Search: “ICONICS”

Sure, Meta’s AI Muse works, but it sure creeps me out

Hands-on review finds Meta's Muse AI agent completes shopping and email tasks but surfaces personal Instagram API data beyond user-visible ad-topic settings.

Meta launched Muse, its first agentic AI productivity assistant, which performs tasks like shopping, email management, trip planning, media generation, and creating webpages or documents via a cloud-based virtual computer. The Verge's hands-on found it successfully deleted thousands of promotional emails and completed an Amazon purchase, but it also revealed detailed personal interests inferred from Instagram and Facebook account API data that is not visible in the apps' ad-topic settings. Meta says Muse only exchanges data needed for third-party integrations and does not share information with advertisers; the reviewer frames privacy unease as the main adoption hurdle.

The Verge · AI·updated 13:01 UTC · 5h ago·first Sep 10, 15:00 UTC · 6d agoAI industry 11 sources

Multiple Vulnerabilities Discovered in a SCADA System

Unit 42 details five vulnerabilities (CVSS 7.0-7.8) in ICONICS Suite SCADA software enabling privilege escalation and DoS.

Unit 42 discovered five vulnerabilities (CVE-2024-1182, CVE-2024-7587, CVE-2024-8299, CVE-2024-8300, CVE-2024-9852) in ICONICS Suite versions 10.97.2 and earlier for Windows during a 2024 security assessment. The flaws, rated CVSS 7.0-7.8, allow DLL hijacking, privilege escalation, information disclosure, denial-of-service and potentially full system compromise. ICONICS Suite is a SCADA solution with hundreds of thousands of installations in over 100 countries, widely used in critical infrastructure, and several dozen servers are internet-exposed per Unit 42 telemetry. ICONICS released patches and advisories with workarounds in 2024.

Palo Alto Unit 42·Aug 19, 12:04 UTC · 28d agoVulnerabilityCVE-2024-1182CVE-2024-7587CVE-2024-8299+2 CVEs1

Multi-Vector (Late Interaction) Embedding Models with Sentence Transformers

Hugging Face details building and using multi-vector late-interaction embedding models with Sentence Transformers for retrieval workloads.

Hugging Face published a guide on multi-vector, late-interaction embedding models (ColBERT-style) supported through Sentence Transformers. The post covers how practitioners can build and use these models for retrieval and RAG pipelines. It is a developer tooling and technique write-up, not a security advisory.

Hugging Face Blog·Aug 18, 00:00 UTC · 29d agoAI tools & infra1

Privileged File System Vulnerability Present in a SCADA System

Unit 42 details CVE-2025-0921 (CVSS 6.5), a privileged file operations flaw in Iconics Suite enabling DoS and privilege escalation.

Unit 42 disclosed CVE-2025-0921 (CVSS 6.5), an execution-with-unnecessary-privileges flaw in the Pager Agent of the AlarmWorX64 MMX feature of Mitsubishi Electric Iconics Digital Solutions GENESIS64. Attackers could misuse privileged file system operations to corrupt critical binaries, causing denial-of-service or integrity loss on vulnerable SCADA systems. The analysis demonstrates a chain with CVE-2024-7587, which grants excessive permissions to the C:\ProgramData\ICONICS directory via the GenBroker32 installer. Iconics released an advisory with a workaround that addresses the reported issues.

Palo Alto Unit 42·Aug 19, 12:04 UTC · 28d agoVulnerabilityCVE-2025-0921CVE-2024-7587

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·Sep 8, 17:57 UTC · 8d agoAI research1

Why Johnny Can't Encrypt: A Usability Evaluation of PGP 5.0 (1999)

Seminal 1999 USENIX study finds most novice users cannot correctly sign and encrypt email with PGP 5.0 in 90 minutes.

Whitten and Tygar's USENIX Security Symposium paper evaluates whether cryptography novices can use PGP 5.0 effectively, using cognitive walkthrough analysis and a laboratory user test. The majority of test participants failed to successfully sign and encrypt a message within 90 minutes, despite PGP 5.0 having a well-regarded graphical interface. The authors argue that security requires usability standards beyond those of general consumer software and propose domain-specific UI design principles for security. The paper is a foundational reference in usable security research.

Lobsters · security·Sep 9, 06:47 UTC · 7d agoResearch

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·Sep 7, 20:00 UTC · 8d agoAI research1

Threat Brief: 3CXDesktopApp Supply Chain Attack (Updated)

Trojanized 3CXDesktopApp installers for Windows and macOS loaded RC4-encrypted shellcode delivering a backdoor to hundreds of thousands of 3CX users.

Threat actors compromised the 3CXDesktopApp build process, adding malicious libraries (ffmpeg.dll, d3dcompiler_47.dll) to installers downloaded from the developer's website. The Windows MSI loader decrypts embedded shellcode with RC4, sleeps for a random 1-4 weeks, then beacons to C2 hidden in icon files on GitHub. Unit 42 fingerprinted 247,277 IPs across 199 countries using 3CX applications and blocked shellcode execution at 127 Cortex XDR customers between March 9-30, 2023.

Palo Alto Unit 42·Aug 17, 12:18 UTC · Aug 17, 2026Threat actor in the wild1

Think Before You Link: Rarity, Reasoning, and Retrieval in Multilingual Entity Linking

Training-free reasoning-plus-retrieval framework lifts multilingual multimodal entity linking accuracy by 6.9% overall and up to 23.3% on rare entities.

The paper broadens rarity measurement in multimodal entity linking using knowledge-graph structural metrics beyond popularity-based pageview metrics, identifying many rare entities that popularity metrics miss. Across these rare-entity slices, state-of-the-art accuracy drops by 15.4-39.9%, exposing distinct failure modes under different rarity definitions. The proposed training-free framework iteratively searches and reasons over Wikipedia with a reasoning-capable vision-language model; experiments show reasoning and retrieval are complementary. On the MERLIN benchmark covering Hindi, Indonesian, Japanese, Tamil, and Vietnamese, the system improves overall accuracy by 6.9% and up to 23.3% on rare entities, with MERLIN-Rare test slices released.

Hugging Face daily papers·Sep 8, 20:00 UTC · 7d agoAI research

Windows 11 cumulative updates KB5124008 & KB5122880 released

Microsoft shipped mandatory September 2026 Patch Tuesday cumulative updates KB5124008 and KB5122880 for Windows 11, fixing vulnerabilities and adding features.

Microsoft released Windows 11 KB5124008 (versions 25H2/24H2) and KB5122880 (23H2) cumulative updates containing the September 2026 Patch Tuesday security patches, described as covering 1,000 vulnerabilities discovered in previous months. The updates are mandatory and add features including movable and resizable taskbars, Start menu customization, a Windows Search option to hide Bing and Store web suggestions, and Administrator Protection, a just-in-time admin privilege feature that is off by default and configurable via Intune OMA-URI or Group Policy.

BleepingComputer·Sep 8, 17:57 UTC · 8d agoAdvisory

Meme Coin Factories: Uncovering Large-Scale Manipulations on pump.fun

Large-scale pump.fun study of 15 million meme coins identifies five manipulation classes including wash trading and a Market-Manipulation-as-a-Service ecosystem.

Researchers analyzed all 15 million coins launched on pump.fun over the last two years plus large random samples of transaction data, identifying five manipulation classes: wash trading, creator address obfuscation, coordinated sells, copycat coins, and social media manipulation. Strategic actors bypass the platform interface and implement strategies in a highly automated, low-latency way by interacting directly with the blockchain. The study also uncovers Market-Manipulation-as-a-Service (MMaaS) third-party tools that let non-technical users run these manipulations, and proposes mitigations for traders, pump.fun, and regulators.

arXiv cs.CR·Sep 9, 14:35 UTC · 7d agoResearch

MILE TEA: Cyber Espionage Campaign Targets Asia Pacific Businesses and Government Agencies

Unit 42 names MILE TEA, a cyber-espionage campaign since 2011 targeting Japanese and Taiwanese businesses and government agencies with e-ticket phishing lures and Elirks-family malware.

Unit 42 tracks the MILE TEA espionage campaign, observed as early as 2011, targeting Japanese trading, petroleum, and mobile companies, a Beijing office of a Japanese public organization, and a Taiwanese government agency. The primary infection vector is spear-phishing emails with attachments, mostly custom executable installers posing as flight e-tickets, dropping Elirks, Micrass, or Logedrut as initial bridgehead malware. Elirks and Logedrut retrieve encrypted C2 addresses from attacker-posted blog articles, decoded with Base64 and TEA or DES ciphers. The campaign's focus shifted from Taiwan to Japan around 2013.

Palo Alto Unit 42·Aug 17, 12:56 UTC · Aug 17, 2026Threat actor in the wild1

Try Google Pics: Easy image creation and editing in Google Workspace

Google rolls out Pics, a Nano Banana-based image creation and editing tool, to Workspace subscribers with integrations in Docs, Slides, and Drive.

Google Pics, built on the Nano Banana image generation model, is rolling out to Google AI Pro and Ultra subscribers and most Workspace business customers. Features include object segmentation, in-image text editing and translation, collaboration, and multiple generations per prompt. Integrations launched in Docs and Slides, with Drive support planned in the coming weeks.

Google · AI·Sep 1, 16:00 UTC · 15d agoAI industry

Grand Theft Auto VI hype leads to malware

Fake GTA6 ISO downloads spread via SEO poisoning, torrents, and forums deliver NJRAT, DCRAT, an infostealer, and wiper ransomware.

Huntress analyzed a malicious ISO masquerading as a leaked Grand Theft Auto VI build; there is no official GTA6 demo or leaked playable copy. The fake gta6installer.exe shows a Russian 'license' message asking victims to email the attacker if the crack fails, then drops GTA6-branded files into %TEMP%. It installs multiple repurposed NJRAT copies (2023-era files) connecting to AWS-hosted IPs 35.157.111.131, 3.68.56.232, 3.67.15.169 and an ngrok tunnel, plus DCRAT communicating with a0700877.xsph.ru (141.8.197.42). The bundle also includes an infostealer and ransomware used as a wiper, in what appears to be an opportunistic, everything-at-once campaign.

Huntress·Sep 9, 14:00 UTC · 7d agoMalware in the wild

Show HN: LLM Attention Visualization

A developer released a browser-based tool that visualizes which past tokens influence each LLM output token using aggregated, value-weighted attention scores.

A Show HN project presents a React application built on Transformers.js that renders per-token attention influence by aggregating attention weights scaled by value-vector magnitudes across all attention heads and layers. To expose internal tensors, the author instrumented the ONNX computation graph, hosted a modified model on Hugging Face, and pre-generated prompts to avoid long model downloads in the browser. Demos with a 600-million-parameter model show how verbatim copying draws heavily on source tokens and how single outputs blend information from multiple phrases.

Hacker News · AI·Sep 8, 16:59 UTC · 8d agoAI tools & infraHN 41↑ · 9 comments

Bisonal Malware Used in Attacks Against Russia and South Korea

Unit 42 details a Bisonal malware variant, active since 2014, targeting Russian and South Korean defense organizations via PDF-disguised spearphishing emails.

In early May, Unit 42 discovered a campaign delivering a Bisonal malware variant against at least one Russian communications security and cryptography company and one unidentified organization in South Korea. The variant, in the wild since at least 2014, introduces a new C2 cipher and rewritten networking and persistence code, with only 14 samples collected to date. Attackers spoofed Russian state corporation Rostec in spearphishing emails carrying an executable disguised with a PDF icon; the dropper decrypts an RC4-encrypted DLL and establishes persistence via a registry Run key. Bisonal has been used since 2013 against government, military, and defense targets in South Korea, Russia, Japan, and India, alongside successors Bioazih and Dexbia.

Palo Alto Unit 42·Aug 17, 12:55 UTC · Aug 17, 2026Malware in the wild

Fortunate Recall: Ontology-Driven Memory Lifecycle Management for Persistent Coherence in LLMs

Fortunate Recall introduces ontology-based lifecycle policies for LLM memory, cutting confabulation roughly in half (e.g., 45.1% to 22.4%) versus Mem0.

Fortunate Recall (FR) is a composable policy layer that classifies personal facts into a 10+1 behavioral ontology and applies category-specific lifecycle rules including differential temporal decay, slot-key supersession, event-time validity, and retrieval routing. FR-Bank scores 76.9% on the new 516-question LifecycleBench, ahead of Mem0, A-MEM, Memory-R1, and MemoryOS (61%-70.5%), and 75.2% on LongMemEval-S. End-to-end, confabulation drops from Mem0's 45.1% to 22.4% over answered queries, with the ranking replicating on open-weight Kimi K2.5 and transferring to the independent BEAM benchmark (46.8% vs 32.9%).

arXiv cs.AI / cs.LG / cs.CL·Sep 9, 16:31 UTC · 7d agoAI research

GoldFactory Weaponizes Open-Source Vwork App Cloner in Gigabud Banking Malware Attacks

Group-IB says GoldFactory's Gigabud Android trojan uses Vwork, a weaponized Shelter fork, to clone banking apps into isolated Work Profiles and evade bank-side detection.

Group-IB's 'Hook for Gold' investigation found GoldFactory ships Vwork, a modified fork of the open-source Shelter app, alongside its Gigabud Android banking trojan, active since 2022. Vwork abuses Android Work Profile provisioning to clone banking apps into an isolated environment, weakening the link between detected malware signals and fraudulent transactions. Gigabud has targeted victims in Southeast Asia, Latin America, the Middle East, Africa, and beyond via fake airline, tax, and government apps requesting Accessibility and overlay permissions. In Indonesia, telemetry recorded about 1,469 compromised devices and roughly $960,939 in estimated losses between February and July 2026.

GBHackers·Sep 9, 10:49 UTC · 7d agoMalware in the wild

Hackers Use Fake GTA 6 Downloads to Deploy RATs, Infostealers and Data-Wiping Malware

Huntress tracked fake GTA 6 downloads distributing NJRAT, DCRAT, Mercurial Grabber, and Chaos ransomware used as a wiper against gamers.

Cybercriminals are pushing fake Grand Theft Auto VI downloads through poisoned search results, gaming forums, torrent sites, and social media, targeting players seeking a leaked build before release. Huntress analyzed a sample bundling NJRAT, DCRAT, Mercurial Grabber, and Chaos ransomware in one package; fake ISO files exceed 100GB largely with junk data, and Russian-language prompts suggest the operators target Russian-speaking gamers. Chaos ransomware acts as a wiper, overwriting files larger than 200MB with random data, deleting shadow copies for admin users, and disabling recovery options rather than offering genuine ransom payment paths.

Cyber Security News·Sep 10, 14:23 UTC · 6d agoMalware in the wild 3 sources

LexFlip: A Dissociation Diagnostic for Legal Meaning Preservation Metrics

LexFlip releases 373 minimal perturbations of Quebec statutory French that reverse legal force while preserving tokens, exposing weaknesses in embedding-based meaning preservation metrics.

LexFlip provides 373 minimal perturbations of Quebec statutory French that reverse legal force while preserving 0.93 of tokens, creating dissociation items that break monotone token-overlap metric validation. The seven embedding and BERTScore metrics tested register only 0.022-0.039 of their identical-to-unrelated range on these edits, versus 0.670 for bidirectional NLI. Against FrJudge, with a measured human ceiling of r=0.597, a bare length feature outscores every semantic metric tested.

arXiv cs.AI / cs.LG / cs.CL·Sep 4, 15:49 UTC · 12d agoAI research

How to opt out of AI chatbot training

Malwarebytes guides users through disabling AI training use of chats in ChatGPT, Perplexity, and Claude after OpenAI's human review program emerged.

404 Media reported that OpenAI's 'Project Lily' hires hundreds of contractors to review ChatGPT prompts, with a 'Privacy Filter' removing personal data and usernames hidden, though user memories summaries can still reveal identifying details. The article provides opt-out steps: ChatGPT Settings > Data Controls > 'Improve the model for everyone' (on by default), Perplexity Settings > Preferences > AI data retention, and Claude Settings > Privacy > 'Help Improve our AI Models'. Opting out does not prevent all human access, which remains allowed for abuse investigation, support, troubleshooting, and legal matters.

Malwarebytes Labs·Sep 15, 15:41 UTC · 1d agoAI industry

Flirty OnlyFans promoters on X may be using AI to appear human

Developer Álvaro Martínez Majado found OnlyFans-promoting accounts on X following rigid scripts yet handling encoded instructions, suggesting generative AI use.

Investigation of flirty X accounts promoting OnlyFans pages showed near-identical openers across accounts plus dynamic behaviors: answering a hexadecimal-encoded instruction with "Pineapple" and failing an exact 12-character count test in an LLM-like pattern. The accounts also sent personalized voice notes reading supplied timestamps and usernames, consistent with automated text-to-speech. Evidence suggests a hybrid scripted/AI system, though no model, provider, or operator was identified.

Malwarebytes Labs·Sep 7, 11:18 UTC · 9d agoAI safety & security

Gigabud Creates Android Work Profiles to Hide From Banking App Malware Checks

Group-IB reports the Gigabud Android banking trojan uses a cloned work profile to hide from banking app malware checks, with infections confirmed in Indonesia.

Group-IB says Gigabud installs a helper app called Vwork, derived from the open-source Shelter tool, which creates an Android work profile and drops a tampered banking app inside it, hiding the trojan from banking apps' malware scans. Gigabud, active since 2022 and linked by Group-IB to the GoldFactory group, abuses Accessibility access and overlay screens to steal credentials and run fraudulent payments while a black screen conceals the operator's actions. Group-IB confirmed the full attack chain on infected devices in Indonesia, counting about 1,469 compromised devices and estimated losses of roughly $960,000 between February and July 2026. Vwork-compatible Gigabud samples have been found targeting 11 countries including Brazil, Mexico, Indonesia, Thailand, and Türkiye, though only the Indonesian chain is confirmed.

The Hacker News·updated Sep 11, 12:14 UTC · 5d ago·first Sep 10, 11:33 UTC · 6d agoMalware in the wild 3 sources1

DeCAL: Towards Physically-Grounded Dexterous Vision-Language-Action Models via Contact-Aware Latent Co-Imagination

DeCAL, a contact-aware dexterous vision-language-action model with visuo-tactile fusion, reports 71% average task success.

DeCAL is a physically-grounded dexterous vision-language-action (VLA) model built on a Mixture-of-Transformers architecture with specialized experts for understanding, imagination, and action generation. It introduces Adaptive Visuo-Tactile Fusion with contact-aware gating and Visuo-Tactile Latent Co-Imagination to jointly model visual and tactile dynamics. It reports state-of-the-art results with a 71% average success rate and 83.4% progress success rate, plus generalization to unseen scenarios.

arXiv cs.AI / cs.LG / cs.CL·Sep 8, 17:47 UTC · 8d agoAI research1

How much of F-Droid is LLM generated?

A FOSS maintainer manually graded 102 F-Droid apps from the September 12, 2026 update batch, finding many show signs of LLM-generated code.

A student and FOSS app maintainer reviewed 102 apps pushed to F-Droid on September 12, 2026, assigning each a three-tier rating for likelihood of LLM-authored code (mostly AI >50%, hard to say/mostly human, no signs of AI). The heuristic relies on commit aesthetics, README and branding style, and the presence of agentic infrastructure like Claude Code or Codex, which automatically places an app in the 'mostly AI' tier. Example ratings include Amber (Nostr event signer) as mostly AI, and Aria for Misskey as showing no AI signs. The author stresses reliable detection of LLM-generated code from text alone is impossible, so ratings are approximate.

Hacker News · AI·Sep 15, 09:47 UTC · 1d agoAI researchHN 46↑ · 23 comments

Technical Manual for a Toolkit for Measuring Contextual Individuation in Transformer Language Models

An open methodology toolkit measures whether transformer language models contextualize fixed word forms across domains using bridge forms and layer-wise silhouette analysis.

The manual documents an open toolkit built around 'bridge forms' - identical written words recurring across two or more subject domains with a different sense in each - to test whether transformer language models individuate word occurrences by context beyond the embedding layer. It covers declarative specification of bridge forms, Wikipedia corpus acquisition, occurrence localization, layer-wise representation extraction, domain-pairwise silhouette measurement, and visualization, justifying each choice against failure modes such as sense contamination and subword-tokenization misalignment. It is a methodological and implementation reference and reports no empirical results.

arXiv cs.AI / cs.LG / cs.CL·Sep 4, 16:38 UTC · 12d agoAI research

Hackers Abuse YouTube Gaming Channels and SEO Poisoning to Deploy RATs and Chrome Hijacker

Unit 42 exposed CL-CRI-1171, a pay-per-install operation using YouTube gaming channels and SEO poisoning to deliver RATs and a Chrome hijacker.

Palo Alto Networks Unit 42 identified over 10,000 OfferLoader samples and 11 YouTube channels pushing trojanized installers via gaming lures and SEO-poisoned searches for tools like WinDirStat and Bluetooth drivers. Trojanized Inno Setup installers spawn three payload branches: Insomnia RAT (dual Node.js/Python backdoor that disables Microsoft Defender), ARKTunnel (undocumented RAT hiding its payload in a bitmap image and tunneling TCP/UDP), and Docro Hijacker (alters Chrome protected preferences to hijack search and affiliate links). The pay-per-install model sells each infection to multiple buyers, reaching gamers as well as corporate and government endpoints.

Cyber Security News·Sep 14, 08:01 UTC · 2d agoMalware in the wild1

Molecular Déjà Vu: Digit-Level Retrieval of Published Values in Frontier Language Models

Audit of 22 frontier models finds widespread verbatim retrieval of published molecular property values, with higher reasoning increasing recall of memorized numbers.

An arXiv audit tests 22 frontier LLMs across 12 molecular regression benchmarks for verbatim retrieval of published values. More than 50% of the LLMs show verbatim retrieval on five datasets, and identical experiments are flagged 89% more often at a high reasoning level than at the lowest one. Suppressing retrieval moves model prediction errors closer together in relative terms, suggesting predictive capability is not determined solely by memorized values.

arXiv cs.AI / cs.LG / cs.CL·Sep 4, 17:32 UTC · 12d agoAI research1

Introducing Meta One: A Subscription Service With More Features and AI to Create, Connect, and Stand Out

Meta launches global Meta One subscriptions bundling Instagram, Facebook, WhatsApp Plus with expanded Meta AI and Muse media generation, priced $2.99-$499 monthly.

Meta introduced Meta One, a global subscription service with plans for individuals, creators, and businesses, launching with more than 50 features across Instagram, Facebook, WhatsApp, and Meta AI. Individual bundles Core ($7.99/month) and Premium ($19.99/month) combine the single-product Plus plans with expanded use of compute-intensive AI capabilities, including image/video generation powered by Muse models and Instagram's Restyle. Business tiers range from Essential ($14.99/month) to Expert ($149/month) and Max ($499/month), with expansion planned to Edits, AI glasses, and more.

Meta Newsroom·Sep 15, 15:00 UTC · 1d agoAI industry

You Can't Prefer Emotions You Don't Sample: Intensity Undershoot in DPO-Tuned LLMs

Study quantifies DPO-tuned LLMs undershooting requested emotional intensity, tracing the gap to candidate-pool extremity rather than conditioning format.

Conditioning an instruction-tuned LLM on continuous valence-arousal targets yields gain of only 0.26 for valence and 0.13 for arousal on Llama-3.1-8B, far below faithful control of 1.0. The authors attribute undershoot to neutral-heavy preference corpora like EmoBank and candidate pools lacking extreme affect, leaving DPO without extreme exemplars. Uniform target coverage with a hotter candidate pool raises valence gain to 0.40 on Llama-3.1-8B and 0.44 on Qwen3-8B, with modest in-distribution cost; arousal gains remain unstable across seeds.

arXiv cs.AI / cs.LG / cs.CL·Sep 7, 17:47 UTC · 9d agoAI research

Claude's new system prompt really doesn't want to reproduce song lyrics

Anthropic published updated Claude consumer system prompts, including changes steering the model away from reproducing song lyrics, likely over copyright concerns.

Anthropic publishes system prompts for Claude.ai and Claude mobile apps, including historic revisions, and has reorganized them into an index with per-model pages such as the Haiku 4.5 page showing the original October 15, 2025 prompt and an updated January 18, 2026 version. The latest consumer prompt strongly discourages reproducing song lyrics, a behavioral constraint likely tied to copyright considerations. Prompts for Claude Cowork and Claude Code are not included in the published set.

Simon Willison·Sep 2, 14:16 UTC · 14d agoAI safety & security

Tracking Elirks Variants in Japan: Similarities to Previous Attacks

Unit 42 links new Elirks backdoor variants attacking Japanese organizations to 2012 Taiwan attacks, delivered via spear-phishing PDFs exploiting Adobe Flash CVE-2011-0611.

Unit 42 analyzed new Elirks backdoor variants found in an attack on a Japanese business, noting strong similarities to 2012 attacks on Taiwanese ministries. The backdoor retrieves its C2 address from attacker-created accounts on Japanese blog and SNS services. Recent deliveries used an airline e-ticket lure named "E-TKT" with a PDF exploiting Adobe Flash CVE-2011-0611. Shared infrastructure and tactics with the Scarlet Mimic campaign suggest possible ongoing cyber espionage across East Asia.

Palo Alto Unit 42·Aug 17, 12:56 UTC · Aug 17, 2026Threat actor in the wildCVE-2011-0611

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

Prompt Injections for Defense

Tracebit researchers show prompt injections placed next to AWS secrets can stop AI hacking agents by triggering forbidden outputs, a technique called context bombing.

Researchers from Tracebit reported that placing prompt injections alongside passwords, SSH keys, and other secrets stored on AWS could shut down attacks by AI hacking agents. The injected prompts order the attacking LLM to perform actions forbidden by its guardrails, such as explaining how to develop inhalable Anthrax spores or referencing Tank Man, causing guarded models to halt. The researchers named the technique context bombing and note it only works against agents with guardrails, not locally run guardrail-free models.

Schneier on Security·Aug 12, 09:57 UTC · Aug 12, 2026AI safety & security

Researchers Show How Meta's 'Pervert Glasses' Are Used to Harass Women

University of Sydney researchers detail how pickup artists use Meta Ray-Ban smart glasses to covertly film and harass women, then post the videos on Instagram.

Researchers Joanne Gray, Milica Stilinovic, Marcus Carter, and Ben Egliston analyzed 350 Instagram videos posted between September 2023 and March 2026 showing unsolicited approaches to women filmed with smart glasses. They found a clear correlation between covert filming and harassment severity, arguing ambient capture creates 'borderline' harassment that evades platform moderation mechanisms. Instagram head Adam Mosseri said the platform would remove harassing pickup-line content, though similar videos remain widespread a month later. Meta's safeguards, such as the recording light, were previously criticized as insufficient, and users have modded glasses to disable the light.

404 Media·Aug 12, 14:05 UTC · Aug 12, 2026AI safety & security

HVNC Backdoor Targets LATAM Organizations with Fake Tax and DocuSign Lures

A custom HVNC backdoor is targeting Latin American financial organizations via fake DocuSign and NFe tax-document lures, giving attackers hidden persistent remote access.

ANY.RUN researchers analyzed a multi-stage phishing campaign delivering a custom HVNC backdoor to banking and financial organizations in Latin America. The chain starts with fake DocuSign and NFe tax-document pages that serve per-visitor ZIP archives, followed by an LNK dropper, an NSIS loader, and a 64-bit backdoor masquerading as Windows Update Assistant. The implant provides hidden remote desktop control, keystroke monitoring, Firefox data theft, Startup-folder persistence, and EDR-aware behavior, communicating over TCP/27015.

ANY.RUN·Sep 8, 10:24 UTC · 8d agoMalware in the wild

Why metaphor may dictate your security strategy

Cisco Talos·Aug 6, 18:00 UTC · Aug 6, 2026Vulnerability1

APT36 Suspected in PATCHCORD Espionage Campaign Using Google Sheets C2

Acronis links the PATCHCORD espionage campaign targeting Afghan telecom and South Asian critical infrastructure to APT36 with moderate confidence.

Acronis Threat Research Unit documented PATCHCORD, a previously undocumented C/C++ backdoor delivered via fake VPN installers impersonating Afghan Telecom (AFTEL), which hijacks Edge, Chrome and Firefox shortcuts for persistence and executes in-memory shellcode. A related Go implant, SHEETCORD, abuses the Google Sheets API for per-victim C2 and was distributed via a domain impersonating India's National Informatics Centre, while a third tool, HACKERAI C2 Agent, uses GitHub Gists and shows signs of LLM-assisted coding. An exposed staging server revealed SuperShell, RAT frameworks, credential-harvesting tools and OpenSSH exploit code, and the researchers attribute the activity to APT36/Transparent Tribe at moderate confidence.

Security Affairs·Aug 16, 07:24 UTC · Aug 16, 2026Threat actor in the wild

LLMs and Contextual Integrity

Bruce Schneier highlights two papers: the CIMemories benchmark shows frontier LLMs leak memory attributes up to 69%, and an RL method reduces inappropriate disclosures.

Bruce Schneier discusses contextual integrity in LLMs, referencing the CIMemories benchmark, which uses synthetic profiles with 100+ attributes per user to test whether models with persistent memory disclose sensitive information appropriately. Evaluation showed frontier models exhibit up to 69% attribute-level violations, with GPT-5's violation rate rising from 0.1% to 9.6% across 40 tasks and reaching 25.1% with repeated prompting, showing unstable leakage behavior. A second paper introduces a reinforcement learning framework trained on a synthetic 700-example dataset that substantially reduces inappropriate disclosure while maintaining task performance, with improvements transferring to the human-annotated PrivacyLens benchmark.

Schneier on Security·Aug 18, 10:40 UTC · 29d agoAI safety & security

Microsoft says Copilot buttons still missing in classic Outlook

Microsoft is still investigating a bug that makes Copilot buttons disappear in classic Outlook for affected M365 Copilot users.

Microsoft confirmed the missing Copilot and Copilot Chat buttons occur after upgrading classic Outlook for Windows to build 20026.20182 and higher, because Outlook cannot locate the MAPI property PR_PROFILE_USER_SMTP_EMAIL_ADDRESS_W. The issue affects Copilot Chat (Basic) and paid M365 Copilot (Premium) customers, and Copilot remains reachable via OWA, new Outlook, and the standalone app. A temporary workaround enables 'Show Apps in Outlook' under Advanced settings, and Microsoft also acknowledged Outlook crashes tied to Kaspersky's Mail Checker (mcou.dll).

BleepingComputer·12:16 UTC · 6h agoAI industry 2 sources