60
Smart search ranks by meaning as well as keywords (one row per story, last 45 days).
47
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.
40
FBI Seizes China-Linked Hacking Platforms QScan and QTRouter Used Against Critical Infrastructure
FBI seizes China-linked QScan and QTRouter hacking platforms used by QTFY to obfuscate intrusions against US federal agencies.
The DOJ and FBI seized domains hard-coded into QScan and QTRouter, two platforms operated by China-based Nanjing Xinjiuwei Network Technology Company on behalf of state-sponsored group QTFY. QScan automatically infected thousands of IoT devices which were added to QTRouter, an obfuscation network routing malicious traffic through compromised and proxy devices outside China. Targets included NASA, the Federal Reserve, Departments of Energy, Justice, and HHS, NIH, and the US Senate, exploiting flaws in Fortinet SSL-VPN, Citrix ADC, Microsoft Exchange, F5 BIG-IP, Log4j, and others.
78
New TCG guidance gives buyers a way to test PQC-ready TPM claims
Trusted Computing Group published PTP 1.07-based requirements defining PQC-ready and PQC-upgradable TPM designations to guide buyers assessing quantum-safe hardware.
The Trusted Computing Group published requirements establishing PTP 1.07, based on TPM 2.0 Library Specification Version 1.85, as the baseline for a 'TCG PQC-ready TPM', and defined a second designation, 'TCG PQC-upgradable TPM', for chips that can be upgraded to that support. The guidance lets buyers request vendor evidence that TPMs meet end-to-end post-quantum requirements for identities, attestation, and hardware-anchored trust rather than single-algorithm compliance. TCG also announced plans to extend its certification programs to certify TPMs meeting PTP 1.07, noting that 90% of businesses still lack a formal PQC roadmap.
20
AI Is Giving Lesser-Resourced Attackers Nation-State-Level Reach, Google Warns
Google's Threat Intelligence Group warns AI now gives lesser-resourced criminal and nation-state attackers nation-state-level speed and scale, citing TeamPCP, Basin Castle, APT42, and APT24 usage.
GTIG documented throughout 2026 that adversaries increasingly use AI to automate and scale attacks. TeamPCP (UNC6780) used an AI coding chatbot with agent instructions to plan and execute a mass credential harvesting campaign in under six hours, and has compromised PyPI, npm, and Docker Hub since March 2026 with its Dustmaker credential stealer plus released tools Shai-Hulud and Miasma. PRC-nexus Basin Castle uses LLMs for target profiling, lure drafting, and malware development; APT42 (Calanque Ion) uses Gemini for OSINT and localized lures; APT24 (Ravine Castle) uses Gemini across the full attack lifecycle; and DPRK's Midnight Neptune (UNC1069) integrates AI into cryptocurrency theft. Google responds by disrupting attacker accounts and hardening models against distillation attacks.
68
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.
20
FBI takes down China-linked hacking network behind attacks on NASA, DOJ and U.S. Senate
FBI seized domains disabling QScan and QTRouter malware run by China-linked QTFY group behind intrusions at NASA, DOJ, Senate and other agencies.
The Justice Department and FBI seized domains hard-coded into two malware tools, QScan and QTRouter, operated by a Chinese state-sponsored group called QTFY, tied to a Nanjing-based company that sold hacking services to China's Ministry of State Security and the PLA. QScan infected IoT devices worldwide while QTRouter combined them with commercial proxies and rented servers to build an obfuscation network that masked attack origins. Victims include NASA, the Federal Reserve, the Departments of Energy and Justice, HHS, NIH, and the U.S. Senate. The FBI and NSA published a joint advisory with indicators of compromise, the latest in operations against Mustang Panda, Flax Typhoon, and Volt Typhoon infrastructure.
80
PLC-DPO: Posterior Label Correction in Noisy and Ambiguous Preference Optimization
PLC-DPO corrects noisy preference labels during DPO by routing each pair as clean, flip, or tie using calibrated policy-reference margins.
PLC-DPO addresses reversed, weak, or ambiguous preference labels in Direct Preference Optimization by using the calibrated policy-reference margin as online evidence to route each pair's training signal as clean, flip, or tie. This actively corrects supervision direction and strength rather than only filtering suspicious examples. Across 57 dataset-model-benchmark cells, PLC-DPO achieves the best mean win rate of 60.5 versus 55.5 for the next-best method. Injected-noise and tie stress tests plus human disagreement analysis show the routing stays stable and distinguishes flipped from weakly directional pairs.
28
Red Heron Hackers Exploit Critical Gitea RCE to Steal Source Code and Deploy Linux Rootkit
PRC-linked Red Heron exploits critical Gitea RCE CVE-2026-60004 to steal source code and deploy JITTERLY implant with SIXZUT LD_PRELOAD rootkit; victims span five countries.
Acronis Threat Research Unit attributes a campaign to Chinese-speaking threat actor Red Heron, which weaponized CVE-2026-60004, a CVSS 9.8 RCE in Gitea versions 1.17 through 1.27.0, patched in 1.27.1 on July 27, 2026. The actor built an automated exploitation framework after a public PoC appeared, scanned 1,386 internet-exposed Gitea instances across seven countries, and separately listed 477 Taiwan-based systems across defense, energy, elections, and AI sectors. Confirmed victims include organizations in Canada, Argentina, Taiwan, the US, and Sri Lanka, with a Canadian renewable-energy firm hit in 22 sessions and a Taiwanese industrial automation firm losing hundreds of repositories including SCADA/HMI tools. Red Heron deploys the JITTERLY Linux implant (30+ commands, AES-128-GCM, Adaptix-like protocol) and the SIXZUT LD_PRELOAD rootkit disguised as libglthread.so.2, and moved laterally into a Synology/Proxmox environment to steal VM backups.
84
ECHO: A Matched-Contrast Benchmark for Context-Sensitive Turn-Taking in Full-Duplex Dialogue
ECHO benchmark pairs identical-overlap Chinese dialogue examples with contrasting contexts, revealing most full-duplex systems bias toward yielding the floor.
ECHO is a paired diagnostic benchmark for Chinese full-duplex turn-taking that matches examples with identical overlap transcripts but contrasting preceding multi-turn contexts, one requiring Yield and the other Keep. It also includes off-talk examples for diagnosing unnecessary yielding and introduces pair accuracy, which grants no credit to constant-action policies. Experiments across multiple full-duplex systems show most exhibit a pronounced Yield bias, performing substantially better on interruptions than backchannels. The benchmark and metadata will be publicly released.
40
Cognition launches new SWE-2 model, Rivaling Fable 5.1 and GPT-Astra
Cognition released SWE-2, a coding model post-trained from Kimi K3 that scores 50.0% on FrontierCode 1.1 Main, near Fable 5.1 at 64% lower cost.
Cognition introduced SWE-2, its most advanced coding model, post-trained from the 2.8T-parameter Kimi K3 base model. It achieves 50.0% on FrontierCode 1.1 Main, 73.0% on DeepSWE 1.1, and 92.8% on Terminal-Bench 2.1, beating Grok 4.6 and SWE-1.7 while matching Fable 5.1 and GPT-5.6 Sol at a fraction of the price. The company says it scaled reinforcement learning to the multi-trillion-parameter regime for the first time, using Pareto-informed cost penalties that train all reasoning-effort levels in a single run, tripled RL environments, and NVFP4/FP8 quantization-aware training. SWE-2 is available today in Devin Desktop and CLI, with rollout on Devin Web and Fusion.
72
China spy chief points at US AI models in cyber threat warning
China's MSS chief Chen Yixin named Anthropic's Claude Mythos and OpenAI's GPT-5.5-Cyber as cyber threats to Chinese critical infrastructure.
Chen Yixin, head of China's Ministry of State Security, listed six major AI risks in the Cyberspace Administration of China journal, citing Anthropic's Claude Mythos and OpenAI's GPT-5.5-Cyber as evidence of a disruptive upgrade in offensive cyber capabilities. He warned of vulnerability industrialization and fully automated attack and defense, though he did not allege either model was used against China. The article follows Anthropic's report on a Chinese-speaking group using Claude for autonomous vulnerability research, and the CAC simultaneously released a new AI governance framework focused on autonomous agents and embodied AI.
55
Google researchers uncover criminal zero-day exploit likely built with AI
Google links a likely LLM-built criminal zero-day for an open-source admin tool to planned mass exploitation and maps AI-assisted threats.
Google Threat Intelligence Group linked a zero-day exploit for a popular open-source web-based administration tool, enabling 2FA bypass with valid credentials via a semantic logic error, to a criminal group, citing educational docstrings, a hallucinated CVSS score, and textbook Python as signs of LLM authorship; the vendor was notified before a planned mass exploitation campaign. The report also details Russia-nexus malware families CANFAIL and LONGSTREAM using AI-generated decoy code, the PROMPTSPY Android backdoor driving the UI through the Gemini API, APT27 using Gemini to build relay tooling, and the TeamPCP (UNC6780) supply chain compromise of LiteLLM and Trivy repositories that planted the SANDCLOCK credential stealer.
78
PIA-Bench: Towards Automated Privacy Impact Assessment with Large Language Models
Researchers release PIA-Bench, the first open benchmark evaluating how accurately LLMs can automate privacy impact assessments using 73 curated federal PIAs.
PIA-Bench is the first open benchmark for evaluating large language models on real-world privacy impact assessments (PIAs). The authors audited 499 expert-authored PIAs published by US federal agencies and curated 73 structured PIAs comprising 451 privacy risk items and 831 mitigation items. Off-the-shelf LLMs were found to produce meaningful assessments while identifying clear avenues for improvement. The paper calls for domain-specific LLM agent workflows, accountable LLM infrastructure, and new quality standards for PIAs.
48
GTIG AI Threat Tracker: From Prompting to Autonomy – The Evolution of Adversarial AI
GTIG's Q2 2026 tracker shows adversaries adopting agentic AI workflows, including credential harvesting in under six hours and supply chain attacks by UNC6780.
Google Threat Intelligence Group's Q2 2026 report documents adversaries moving from basic prompting to agentic AI workflows and automation, including a cloud compromise followed by agent-enabled mass credential harvesting executed in under six hours. It tracks financially motivated actor UNC6780 (TeamPCP) conducting large-scale open source supply chain compromises across PyPI, npm, and Docker Hub since March 2026, deploying credential stealers. The report also highlights growing targeting of proprietary AI models, source code, prompts, and API credentials, plus LLMJacking practices where adversaries steal developer credentials or hijack cloud infrastructure to run unauthorized AI workloads.
65
Claude is a Contrarian
Opinion piece argues Claude habitually contradicts explicit user instructions, injecting contrarian content despite CLAUDE.md rules and user objections.
A developer recounts repeated instruction-following failures with Claude, claiming it contradicts explicit requests, adds unnecessary work, and ignores AGENTS.md and CLAUDE.md directives. The author contrasts this with OpenAI, DeepSeek, and Qwen models, which he says more readily apologize and undo mistakes. He theorizes Claude's training makes it assume the human is wrong and needs correcting. The post is personal commentary with no benchmarks or systematic evaluation.
12
I Think the Military Commissary Freezers Were Hacked
Refrigeration failures at six-plus US military commissaries prompt speculation of a cyber attack on DeCA's remote monitoring systems; Pentagon acknowledges possible disruption.
The author documents near-simultaneous freezer and refrigeration failures at confirmed installations including Fort Huachuca, F.E. Warren AFB, Fort Irwin and Travis AFB on August 26-27, with freezers entering defrost mode that heated and spoiled food. DeCA's Remote Monitoring Control System controls defrost across roughly 182 locations, and an unverified comment attributed the Fort Huachuca failure to a network issue. Stars and Stripes and Military Times independently reported the multi-base failures, and the Pentagon acknowledged a 'possible refrigeration disruption,' though no evidence of hacking has been confirmed.
45
60
Nearly half of enterprises have no one leading PQC migration
Axiad survey finds 46% of enterprises lack a single leader for post-quantum cryptography migration despite most maintaining cryptographic inventories.
Axiad research on post-quantum cryptography (PQC) readiness found 75% of respondents maintain continuously updated inventories of certificates, keys, and algorithms, but 46% have no single person owning the migration. About half have never formally assessed whether public-facing infrastructure supports post-quantum key exchange. 67% said the harvest-now-decrypt-later threat is an active priority, with competing priorities, budget constraints, and lack of regulatory guidance cited as top obstacles; executives were more confident than PKI practitioners.
22
Threat actors are giving AI agents a bigger role in cyberattacks
Google Threat Intelligence's Q3 2026 tracker shows threat actors using AI agents for autonomous credential harvesting, including a six-hour campaign compromising thousands of credentials.
Google Threat Intelligence Group's Q3 2026 AI Threat Tracker, based on Mandiant incident response and platform telemetry, documents adversaries shifting toward AI-driven multi-step attack workflows. Mandiant investigated a financially motivated actor that compromised cloud infrastructure and used an AI coding chatbot plus agent instructions to run vulnerability scanning, troubleshooting, and IP rotation autonomously, harvesting thousands of third-party credentials in under six hours; an exposed 'Recon' C2 server held over 23,800 harvested secrets including cloud and AI API keys. Separately, two PRC-nexus espionage actors experimented with Gemini, Claude, and Codex to build automated penetration testing and exploitation frameworks, with Shai-Hulud used for C2 and credential harvesting. GTIG says no fully autonomous attack pipelines have yet been observed in the wild.
62
openbmb/MiniCPM5-2B-GGUF — new model trending #30 on Hugging Face
OpenBMB released MiniCPM5-2B, a dense 2B on-device model claiming open-source SOTA among 2B-class models.
OpenBMB released MiniCPM5-2B, the second model in the MiniCPM5 series following MiniCPM5-1B, as a dense 2B Transformer built for on-device and resource-constrained deployment with GGUF weights on Hugging Face. The team claims 2B-class open-source state-of-the-art performance, remaining competitive with 4B-class models in coding, mathematics, long-context understanding, tool use and agentic tasks. The release includes a tech report, GitHub repository and online demo, and is currently trending on Hugging Face.
45
FBI cyber chief worries private sector not sharing enough cyber threat information
FBI cyber chief Brett Leatherman urged companies to share breach information with the bureau as it publishes a victim-focused cyber strategy.
FBI Cyber Division assistant director Brett Leatherman said at the Billington CyberSecurity Summit that private-sector hesitancy to engage the FBI stems from misconceptions, including a belief that shared incident data is passed to regulators. He warned that organizations breached by PRC nation-state actors risk more by handling intrusions alone, since FBI involvement speeds eradication. The bureau published a new cyber strategy Wednesday emphasizing victim aid, adopting a 'share until it hurts' posture on releasing threat intelligence.
38
Beijing Hits Back at Anthropic CEO’s Call to Curb China’s AI Development
China's Foreign Ministry rejected Anthropic CEO Dario Amodei's call to curb Chinese AI development as 'Cold War playbook' fearmongering.
China's Ministry of Foreign Affairs pushed back against Anthropic CEO Dario Amodei's essay warning that a Chinese AI lead would endanger the US, with spokesperson Guo Jiakun saying fearmongering and vicious competition disrupt global AI governance. Amodei urged continued restrictions on cutting-edge AI chip sales to China and predicted AI agents could take over the internet within 6-12 months without a global slowdown. The exchange precedes a planned September 24 Trump-Xi meeting covering AI governance, and follows a joint FBI/NSA/CISA advisory alleging Chinese developers distilled capabilities from Claude and GPT.
48
60
Threat actors are coming for your AI assets to operationalize their use of AI
Google GTIG reports espionage and crime groups stealing AI models, prompts, and API credentials, plus distillation campaigns and agentic AI attack automation.
Google Threat Intelligence Group's quarterly AI Threat Tracker reports adversaries stealing proprietary models, source code, prompts, and API credentials from government, healthcare, and media targets, including China-based UNC6508 compromising clouds to run unauthorized LLM workloads. Distillation campaigns against Google's models exceeded 100 million prompts launched via thousands of stolen account credentials through proxy networks. Mandiant also observed a financially motivated actor deploy an autonomous multi-agent framework that harvested thousands of third-party credentials in under 6 hours, and a 'Recon' framework on a live C2 server managing over 23,000 stolen credentials including cloud and AI API keys.
72
G7 Urges Fast-Track on Quantum-Safe Cybersecurity Rules
G7 cybersecurity agencies led by France's ANSSI urged accelerated transition to post-quantum cryptography, prioritizing critical systems and phased, risk-based migration.
Under France's 2026 G7 Presidency, ANSSI, chairing the G7 Cybersecurity Working Group, published a September 3 call to action urging governments and organizations to begin quantum-safe (PQC) transitions now, reframing the quantum threat as near-term. The document, signed by the national cyber agencies of all G7 members and supported by the EU Commission and ENISA, outlines five priorities including national PQC strategies, R&D, public-private partnerships, and integrating PQC into cybersecurity requirements. It recommends cryptographic inventories, dependency mapping, prioritizing the most critical systems, and buying PQC-integrated products during normal renewal cycles. ANSSI will stop vetting non-quantum-safe products in 2027, with PQC mandatory in some security product procurement by 2030.
60
DoJ Corrects China Hacking Claim, Says U.S. Agencies Were Targets, Not Victims
DoJ corrected its statement to say US federal agencies were targets, not confirmed victims, of China-linked group QTFY's intrusions.
The US Department of Justice revised its press release to list NASA, the Federal Reserve, Department of Energy, DoJ, HHS, NIH and the US Senate as 'among the targets' of QTFY (aka QT AND QTCYBER), rather than victims. QTFY, active since 2018 and linked to Nanjing Xinjiuwei Network Technology Co with payments suggesting MSS sponsorship, operates the QScan vulnerability scanning platform and QTRouter obfuscation network, which underpin the Fast Labyrinth encrypted ORB relay network built from infected IoT devices and leased VPSs. The FBI seized qtproxy[.]xyz, qt-proxy[.]org and qt-team[.]com, disrupting QScan and QTRouter, while Lumen Black Lotus Labs reported the actor industrialized ORB networks for China-linked espionage. A 2019 NASA intrusion attempt exploited CVE-2019-11510, a critical Pulse Secure VPN flaw.
70
New Guidance Helps Businesses Verify Quantum-Safe Hardware Claims
TCG issued guidance to help businesses verify that trusted platform modules genuinely meet quantum-safe requirements.
The Trusted Computing Group (TCG) published new guidance aimed at proving that trusted platform modules (TPMs) genuinely satisfy essential quantum-safe requirements. The document gives businesses a way to validate vendor claims about post-quantum readiness in hardware security rather than trusting marketing assertions. This is a guidance publication, not an incident or vulnerability disclosure.
25
China-Linked Hackers Exploit Sogou One-Click RCE to Deploy GRAYRABBIT Backdoor
China-linked UNC3569 exploited CVE-2026-51990 in Sogou Input Method to deploy the GRAYRABBIT backdoor in active espionage intrusions.
Gen Threat Labs discovered UNC3569 exploiting CVE-2026-51990, a one-click RCE in Tencent's Sogou Input Method for Windows that chains an insecure sgbiz: protocol handler with an unsandboxed Chromium 80 CEF webview. The chain weaponizes CVE-2021-38003 (V8 type confusion) to run shellcode that DLL-sideloads via 7z.exe/7z.dll and deploys the GRAYRABBIT backdoor, which beacons over RC4-encrypted raw TCP 443 to mail.uaiubifas[.]top. Tencent patched the issue in version 16.3.0.3498, released via automatic updates on April 21, 2026. UNC3569 is a PRC-nexus espionage actor targeting government, education, technology, and financial sectors across East and Southeast Asia.
82
ZGCM-1: A Fully Open and Extremely Efficient Foundation Model for Math and Agentic Search
ZGCM-1 is a fully open 7B foundation model with 256K context that stays competitive with frontier models on math reasoning and agentic search.
ZGCM-1 is a fully open 7B dense foundation model trained from scratch using an efficiency-focused recipe: interleaved gated sliding-window and full attention, a stable FP8 Muon optimizer, and MDP-based mid-training with context scaling across 16K, 64K, and 256K. On mathematical reasoning and agentic search suites it remains competitive with much larger frontier models such as Qwen3-235B-A22B and GLM-5.1. The recipe yields a ~4.2x improvement in 16K pre-training time-to-loss, and all weights, checkpoints, training code, data recipes, and W&B logs are open-sourced.
50
FBI Disrupts China-Linked QTFY Infrastructure Used to Steal Data From U.S. Organizations
FBI and Lumen disrupted QTFY's QScan and QTRouter botnet platforms used by Chinese state-sponsored hackers to conceal intrusions into U.S. agencies.
The U.S. DoJ announced court-authorized seizure of domains behind QScan and QTRouter, operated by the Chinese state-sponsored group QTFY and employed by Nanjing Xinjiuwei Network Technology Company. QTFY has been active since May 2018 and targeted NASA, the Federal Reserve, the Department of Energy, DoJ, HHS, NIH, the U.S. Senate, and academic institutions. QScan exploits vulnerable IoT devices, feeding them into QTRouter, an OpenWrt-based proxy obfuscation network likened to an operational relay box (ORB) that masks attack origins. The group exploited zero-days such as Ivanti CSA flaws CVE-2024-8190, CVE-2024-8963, and CVE-2024-9380, plus numerous N-days, and maintained persistence with RATs, web shells, and legitimate credentials.
78
Comfy-Org/YuE2 — new model trending #30 on Hugging Face
m-a-p's YuE2-3B music generation model and SheetSage2 audio encoder are repackaged in bf16 for ComfyUI and trending #30 on Hugging Face.
Comfy-Org published repackaged bf16 safetensors files for m-a-p's YuE2-3B model and its SheetSage2 audio encoder, organized into ComfyUI checkpoints and audio encoder folders. The repository links to the original m-a-p/YuE2-3B and m-a-p/SheetSage2 model pages and is currently trending #30 on Hugging Face.
20
Major Cyber Threat Detection Vendors Shift from MITRE to UK Testing Program
SE Labs launched PIVOT, a six-month vendor detection testing program backed by CrowdStrike, Fortinet, Palo Alto Networks and Sophos, as major vendors exit MITRE evaluations.
SE Labs unveiled PIVOT on September 15, a six-month testing program in which its ethical hackers replicate nation-state and criminal attack chains against participating vendor products, with results due January 2027. Broadcom (Symantec/Carbon Black), CrowdStrike, Fortinet, Palo Alto Networks and Sophos have confirmed participation, and Gartner and Forrester analysts will verify the underlying evidence before publication. The launch follows declining participation in MITRE Engenuity ATT&CK Evaluations: Enterprise, which fell from 30 vendors in 2023 to 11 in 2025 after public withdrawals by Microsoft, SentinelOne and Palo Alto Networks.
40
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.
44
Hackers Weaponize Agentic AI to Automate Reconnaissance, Exploitation and Post-Exploitation
Google GTIG reports threat actors using agentic AI to automate reconnaissance, exploit selection, and credential harvesting, compromising thousands of secrets.
Google Threat Intelligence Group's Q3 2026 AI Threat Tracker documents threat actors operationalizing agentic AI: in one Mandiant investigation, a financially motivated actor built and executed a credential-harvesting operation in under six hours, with an exposed 'Recon' framework managing more than 23,800 harvested secrets including cloud and AI-service API keys. A Chinese-speaking actor tracked as knaithe used a DeepSeek-powered Hermes Agent for automated reconnaissance and vulnerability enumeration, pivoting from Langflow to n8n and enabling manual exploitation of exposed Citrix NetScaler, Marimo, Apache Tomcat, and VPN infrastructure. Operators harvested Citrix session cookies from process memory to bypass MFA, obtained AWS credentials from compromised Marimo instances, and deployed the Go-based NKAbuse backdoor, with reported RCE and data exfiltration. Google notes fully autonomous end-to-end AI attack pipelines have not yet been observed in the wild.
72
Chinese Hackers Use AI Agents in Multi
China-linked campaign used the SecFlow AI-agent framework (Claude, Qwen, DeepSeek) to automate intrusions against government targets in Taiwan, Indonesia, China and Vietnam.
Hunt.io documented a second China-linked campaign wiring commercial AI models into live cyberespionage, reconstructing the SecFlow orchestration system from five accidentally exposed open directories. Targets included Taiwan's Kuomintang Party archives, Indonesia's Ministry of Foreign Affairs, mainland Chinese government and education systems, and Vietnamese industrial hosts. The most extensive compromise hit a Fengtai District government Office Automation environment, yielding LSASS and registry hive theft, 822 user records and 1.28GB of attachments including patient health data. Tooling included a GLUTTON webshell hiding payloads in PNG pixels via steganography and a fake MySQL deserialization service for client-side code execution.
72
MP-Bench: Evaluating Voice Agents as a Multiparty Conversation Participant
MP-Bench is the first benchmark for voice agents in multiparty conversations, finding real-time agents near chance on turn-taking.
MP-Bench is the first benchmark designed to objectively evaluate conversational speech systems as active participants in multi-party conversations. It assesses agents on turn-taking awareness and response appropriateness, with comprehension-based question-answering as a complementary evaluation. Benchmarking 12 voice agents shows real-time agents score at or below 22% on multiparty comprehension and remain near chance on multiparty turn-taking.
20
57
Making Knowledge Distillation Cheap Enough to Run at Scale
Hugging Face blog by Multiverse Computing describes techniques making knowledge distillation cheap enough for large-scale training.
A Hugging Face blog post from Multiverse Computing (CAI) presents methods for reducing the cost of knowledge distillation so it can be run at scale. The post is aimed at practitioners compressing large models into smaller, cheaper ones for production use.
30