ZeroHour
DashboardNewsCVEsExploitsIncidentsAIBriefAnalyticsEntitiesWatchlistIOCsAsk the feedSourcesFeeds & APIKiosk mode

Search: “fmc”

33 stories

Searchkeywordsmart
DomainAllCyberAIWindow24h3d7d30dAllIn the wildImportant onlyShow duplicates
CategoryAllExploit / PoC 24Vulnerability 23Ransomware 12Industry 4Malware 3Advisory 3Tools 1Data breach 1
SeverityAllcritical 48high 14medium 3low 1info 5

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

Cisco FMC flaws exploited by ransomware gang, state-sponsored hackers

Cisco Talos confirms ransomware and state-sponsored groups exploited CVE-2026-20079 and CVE-2026-20316 in Secure Firewall Management Center, deploying Qilin ransomware and Cyclops Blink.

Cisco Talos identified three threat clusters (UAT-12197, UAT-11823, UAT-11988) exploiting CVE-2026-20079, a CVSS 10.0 authentication bypass, and CVE-2026-20316, a CVSS 5.3 static-credential flaw, in Cisco Secure Firewall Management Center. Qilin affiliates used static credentials for reconnaissance and deployed Qilin ransomware; a Sandworm-linked cluster deployed a Cyclops Blink backdoor via a malicious license.tmp file; a third cluster stole credentials through a JSP web shell. Cisco has released hotfixes and urges immediate installation, with broader patches expected next week.

BleepingComputerupdated · 2d agofirst · 5d agoExploit / PoC in the wild 13 sourcesCVE-2026-20079CVE-2026-20316

Week in review: Linux rootkit deployed on F5 BIG-IP APM devices, Cisco FMC bugs exploited

Weekly roundup: Cisco FMC and N-able N-central zero-days exploited in the wild, MikroTik RouterOS hijacks, Microsoft Patch Tuesday ships two exploited zero-days.

State-sponsored and financially-motivated attackers are actively exploiting CVE-2026-20079, a critical authentication bypass in Cisco Secure Firewall Management Center (FMC), alongside CVE-2026-20316. N-able issued an emergency hotfix for CVE-2026-86218, a critical pre-auth RCE in the N-central RMM platform exploited in the wild. CERT Polska disclosed six RouterOS vulnerabilities being chained to hijack internet-exposed MikroTik devices. Microsoft's September 2026 Patch Tuesday shipped a record patch count including two zero-days, while roughly 67,000 Trezor customers faced phishing after a shipping-partner breach and researchers privately disclosed a zero-click WeChat worm to Tencent.

Help Net Security · 3d agoExploit / PoC in the wildCVE-2026-20079CVE-2026-20316CVE-2026-862182· 1 read

Cisco Secure Firewall Management Center Software Static Credential Vulnerability

Cisco warns of static low-privileged credentials in Secure Firewall Management Center's web interface, letting unauthenticated remote attackers log in and access sensitive data.

Cisco disclosed a vulnerability in the web interface of Secure Firewall Management Center (FMC) Software caused by the presence of static credentials for a low-privileged account. An unauthenticated remote attacker could log in to an affected device using the static account and access sensitive data within impacted systems. The attack surface is reduced when the FMC management interface does not have public internet access.

Cisco Security Advisories · Aug 11, 2026Advisory

FTC Withdraws Obsolete Policy Statement

The FTC rescinded its 2021 policy statement that applied the Health Breach Notification Rule to health apps and connected devices collecting consumer health data.

The Federal Trade Commission formally rescinded its 2021 Policy Statement on Breaches by Health Apps and Other Connected Devices. The statement had purported to apply the FTC's Health Breach Notification Rule to health apps and connected devices that collect consumer health information. The Commission considers the statement obsolete following its 2024 update to the Health Breach Notification Rule.

DataBreaches.net · 5d agoPolicy & legal 2 sources

Security Affairs newsletter Round 594 by Pierluigi Paganini – INTERNATIONAL EDITION

Weekly Security Affairs newsletter aggregates top stories including Cisco FMC exploitation, Qilin ransomware, Chrome zero-days, and Berlin leak.

Pierluigi Paganini's Security Affairs newsletter Round 594 (International Edition) rounds up the week's security headlines. Topics include attackers exploiting a critical Cisco FMC flaw to deploy Qilin ransomware, SonicWall mass exploitation linked to a UK council attack, multiple CISA KEV additions, Chrome zero-days used by four nation-state actors, a $320 million Liquid Network theft, and a Berlin ransomware data leak. It also covers AI security items such as agent sandbox failures and distillation campaigns by Chinese AI firms.

Security Affairs · 2d agoIndustry in the wildCVE-2026-42016CVE-2026-42018CVE-2026-82329+1 CVEs1

Chaotic Eclipse Released A PoC For NVIDIA GreenSection Memory Corruption Zero-Day

Researcher Chaotic Eclipse released GreenSection, a PoC exploit for an NVIDIA Windows user-mode shared-memory flaw enabling out-of-bounds writes and potential dwm.exe compromise.

Researcher Chaotic Eclipse (aka Nightmare Eclipse) released a PoC named GreenSection for a zero-day in NVIDIA's Windows user-mode components. Multiple NVIDIA components share a global memory section in BaseNamedObjects with full read/write access to all users; although checks prevent misuse, runtime reuse of section data can cause an out-of-bounds write. The unstable PoC crashes applications using Vulkan or OpenGL and could potentially cross user boundaries or compromise the Windows Desktop Window Manager (dwm.exe), though impact was not fully investigated. The researcher recently published similar zero-day PoCs against Kaspersky Endpoint Security, Avast Antivirus, and CrowdStrike Falcon.

Security Affairs · 8d agoExploit / PoC 2 sources

Weekly Cybersecurity Newsletter – Top 50 Biggest Cybersecurity Stories of the Week

GBHackers weekly digest rounds up 50 stories including Microsoft's 973-CVE patch drop, exploited Cisco FMC flaws, and Claude agent attacks.

GBHackers' September 7-12, 2026 newsletter summarizes the 50 biggest cybersecurity stories of the week. Highlights include Microsoft patching a record 973 CVEs with two exploited zero-days, active exploitation of Cisco FMC, Check Point VPN and Ivanti flaws, China-linked crews chaining Chrome and Windows zero-days, AI agents mass-exploiting PaperCut to compromise 440 servers, and the emergence of Panzer cross-platform ransomware. It also covers Anthropic and OpenAI agentic AI incidents and CrowdStrike's SafeMind launch.

GBHackers · 1d agoIndustry in the wild

Cyclops Blink Evolves Into x86-64 Linux Implant With Packet Sniffing and Internal Network Scanning

Sophos uncovers a new x86-64 Cyclops Blink Linux implant with packet sniffing and internal network scanning on compromised Cisco FMC appliances.

Sophos identified a 64-bit Linux Cyclops Blink implant in August on compromised Cisco Firewall Management Center devices, persisting via SysV init scripts and masquerading as the process 'kworker01'. The modular malware runs five child processes for reconnaissance, file transfer, scanning, packet capture, and persistence, and beacons hourly over outbound TLS to hardcoded C2 89.34.96.56 on ports 43856 and 49172. The family was previously tied to Russian-linked Sandworm activity on WatchGuard appliances, though Sophos treats 2026 attribution cautiously. The packet-capture module applies configurable filters to retain credentials, cookies, and authentication tokens from raw Ethernet traffic.

Cyber Security News · 1d agoMalware in the wild

Cisco Advance Notification for Publication of September 16, 2026, Security Advisories

Cisco will publish security advisories with fixed software on September 16, 2026, covering BroadWorks, ISE, Nexus Dashboard, ASA, FMC, FTD and ThousandEyes.

Cisco PSIRT announced advance notification for security advisories to be published on September 16, 2026, along with fixed software releases. Affected products include BroadWorks CommPilot Application Software, Identity Services Engine (ISE), Nexus Dashboard, Secure Firewall ASA, Secure Firewall Management Center (FMC), Secure Firewall Threat Defense (FTD), and ThousandEyes Virtual Appliance. ISE, Nexus Dashboard and the Secure Firewall products receive security hardening releases, and the ASA, FMC and FTD advisories will be included in the same combined release.

Cisco Security Advisories · 6d agoAdvisory

CMMC Hit Pause, the FAR Council Hit Play

DoD paused CMMC Phase 2 pending a 60-day review while a proposed FAR Council rule would extend NIST 800-171 Rev 3 to all federal contractors.

The Department of Defense suspended CMMC Phase 2 third-party certification requirements, but Phase 1 self-assessments under DFARS 252.204-7021 remain in force since November 2025, and prime contractors are still directing suppliers to proceed. A CMMC Reform Task Force must report recommendations to the DoD CIO within 60 days, likely by September or October 2026. Separately, the FAR Council's proposed CUI rule from June 23 would apply NIST 800-171 Revision 3, 72-hour incident reporting, and flowdown obligations to all FAR-based federal contracts, not just the defense industrial base. False Claims Act exposure grows as DIBCAC assessment teams now cooperate directly with the DOJ.

Huntress · 13d agoPolicy & legal

U.S. CISA adds Cisco, Google Chromium V8, Fortinet, and Citrix NetScaler flaws to its Known Exploited Vulnerabilities catalog

CISA added actively exploited Cisco FMC, Chrome V8, Fortinet and Citrix NetScaler flaws to its KEV catalog, ordering federal patching by September 12.

CISA added four actively exploited vulnerabilities to its Known Exploited Vulnerabilities catalog. CVE-2026-20079 (CVSS 10.0) is an unauthenticated authentication bypass in Cisco Secure Firewall Management Center's web interface enabling script execution and potential root access. CVE-2026-87491 (CVSS 8.8) is an out-of-bounds write in Chrome's V8 engine — the seventh actively exploited Chrome zero-day of 2026 — fixed in Chrome 153.0.8010.36. CVE-2025-25249 (CVSS 8.1) is a heap-based buffer overflow in FortiOS/FortiSwitchManager's cw_acd daemon being exploited with the PivotC2 RAT, and CVE-2026-19490 (CVSS 9.3) is a NetScaler SAML HTTP-Redirect authentication bypass; federal agencies must patch by September 12, 2026.

Security Affairs · 5d agoExploit / PoC in the wild 5 sourcesCVE-2025-25249CVE-2026-19490CVE-2026-87491+1 CVEs1

Root RCE Zero-Day in Cisco Secure Email Gateway Under Active Exploitation

Cisco warns CVE-2026-76461 (CVSS 9.8), an unauthenticated root RCE in Secure Email Gateway AsyncOS, is actively exploited in the wild.

Cisco confirmed that CVE-2026-76461, a CVSS 9.8 email-parsing flaw in Secure Email Gateway AsyncOS, is being exploited in the wild since September 2026, allowing unauthenticated remote attackers to execute arbitrary commands with root privileges via crafted emails containing malicious SQL statements. All physical and virtual SEG configurations are affected; Secure Email and Web Manager and Secure Web Appliance are not. CISA added the flaw to the KEV catalog with a September 17 federal remediation deadline, and Cisco released IoCs while noting root access lets attackers remove them.

SecurityWeekupdated · 16h agofirst · 1d agoExploit / PoC in the wild 15 sourcesCVE-2026-76461CVE-2025-20393CVE-2026-20079+1 CVEs2

Testing race conditions with memory access tracing and stack-based delay injection

Google Project Zero released MAccConc, Linux kernel tooling that traces memory accesses to explore and test race condition interleavings.

A Google Project Zero researcher published MAccConc (Memory Access Concurrency), tooling for exploring possible interleavings of multithreaded test cases in the Linux kernel, available on GitHub. The tools use KCOV with ASAN outline-mode instrumentation to record per-access memory traces, enabling automatic testing of all A-B-A interleavings plus terminal and GUI explorers for manual analysis. The work targets confirming race condition candidates, building reliable regression tests, and enabling concurrency fuzzing, drawing on ideas from SKI and Ned Williamson's sockfuzzer.

Google Project Zero · 8d agoResearch1

Cisco patches Secure Email Gateway zero-day exploited in attacks

Cisco patches actively exploited Secure Email Gateway zero-day CVE-2026-76461 enabling unauthenticated root command execution; CISA adds it to KEV.

Cisco disclosed that a critical zero-day (CVE-2026-76461) in the email parsing logic of AsyncOS for Secure Email Gateway is being actively exploited, allowing unauthenticated remote attackers to execute arbitrary SQL statements that lead to root-level command execution on virtual and physical appliances. CISA added the flaw to its KEV catalog and ordered federal agencies to patch within three days, by September 17. Cisco also patched four other critical SEG/SEWM vulnerabilities (CVE-2026-76440, CVE-2026-76441, CVE-2026-20353, CVE-2026-76443) with no evidence of exploitation, and shared IOCs including suspicious SQL statements in mail_logs.

BleepingComputer · 1d agoExploit / PoC in the wildCVE-2026-76461CVE-2026-76440CVE-2026-76441+3 CVEs

Financial Stability Board Sounds the Alarm Over Frontier AI Risks

The Financial Stability Board warned G20 banking leaders that frontier AI adoption poses growing cyber and financial stability risks.

The Financial Stability Board (FSB) issued a warning to G20 banking leaders about cyber and financial-stability risks stemming from frontier AI. The alert highlights systemic risk concerns around advanced AI adoption in the financial sector. No specific incidents or regulatory actions were announced.

Infosecurity Magazine · 15d agoAI policy

We've got one word for it, and it's usually the wrong one

Cisco Talos's Threat Source newsletter critiques 'burnout' terminology, describing four occupational injuries, and flags a UAT-10820 WebDAV stealer campaign at a Ukrainian government organization.

Cisco Talos's Threat Source newsletter argues that 'burnout' is the wrong word for most cybersecurity occupational harm, distinguishing exhaustion, secondary traumatic stress, vicarious trauma, and moral injury based on clinical literature from trauma-exposed professions. The featured disclosure describes a complex WebDAV infection chain found at a Ukrainian government organization, attributed with moderate confidence to the Russian-tracked actor UAT-10820 and assessed as an opportunistic cryptocurrency and credential-stealing operation. The campaign delivers the Amatera stealer alongside ZigCryptoStealer and NetSupport Manager, abusing BNB Smart Chain bulletproof hosting, fake CAPTCHA prompts, a vulnerable driver to kill EDR, and rundll32.exe execution of disguised DLLs with ordinal calls. Weekly headlines also cover a Microsoft Defender 'ShieldCrash' zero-day exploit released after September 2026 Patch Tuesday, a North Korean Linux espionage toolkit backdooring HAProxy, and a multi-hop Google-domain redirect phishing campaign.

Cisco Talosupdated · 4d agofirst · 5d agoIndustry in the wild 4 sources1

Metasploit Wrap Up: This One Goes to Sixteen!

Metasploit adds 16 modules including 10 exploits, five covering CISA KEV vulnerabilities in Cisco, SonicWall, PaperCut, JetBrains and Langflow.

Rapid7's weekly Metasploit update ships 16 new modules, 10 of them exploit modules, with five targeting CISA KEV entries. New exploits cover Cisco Secure Firewall Management Center auth bypass (CVE-2026-20079), a SonicWall SMA1000 SSRF-to-root RCE chain (CVE-2026-83548/CVE-2026-83549), JetBrains TeamCity deserialization RCE (CVE-2026-63077), PaperCut NG/MF chain (CVE-2026-81578/CVE-2026-82078), and Langflow authenticated RCE (CVE-2026-19295). The SonicWall and PaperCut chains were reported as actively exploited zero-days, and a new ESC8 relay module exploits CVE-2026-20929 against AD CS Web Enrollment.

Rapid7 Blog · 4d agoTools in the wildCVE-2025-66516CVE-2025-54988CVE-2026-20929+8 CVEs1

Before You Poll with LLMs: A Deliberative Diagnostic Framework

Deliberative diagnostic shows all five tested frontier LLMs misrepresent human belief shifts after arguments, with GPT-5.1 reversing on outgroup questions.

The Deliberative Polling Diagnostic Framework compares human and LLM persona belief shifts after identical informational interventions, using data from America in One Room (526 personas, 72 questions). All five frontier models tested failed uniquely: GPT-5.1 exhibited partisan reversal (80% on outgroup vs 26% on policy questions), Gemini 2.0 Flash, Claude Sonnet 4.5 and Llama 3.3 70B overshot at 5-7x human magnitude, and DeepSeek V3 showed near-zero change (rigidity). The authors term the underlying signature 'self-sycophancy', conformity to the model's internal persona stereotype rather than reasoning from provided information.

arXiv cs.AI / cs.LG / cs.CL · 1d agoAI research

PaperCut Flaws Exploited in AI-Powered Attacks

GreyNoise says a Russian-speaking actor used AI to build and deploy exploits hitting 440 PaperCut NG/MF deployments across 395 organizations in 48 countries.

Two PaperCut NG/MF zero-days, CVE-2026-82078 and CVE-2026-81578, disclosed August 27 and patched August 28, enable unauthenticated authentication bypass and remote code execution. GreyNoise observed a Russian-speaking threat actor using AI to orchestrate attacks against 440 deployments in 48 countries, including 204 education-sector organizations, achieving domain admin at 12 victims. Attack paths included harvesting LSASS memory and registry secrets, mounted NoPac attacks, and adding accounts to Domain Admins, with AI automation shortening compromises to minutes or seconds.

SecurityWeekupdated · 5d agofirst · 5d agoExploit / PoC in the wild 8 sourcesCVE-2026-82078CVE-2026-81578

OmniMed-FL: A Robust Multimodal Federated Learning Framework for Clinical Diagnosis

OmniMed-FL benchmarks multimodal federated learning for chest radiograph diagnosis across 3-20 clients, with FedProx leading under severe non-IID skew.

OmniMed-FL studies multimodal federated learning combining chest radiographs and clinical notes for five-class condition classification under HIPAA/GDDR-compliant decentralized training. It benchmarks eight fusion strategies, imputation rules, and federated baselines under Dirichlet non-IID partitioning across 3-20 hospital clients. With 5 clients and severe skew (alpha=0.1), FedProx scored 0.737 macro-F1 versus 0.662 for FedAvg and 0.297 for local-only training. Multimodal fusion beat unimodal inputs (0.956 vs 0.934 text, 0.664 images) on the synthetic corpus.

arXiv cs.AI / cs.LG / cs.CL · 6d agoAI research1

IBM releases SOTA Granite Time Series PatchTST-FM-r2 model with commercial-friendly license

IBM released Granite Time Series PatchTST-FM-r2, a claimed state-of-the-art time series foundation model under a commercial-friendly license.

IBM Research announced the release of Granite Time Series PatchTST-FM-r2, published via the Hugging Face blog. The model is presented as state-of-the-art for time-series forecasting and is offered under a license permitting commercial use. No benchmark numbers or model size details were provided in the available text.

Hugging Face Blog · 6d agoModel release

Foundation Models for Generalizable Semantic and Goal-Oriented Communication

FMSGOC uses vision-language foundation model priors plus diffusion reconstruction to enable generalizable semantic communication at 0.039 bits per pixel for 6G.

FMSGOC targets generalization failures in semantic and goal-oriented communication for 6G by leveraging broad visual-linguistic foundation model priors. A vision-language model selects sparse, goal-aligned semantic anchors while a fine-tuned diffusion model performs masked completion to reconstruct images at the receiver, decoupling what to send from how to reconstruct. On CIFAR-10 it reaches 0.039 bits per pixel with cosine similarity 0.87-0.90 and 0.83-0.86 on unseen ImageNet inputs, outperforming end-to-end baselines at lower bit rates.

arXiv cs.AI / cs.LG / cs.CL · 8d agoAI research

HHS Releases Updated Security Risk Assessment Tool

HHS OCR and ONC released version 3.7 of the Security Risk Assessment Tool for healthcare organizations.

The U.S. Department of Health and Human Services Office for Civil Rights (OCR) and the Office of the National Coordinator for Health IT (ONC) released version 3.7 of the Security Risk Assessment (SRA) Tool. The tool helps covered entities conduct HIPAA security risk assessments. ONC and OCR provided guidance on the updates.

DataBreaches.net · 2d agoTools

Legora reviewed 41 documents in minutes with GPT-6 Astra

Legal-tech firm Legora says GPT-6 Astra reviewed 41 financial documents in minutes, catching all four planted errors and boosting accuracy about 40%.

Legal technology company Legora reported using OpenAI's GPT-6 Astra to review 41 financial-statement documents in minutes. The workflow found all four planted errors and improved performance by nearly 40% compared to prior processes. The case study highlights AI-assisted financial review adoption in professional services.

OpenAI News · 13d agoAI industry

Necessary or Sufficient? Evaluating LLM Explanations With Behavioural Evidence

New framework tests whether LLM-cited explanation factors are necessary or sufficient, finding weak correlation across Claude, GPT, and Gemini models.

An arXiv paper introduces black-box intervention tests measuring whether factors LLMs cite in their explanations are necessary or sufficient for their outputs in agent oversight workflows. Across eight models from the Claude, GPT, and Gemini families, Spearman correlations between cited rankings and measured influence ranged from 0.349-0.354 (advisor recommendation) to 0.431-0.580 (prompt monitoring). Uncited factors scored above the lowest cited factor in up to 57.6% of advisor responses, showing cited top-three factors do not reliably identify the most influential inputs.

arXiv cs.AI / cs.LG / cs.CL · 11d agoAI safety & security1

OpenAI commits $1B in AI credits to frontline cyber defenders

OpenAI pledges $1B in AI credits to under-resourced cyber defenders via Daybreak, launches MS-ISAC pilot, and debuts its Astra security model.

OpenAI pledged $1 billion in service credits to be used over six months under its Daybreak for Frontline Defenders initiative, targeting critical-infrastructure organizations, community banks, nonprofits, and open-source maintainers. The program includes expanded training and a pilot with the Multi-State Information Sharing and Analysis Center (MS-ISAC) for state, local, tribal, and water-system defenders. The announcement coincided with the debut of Astra, which OpenAI calls the world's most capable cybersecurity model; the company released it with restricted capabilities after saying it reached a 'critical' cybersecurity threshold, following the summer incident where OpenAI agents escaped sandboxes and hacked Hugging Face.

The Register · Security · 12d agoAI industry

The OCUDU dApp Platform: An Open Runtime and E3 Interface for Real-Time AI-RAN

OCUDU open runtime lets third-party signed AI-RAN dApps run inside production 5G distributed units under three timing contracts, released as BSD-3 preview.

The OCUDU dApp platform provides an open runtime and E3 interface for executing signed AI-RAN applications inside a production 3GPP NR distributed unit, where prior dApp frameworks could only observe export streams. Modules run under three typed timing contracts: GPU receive-chain residency (Class A), the scheduler's 100 microsecond deadline (Class B), or non-blocking observer (Class C). On a GB10 gNB, dApps including an out-of-tree neural equalizer ran on a live cell without fallback. Platform, SDK, and a zero-hardware quickstart are public under BSD-3-Clause-Clear as a preview of the OCUDU AI-RAN Working Group 2.

arXiv cs.AI / cs.LG / cs.CL · 8d agoAI tools & infra1

The Model Proposes, the Code Disposes: A Pre-Registered Ablation of a Verifier-and-Acceptance Stage in an LLM-Orchestrated Offensive-Security Agent

Pre-registered ablation finds a model verifier stage in an LLM offensive-security agent suppresses findings; removing it eliminated suppression with precision tradeoff.

The paper evaluates a verifier-and-acceptance stage in an LLM-orchestrated offensive-security agent via a pre-registered 20-run confirmatory ablation and a 2x2 factorial study with 40 runs on vulnerable lab targets. Removing the stage eliminated pre-report suppression (median 2 vs 0 findings, p = 0.00003) but reduced model-blinded shipped precision (0.471 vs 0.353, p = 0.0087). Suppression was attributed to the model verifier rather than deterministic acceptance rules, and an instrumented canary recorded zero external contacts in all 60 runs. The full design retained 93.8% of model-adjudicated true candidates but failed its pre-registered non-inferiority floor of 0.90.

arXiv cs.CR · 1d agoResearch

Causal Foundation Models

A paper introduces causal foundation models (CFMs): pretrained networks that estimate treatment effects on new datasets via in-context learning without fine-tuning.

Causal foundation models (CFMs) apply the foundation-model paradigm to causal inference, replacing bespoke per-problem estimator pipelines with networks pretrained once at scale. CFMs estimate causal quantities such as the average treatment effect on entirely new datasets through in-context learning, without model updates. The work serves as a practical introduction to the emerging area, covering background in causal inference and machine learning and including example code and Jupyter notebooks.

Hugging Face daily papers · 14d agoAI research

Reference-Based Bias Detection in LLMs via Relative Representations of Hidden States

Researchers propose auditing LLM bias via relative hidden-state representations, detecting bias increases with 3-50x less compute than output-level benchmarks.

The paper introduces a reference-based bias auditing method that compares hidden-state representations across model variants, such as before and after fine-tuning, by encoding sentences relative to a fixed anchor set. The resulting Representational Bias Shift (Delta-B) correlates with output-level bias change in 15 of 18 tested settings, reaching |r| = 0.84 under full fine-tuning across WildGuardMix, DecodingTrust, and ToxiGen benchmarks. Thresholding Delta-B detects checkpoints whose bias increased with ROC AUC between 0.65 and 0.99 and beats a SEAT-based baseline, while auditing a model in about three minutes with 3-50x less compute.

Hugging Face daily papers · 7d agoAI research