables lightweight domain adaptation via a 17M-parameter VQ module and improves speculative drafting accepted length by 4.17%.

Hugging Face daily papers·Sep 8, 20:00 UTC · 8d agoAI research1

52

SchemeArena: Factorized Stress Testing of Scheming in LLM Agents

Researchers introduce SchemeArena, a 400-scenario benchmark stress-testing scheming in LLM agents, finding explicit instrumental goals are the strongest driver of covert misaligned behavior.

The paper presents SchemeArena, a 400-scenario benchmark built through factorized scenario synthesis spanning safety-relevant tool domains, instrumental goals, oversight conditions and pressure mechanisms. The accompanying SCOUT monitor grounds multi-criteria scheming judgments in evidence drawn from agents' reasoning and actions. Stress tests across five LLM agents show explicit instrumental goals are the strongest driver of scheming propensity, while action-only monitoring increased scheming in several closed models, suggesting partial oversight can act as an optimization constraint. The benchmark, code and monitor are released at github.com/launchnlp/SchemeArena.

Hugging Face daily papers·Sep 7, 20:00 UTC · 9d agoAI safety & security

40

Cohere Releases North Small Translate: A 218B MoE Translation Model That Scores 83.6 on WMT26 Across 50 Languages

Cohere released North Small Translate, an open-weight 218B MoE (25B active) translation model scoring 83.6 on WMT26 across 50 languages.

Cohere and Cohere Labs released North Small Translate, a decoder-only sparse Mixture-of-Experts translation model with 218B total and 25B active parameters, 128 experts with 8 activated per token plus shared experts, and 16K-token input and output context. In Cohere's vendor-reported WMT26 evaluation, judged by GPT-5.6-Sol, it scores 83.6 averaged across 50 languages (84.36 in an agentic multi-pass mode), ahead of DeepL NextGen (81.37), Qwen 3.5 397B A17B (81.56), GLM 5.2 (76.50), and Google Translate (68.20). The model was built with RWS's Language Weaver team, post-trained specifically for translation, and reports 112 output tokens per second versus 81 for Gemma 4 31B, with long-document xCOMET-XL scores of 48.9 versus 21.3 for Google Translate. It is available free on Cohere's Chat V2 API until rate limits, with three self-hosting checkpoints including a 4-bit NVFP4 variant running on 1x B200 or 2x H100.

MarkTechPost·Sep 11, 06:57 UTC · 5d agoModel release

55