Search: “NoName057”

33 stories

DomainAll Cyber AIWindow24h 3d 7d 30d AllIn the wild Important only Show duplicates

CategoryAll Threat actor 35 Ransomware 27 Malware 24 Exploit / PoC 20 Policy & legal 10 Data breach 10 Vulnerability 7 Industry 5 Phishing & fraud 3 Research 2 Advisory 1

SeverityAll critical 83 high 20 medium 13 low 27 info 1

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

Risky Bulletin: Expired cards can be used for new transactions

Researchers show expired Visa contactless cards can be revived via NFC man-in-the-middle relay to run fraudulent transactions; roundup also covers major breaches.

University of Massachusetts Amherst researchers built an NFC man-in-the-middle rig that updates a card's expiration date in transit and relays the modified payment to POS terminals, reviving expired contactless cards; Visa terminals and the backends of all five banks studied failed to catch the manipulation. The same roundup reports Iranian hackers shut down a small UK power plant for four days, Lazarus breached South Korea's Presidential Office as part of a campaign exceeding 100 victims, and French telecom SFR suffered a breach affecting over 2.1 million customers.

Risky Business NewsAug 24, 04:40 UTC · 23d agoResearch1

CVE-2026-73579: Apache Syncope: Non-recursive Any search could skip Realms restrictions

Apache Syncope non-recursive Any search can skip Realms restrictions, exposing objects outside an administrator's authorized realm (CVE-2026-73579).

CVE-2026-73579 is an incorrect authorization vulnerability in Apache Syncope where non-recursive Any search requests are transformed in a way that skips Realms restrictions, returning objects outside the administrator's authorized realm. Affected component is syncope-core-persistence-common 3.0.0-M0 through 3.0.16, 4.0.0-M0 through 4.0.7, and 4.1.0-M0 through 4.1.2. Apache rates the issue moderate severity.

oss-securitySep 14, 15:31 UTC · 2d agoVulnerability CVE-2026-73579

Risky Bulletin: Ukraine's top prosecutor resigns amid scam call center scandal

Ukraine's top prosecutor Ruslan Kravchenko resigned after NABU arrested a deputy for taking bribes protecting scam call centers running fake investment platforms.

Ukraine's anti-corruption bureau NABU arrested Serhiy Kropyva, Deputy Head of International Cooperation at the Prosecutor General's Office, alleging officials took monthly protection fees from a network of 100-500 scam call centers luring victims into fake investment platforms, with bribes reportedly growing from $700,000 to $3.5 million per month. Prosecutor General Ruslan Kravchenko resigned on Monday, calling it a political decision, while Kropyva was fired with bail set at 120 million hryvnias ($2.7 million) and over 100 call centers shut down in the past month. The newsletter also briefly covers a cyberattack crippling more than 80 Luxembourg medical practices via payment vendor BMS Engineering, ShinyHunters' claimed theft of 200,000 Florida DMV driver records, a cyberattack on the American Meteor Society, and school closures in Springfield, Massachusetts.

Risky Business NewsSep 9, 06:06 UTC · 7d agoPhishing & fraud

CVE-2026-86219: Authen::SASL::Perl::DIGEST_MD5 versions before 2.2100 for Perl accept replayed authentication responses via unverified nonce in server_step

Authen::SASL::Perl::DIGEST_MD5 before 2.2100 for Perl accepts replayed DIGEST-MD5 authentication responses via unverified nonce handling (CVE-2026-86219).

CVE-2026-86219 affects Authen::SASL::Perl::DIGEST_MD5 versions before 2.2100 for Perl. The server_step function does not verify the nonce, allowing replayed authentication responses to be accepted in DIGEST-MD5 SASL exchanges. The fix is available in version 2.2100 of the perl-authen-sasl distribution.

oss-securitySep 6, 18:55 UTC · 9d agoVulnerability CVE-2026-862191

Risky Bulletin: Russia starts blocking DoH and DoT

Russian users report blocks on DoH and DoT servers, including Cloudflare 1.1.1.1 and Google 8.8.8.8, in an apparent censorship crackdown.

Russian internet users began reporting failures connecting to DNS-over-HTTPS and DNS-over-TLS servers, suggesting a government crackdown on the two privacy protocols. The blocks reportedly cover Cloudflare's 1.1.1.1 and Google's 8.8.8.8 resolvers; Roskomnadzor has not officially confirmed the action. The agency tested a similar block in March on Beeline's network and had named DoH for blocking as early as 2021. The bulletin also briefly notes state-sponsored phishing of EU officials, a DDoS against Norway's Digdir, the ReliaQuest/ShinyHunters dispute, and older ransomware and breach disclosures.

Risky Business NewsAug 26, 03:07 UTC · 21d agoPolicy & legal1

The 2026 PNPL Competition: Word Classification and Efficient Cross-Subject Generalisation in LibriBrain100

The 2026 PNPL competition releases LibriBrain100, a MEG speech dataset with 32 extra subjects, targeting word classification and cross-subject BCI generalization.

The 2025 PNPL competition on non-invasive speech decoding from MEG achieved F1-macro scores of 95.6% for speech detection and 73.6% for phoneme classification, built on LibriBrain's ~50 hours of single-subject data. The 2026 edition extends this with LibriBrain100, adding 32 subjects (~40 minutes each) plus ~80 hours of within-subject data. Two tracks target within-subject word classification at scale and cross-subject generalization with subject-specific fine-tuning shrinking from ~40 to ~20 to ~10 minutes, aiming at clinically feasible non-invasive BCIs for people with profound paralysis.

Hugging Face daily papersSep 2, 20:00 UTC · 13d agoAI research

Untracked Nightmares: The Threats Hiding Behind Commodity Infrastructure

Unit 42 exposes CL-CRI-1171, a pay-per-install network spreading malware like Insomnia RAT via YouTube channels and SEO poisoning for over two years.

Palo Alto Networks Unit 42 details CL-CRI-1171, a cybercrime cluster operating a pay-per-install (PPI) marketplace that has delivered multiple malware families for at least two years. The group used at least eleven YouTube gaming channels with hundreds of thousands of followers, plus SEO poisoning promoting trojanized software such as a Bluetooth driver and WinDirStat, infecting gamers and corporate endpoints including critical infrastructure and government entities. A single shared loader delivered payloads including Insomnia RAT, ARKTunnel, Docro Hijacker, GCleaner and Socks5Systemz between July 2025 and April 2026, with more than 10,000 distinct loader samples and over 200 rotating C2 domains identified. YouTube terminated the malicious channels after Unit 42 notified the platform.

Palo Alto Unit 42Sep 9, 10:00 UTC · 7d agoMalware in the wild1

Ncsc Raises Alarms Prompt

The UK NCSC raised alarms about prompt injection risks in LLM-integrated systems, urging organizations deploying AI to review exposure.

The UK National Cyber Security Centre (NCSC) has raised alarms about prompt injection attacks against systems using large language models. The warning highlights how attackers can manipulate model instructions to bypass safeguards, exfiltrate data, or trigger unintended agent actions. Organizations deploying LLM-based features are advised to assess and mitigate their exposure to this technique.

Infosecurity MagazineAug 19, 00:00 UTC · 28d agoAI safety & security

Safety for Whom? Refusing the Right Subset of a Topic, Not the Whole Topic

Multiverse Computing's Hugging Face post argues language models should refuse only the relevant subset of a topic instead of over-refusing whole subjects.

A Hugging Face blog post by Multiverse Computing examines refusal granularity in language models, arguing models should refuse the relevant subset of a topic rather than the entire topic. No full article text was available for additional technical detail.

Hugging Face BlogSep 8, 14:23 UTC · 8d agoAI safety & security

Evaluating Practical Enumeration and Blocking Attacks on the Snowflake Circumvention System

Ethical measurements enumerated 21,000+ Snowflake proxy IPs across ~1,000 ASes; blocking top 1% of ASes disrupts 30% of Snowflakes.

Researchers tested Snowflake's assumptions that proxy IPs cannot be easily enumerated and that blocking them causes unacceptable collateral damage. Over 48 days of real-world measurements (May-June 2025), malicious-client-style enumeration collected over 21,000 unique proxy IPs across almost 1,000 autonomous systems. Blocking the top 1% of ASes blocks more than 30% of observed Snowflakes while affecting about 2.5% of Tranco Top 1M domains, and the broker's load-aware matching leaks stable high-capacity proxies to attackers. Some proposed mitigations have already been integrated into Snowflake.

arXiv cs.CRSep 10, 22:01 UTC · 5d agoResearch1

llm 0.35

llm CLI tool version 0.35 adds support for OpenAI's new GPT-6 Astra model exposed as gpt-6-astra.

Simon Willison released llm 0.35, which adds an OpenAI model definition for GPT-6 Astra under the model ID gpt-6-astra. No other release details were provided in the post.

Simon WillisonSep 7, 23:54 UTC · 8d agoAI tools & infra

Palo Alto Networks product-category page describing Next-Generation Firewall threat detection, including encrypted traffic; no standalone article.

The page is a product category listing for Palo Alto Networks Next-Generation Firewalls, stating the products detect known and unknown threats, including in encrypted traffic, using intelligence from many thousands of customer deployments. It contains no research findings, vulnerabilities, or incident details relevant to defenders.

Palo Alto Unit 42Aug 19, 12:05 UTC · 28d agoIndustry

ToxicPanda 2.0 Gets a Major Upgrade, Expanding Attacks Across 16 Countries

Zimperium documents ToxicPanda 2.0, an Android banking trojan now targeting 349 financial institutions in 16 countries via ADB privilege escalation and overlay credential theft.

Zimperium's zLabs documented ToxicPanda 2.0, an Android banking trojan expanding from 16 targeted apps to 349 financial institutions across 16 countries, with 167 remote commands. It poses as a dropper, abuses VPN permissions to block Google Play Protect while installing a hidden payload, then uses the Accessibility Service for screen monitoring and overlay-based credential theft. It automates enabling Android Wireless Debugging and completes the pairing handshake to gain ADB shell access for privilege escalation, and overlays fake lock screens to steal device PINs. Previously unfinished commands are now operational and samples are served from AWS-hosted storage buckets.

Security AffairsAug 22, 16:50 UTC · 25d agoMalware in the wild

Recognition-Refusal Misalignment in LLMs: Why Models Answer Structurally Unanswerable Questions

A linear hidden-state direction encodes question impossibility in 1.7B-70B LLMs, but misalignment with the safety-refusal pathway explains why models answer unanswerable questions.

The study examines why instruction-tuned LLMs from 1.7B to 70B parameters answer structurally unanswerable math and code questions instead of abstaining. A single linear direction in the hidden state separates answerable from impossible prompts, showing models represent impossibility before generation, but this direction is nearly orthogonal to the canonical safety-refusal direction. Generation-time steering along the recognition direction changes invalidity-aware behavior dose-responsively, and the geometry is present even at the pretraining endpoint, indicating a routing failure rather than an encoding failure.

Hugging Face daily papersAug 28, 20:00 UTC · 18d agoAI safety & security

Cortex Xpanse

Palo Alto Unit 42 Cortex Xpanse product-category page published with no article text available for classification.

The URL resolves to the Cortex Xpanse product category listing on the Unit 42 blog rather than a research article. No article body was provided, so no vulnerabilities, incidents, or research findings can be extracted.

Palo Alto Unit 42Aug 19, 12:04 UTC · 28d agoIndustry

FrostyGoop’s Zoom-In: A Closer Look into the Malware Artifacts, Behaviors and Network Communications

Unit 42 dissects FrostyGoop OT malware, which sent Modbus TCP commands to disrupt heating for 600+ apartment buildings in Ukraine.

Palo Alto Unit 42 analyzed new FrostyGoop/BUSTLEBERM samples, the ninth known ICS-centric malware, first used against a Ukrainian municipal energy company by sending Modbus TCP commands to ENCO control devices. The attack caused a two-day heating outage affecting over 600 apartment buildings during sub-zero temperatures. The Go-compiled malware accepts Modbus register operations and timing parameters via command-line arguments or a JSON configuration file, uses an obscure open-source Modbus library, and checks the Windows BeingDebugged flag to evade debuggers. Unit 42 telemetry recorded over 1 million Modbus TCP devices exposed to the internet between September and October 2024.

Palo Alto Unit 42Aug 19, 12:04 UTC · 28d agoMalware in the wild

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.CRSep 11, 08:34 UTC · 5d agoResearch1

Researcher Publishes CrowdStrike Privilege Escalation Zero Day

Researcher NightmareEclipse published FalconFlank, a zero-day local privilege escalation in CrowdStrike Falcon Sensor abusing its Office malicious macro remediation feature.

A researcher known as Nightmare Eclipse published details of FalconFlank, a zero-day privilege escalation that abuses the Microsoft Office malicious macro removal feature in CrowdStrike Falcon Sensor, verified working on fully updated Windows 11 25H2 and Windows Server 2025 with Falcon Phase 3 Optimal Protection. CrowdStrike urged customers to disable the Office File Suspicious Macro Removal policy while it investigates, noting customers remain protected via Cloud Anti-malware for Microsoft Office Files; no CVE has been assigned. The same researcher previously released the Exploitarium dump of over 30 PoC exploits, and researcher Kevin Beaumont confirmed FalconFlank works.

Infosecurity MagazineSep 7, 08:28 UTC · 9d agoExploit / PoC

Knowing What Not to Answer: Selective Non-Compliance in Vision-Language Models

Researchers introduce KoNA, a benchmark exposing vision-language models' failures at selective non-compliance, plus fine-tuning that improves refusal and abstention accuracy.

KoNA is a benchmark for evaluating selective non-compliance in vision-language models across five categories: False Premise, Visual Inaccessibility, Universal Unknown, Task Feasibility and Safety. It tests both query-level and component-level non-compliance using paired single and compound queries, and evaluations across diverse VLMs show models often fail to refuse, correct or abstain appropriately, with failures worsening on compound queries. Fine-tuning VLMs on KoNA examples substantially improves non-compliance accuracy while largely maintaining performance on fully answerable tasks.

Hugging Face daily papersSep 3, 20:00 UTC · 12d agoAI research1

Cardinal RAT Sins Again, Targets Israeli Fin

Unit 42 documents updated Cardinal RAT attacks against Israeli FinTech firms, using BMP steganography, MD5-hash obfuscation, and process injection to hinder analysis and detection.

Unit 42 tracked a series of attacks using an updated Cardinal RAT (version 1.7.2) targeting the Israeli financial technology sector. The .NET loader hides a second-stage DLL inside an embedded BMP image decrypted with a single-byte XOR key, and the payload renames functions, methods, and variables to MD5 hashes for obfuscation. The malware installs a startup-folder LNK file and injects its final payload into RegSvcs.exe or RegAsm.exe, communicating with affiliatecollective[.]club over port 443. A possible relationship with the EVILNUM JavaScript malware used against similar organizations was also noted.

Palo Alto Unit 42Aug 17, 12:21 UTC · Aug 17, 2026Malware1

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-securityupdated Sep 13, 15:08 UTC · 3d agofirst Sep 13, 14:51 UTC · 3d agoIndustry 12 sources

Updated ToxicPanda Variant Targets 140+ Banking and Crypto Apps

Zimperium reports ToxicPanda 2.0, an updated Android banking trojan now targeting more than 140 banking and cryptocurrency apps.

Zimperium has published analysis of ToxicPanda 2.0, an updated Android banking trojan variant. The new variant expands its target list to over 140 banking and cryptocurrency applications. The report lifts the lid on the updated capabilities of the mobile malware family.

Infosecurity MagazineAug 20, 10:00 UTC · 27d agoMalware

Cortex XDR™ Detects New Phishing Campaign Installing NetSupport Manager RAT

Cortex XDR threat hunters uncovered a phishing campaign delivering the NetSupport Manager RAT via a fake password-protected NortonLifelock Word document.

Unit 42 identified a January 2020 phishing campaign using a Microsoft Word document disguised as a password-protected NortonLifelock file. Enabling macros triggered an obfuscated command that built alpaca.bat in the temp directory, which used msiexec to download an MSI payload from quickwaysignstx.com, filtered on the Windows Installer user-agent string. The payload installed a PowerShell script and the campaign, which has delivered NetSupport Manager RAT since at least 2018, showed related activity dating back to early November 2019.

Palo Alto Unit 42Aug 17, 13:18 UTC · Aug 17, 2026Phishing & fraud in the wild1

GuardBreaker: Derailing AI-assisted malware analysis with a code comment

ESET names 'GuardBreaker': UAC-0099 embeds a nuclear-weapon question in VBScript comments to trip LLM scanner guardrails during analysis of its MATCHBOIL loader.

ESET researchers observed the Russia-aligned group UAC-0099 inserting a decoy prompt injection into a VBScript used to install its MATCHBOIL loader in an attack against a Ukrainian target, aiming to make LLM-based code scanners refuse and stop inspecting the file. The comment triggers safety guardrails with a request about building a nuclear weapons but has no runtime effect. Similar LLM-thwarting tricks have appeared in malicious PyPI and npm packages reported by Socket and StepSecurity. ESET recommends multi-model cross-validation of AI-assisted analysis and treating missing LLM output as requiring further checks.

ESET WeLiveSecurityupdated Sep 11, 14:32 UTC · 5d agofirst Sep 10, 09:00 UTC · 6d agoAI safety & security 3 sources1

Decoy Direction Optimization: A Post-Hoc Defense Against LLM Abliteration

Researchers introduce Decoy Direction Optimization, a cheap weight-editing defense that blinds refusal-direction ablation attacks against open-weight LLM safety guardrails.

Refusal Feature Ablation bypasses safety guardrails in open-weight LLMs by projecting out a linear refusal direction, often with high attack success rates. Decoy Direction Optimization injects a high-magnitude nonlinear decoy into MLP neurons so attackers' contrastive estimators ablate a harmless orthogonal feature instead. Evaluated across six model families, DDO keeps ASR below 10% under standard RFA and on Llama-3-8B-Instruct reduces Heretic weight-level attack ASR from 88.7% to 18%. It costs 30 to 450 times less per configuration than trained defense baselines.

Hugging Face daily papersSep 13, 20:00 UTC · 2d agoAI safety & security

Safety for Whom? Boundary-Aware Self-Distillation for Controlled LLM Safety Refusal

A self-distillation safety framework tunes narrow-boundary refusals in Qwen3-8B, raising target-domain refusal to 84.75% while cutting over-refusal from 15.20% to 5.20%.

The paper formulates narrow-boundary safety, where deployments need refusals within specific topics rather than whole subjects, and proposes an offline self-generated framework with controlled topic generation, escalating retries, and harmful-benign boundary pairs. On political persuasion with Qwen3-8B, the method raised target-domain refusal from 9.47% to 84.75% and cut the mean unsafe-response rate across three broader benchmarks from 26.26% to 0.14%. Verified target-model responses reduced over-refusal from 15.20% to 5.20%, and boundary-pair data cut comply-side over-refusal on held-out pairs from 32.94% to 4.16%. Results show data composition controls the safety-usability trade-off and alignment should be evaluated on both sides of the refusal boundary.

Hugging Face daily papersSep 2, 20:00 UTC · 13d agoAI safety & security1

CVE-2026-34486: Apache Tomcat EncryptInterceptor Fail-Open Bypass

Technical analysis shows CVE-2026-34486 is a one-line fail-open regression in Tomcat Tribes enabling unauthenticated RCE via Java deserialization.

CVE-2026-34486 is a fail-open regression in Tomcat's Tribes EncryptInterceptor, introduced while fixing the CVE-2026-29146 padding-oracle flaw; Apache rated it Important and NVD scores it 7.5 High. Decryption failures are now forwarded to an unfiltered ObjectInputStream, and public PoC repositories demonstrate unauthenticated RCE on default port 4000 when clustering, the EncryptInterceptor, a reachable receiver and deserialization gadgets are all present. Affected releases 9.0.116, 10.1.53 and 11.0.20 are fixed in 9.0.117, 10.1.54 and 11.0.21; Tomcat 8.5 is unaffected.

Web discovery (articles for new exploits & KEV entries)Sep 8, 20:09 UTC · 7d agoExploit / PoC CVE-2026-34486 CVE-2026-291461

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 · AISep 7, 09:26 UTC · 9d agoAI tools & infra HN 141↑ · 52 comments1

How Fragile Is Safety Alignment at Frontier Scale? A Single-Direction Attack on a 320B MoE

Researchers show directional ablation breaks refusal in GLM-5.3-Flash, a 320B-parameter MoE, cutting refusal by 41–89 points across seven benchmarks.

The study extends directional ablation, a white-box attack that removes an aligned LLM's refusal behavior, from dense models up to ~70B parameters to GLM-5.3-Flash, a 320B-parameter mixture-of-experts model with 288 routed experts, four-wide hyper-connection residual, and block-FP8 quantization. Editing attention, dense, and routed-expert writers jointly removes 0.776 of refusal, with 74% of the effect existing only under the joint intervention; the conventional module-name-based recipe reaches only 0.066 and fails silently on MoE architectures. The attack yields 41–89 percentage-point reductions in refusal across seven harmful benchmarks with no detected capability change, and a category-concentrated refusal residue survives all edits at ranks 1 to 12.

arXiv cs.CRSep 9, 06:48 UTC · 7d agoAI safety & security

PuzzleMask: Abusing Plain Prose as a Covert AI Attack Vector

Check Point details PuzzleMask, a plain-prose technique that bypasses LLM gatekeeper policy checks, letting hidden payloads reach target models unreviewed.

Check Point Research describes PuzzleMask, a prompt-crafting technique that hides policy-violating payloads inside plain-English prose wrappers, bypassing quick LLM-based policy checks without emojis, Base64, or invisible formatting. The researchers tested 23 automated prompts against gatekeepers including GPT-4o-mini, GPT-OSS-Safeguard 20b, Claude 3 Haiku, and Llama Guard 3, and all were classified as safe despite policies that flagged the plain versions. When submitted to GPT-5 in thinking-high mode with a Python interpreter, the target model extracted and acted on the payload in over 90% of trials. The technique is not itself a jailbreak but can carry a jailbreak prompt as payload; mitigations include input paraphrasing, hardened gatekeeper policies, and output monitoring.

Check Point Researchupdated Sep 11, 05:53 UTC · 5d agofirst Sep 10, 14:32 UTC · 6d agoAI safety & security 2 sources

How Lossless Is Lossless Speculative Decoding? The Role of Numerical Precision in Orthrus

Reproduction study finds Orthrus speculative-decoding trajectories match the reference model in only ~45% of cases under BF16, but 100% under FP32.

Researchers independently reproduced Orthrus, a hybrid autoregressive-diffusion architecture claiming lossless speculative decoding via intra-model consensus, testing exact trajectory matching on 1,190 prompts across 12 domains. Under BF16, exact matching occurred in only 45% of cases for the authors' checkpoint and 43% for an independently trained model, with matching probability strongly tied to reference-model response-conditional perplexity. Despite trajectory divergence, downstream lm-eval-harness benchmarks showed no systematic degradation, while FP32 evaluation yielded exact matching on all prompts.

Hugging Face daily papersSep 13, 20:00 UTC · 2d agoAI research1

Perturbation Probing: A New Diagnostic for the Fragility of LLM Safety

Unit 42 research shows LLM safety refusals concentrate in a thin neural layer, motivating external, multi-layered AI security controls.

Palo Alto Networks Unit 42 introduces Perturbation Probing, a diagnostic technique for measuring the fragility of LLM safety mechanisms. The research finds that safety refusal behavior is localized within a thin neural layer, implying small perturbations can undermine built-in refusals. The authors argue this motivates external, multi-layered security defenses on top of model-internal safety training.

Palo Alto Unit 42Aug 28, 22:00 UTC · 18d agoAI safety & security

ImpossibleRubrics: Stress-Testing Generated Rubrics as Reward Signals

ImpossibleRubrics benchmark shows LLM-generated rubric reward signals are exploited 8-26% of the time by adversarial answers on impossible tasks.

ImpossibleRubrics is a benchmark of 169 impossible tasks across six impossibility categories, each paired with a verifiable oracle certificate, plus 48 answerable controls, for stress-testing LLM-generated rubrics used as reward signals. Eleven rubric generators were exploited 8-26% of the time on an unbiased 150-task cut and up to 36% on a stress cut, while a certificate-faithful rubric scored 0%. A single generic 'be decisive, penalize hedging' rubric was exploited 64% of the time, suggesting tailored criteria can reveal which claims attackers should fabricate.

Hugging Face daily papersSep 14, 20:00 UTC · 1d agoAI research