THE DAILY EDITION

Cyber security

8 items · 44 articles in this topic

Patching dominated: Java4Shell, Puppet, Apache Roller and Red Hat Quay all need action, and the first three are exploited or have a CISA advisory. Supply-chain and credential-theft stories ran alongside them, along with hardware and AI-agent attack research.

  1. 1

    Apache released updates for a critical Roller RCE flaw that threat actors are already exploiting for code execution. Roller operators should update now.

    HighBleepingComputer ↗2 min read
  2. 2

    A critical Red Hat Quay flaw lets attackers read and overwrite all private container images in a registry. Overwriting images makes it a supply-chain risk for anything pulling from that registry.

    HighLobsters ↗5 min read
  3. 3

    Cloud credentials exposed in public Docker Hub images are harvested by attackers within eight seconds of publication. Revoking a leaked key after the fact is effectively too late.

    HighBleepingComputer ↗3 min read
  4. 4

    TU Berlin researchers show that opening the PC case can reset the TPM lockout. That allows fast guessing of Windows Hello and BitLocker PINs on AMD and some Intel PCs, in days rather than years.

    HighThe Hacker News ↗9 min read
  5. 5

    Russia-linked Cold Relic (COLDRIVER) used data from a stolen Google Workspace compromise to breach a second US organization, per Google GTIG.

    HighThe Hacker News ↗5 min read
  6. 6

    Researchers took control of the Hugging Face account hosting the popular Kokoro text-to-speech model because of weak ownership verification. Separately, the Trigram campaign impersonates about 50 AI and blockchain brands to push infostealers at developers, per Stormshield.

    HighBleepingComputer ↗3 min read
  7. 7

    Japan's RIKEN Center took the Fugaku supercomputer offline after a security breach. Separately, Tor released an emergency update while responding to an ongoing DDoS attack on its network.

    HighBleepingComputer ↗2 min read
  8. 8

    A study extracted system-prompt secrets from 296 of 363 OpenRouter-hosted LLMs using persuasion techniques in an agentic capture-the-flag setup. Separately, NebuSec demonstrated two RCE exploits via malicious bot widgets that bypass Telegram's new Android filter.

More in Cyber security · 36 articles›
  • CISA adds actively exploited critical Perforce Puppet RCE flaw CVE-2026-115731 (CVSS 9.4) to its KEV catalog.

    The Hacker Newscvepuppetkevactively-exploitedrce
    Feed excerpt

    The U.S. Cybersecurity and Infrastructure Security Agency (CISA) on Tuesday added a critical security flaw impacting Perforce Puppet, an enterprise infrastructure-as-code (IaC) and configuration management platform, to its Known Exploited Vulnerabilities (KEV) catalog, citing reports of active exploitation. The vulnerability, tracked as CVE-2026-115731 (CVSS score: 9.4), is a case of

  • CISA urges immediate patching of critical Oracle Java flaw CVE-2026-114938 (CVSS 8.1) enabling DoS or remote code execution.

    The Hacker Newscvejavacisarce
    Feed excerpt

    The U.S. Cybersecurity and Infrastructure Security Agency (CISA) has urged organizations to immediately patch a newly disclosed critical security flaw impacting Oracle Java. The flaw, tracked as CVE-2026-114938 (CVSS score: 8.1), refers to a case of out-of-bounds write that can lead to denial-of-service (DoS) or remote code execution under certain conditions. "This critical memory-corruption

  • Attackers published malicious versions 1.26.1 and 1.26.2 of Microsoft's agent-framework PyPI package carrying a credential stealer that runs on every Python start.

    The Hacker Newssupply-chainpypicredential-theftmicrosoft
    Feed excerpt

    Threat actors have compromised Microsoft's agent-framework package to push a credential stealer, according to reports from Aikido, Chainguard, Endor Labs, OX Security, SafeDep, Socket, StepSecurity, and Google-owned Wiz. The malicious versions, 1.26.1 and 1.26.2, were published 33 minutes apart on November 17, 2026, and are designed to run a credential stealer on every Python start. As of

  • Java4Shell, a Java flaw reported by Calif, enables unauthenticated remote code execution in Jira, Jenkins, and Keycloak on AVX-512 VBMI servers.

    The Hacker Newscvercejavajirajenkins
    Feed excerpt

    A Java flaw called Java4Shell can let attackers without an account run their own code on Jira, Jenkins, Keycloak, and three other widely used applications, says Calif, the security research firm that disclosed it this week. The flaw is present in Java 17.0.6 and later, and the vulnerable code runs only on x86-64 Intel and AMD chips that support the AVX-512 VBMI feature. Tracked as

  • Attackers exploited an Optimizely vulnerability to inject malware into about 178 top websites, including BBC, CNN, and Zoom.

    Krebs on Securitysupply-chainmalwareweb-securitythird-party-script
    Feed excerpt

    At least 178 top web destinations unwittingly served a variety of malware exploits against visitors over the past 24 hours thanks to a security vulnerability involving Optimizely, a company that helps websites customize user experiences and test out new features. Experts say the incident, which saw data-stealing malicious software silently injected into sites like the BBC, CNN, Equifax, Ledger and Zoom, was exacerbated by the reality that few popular websites enforce strict rules on the sources of third-party content.

  • A critical flaw dubbed Java4Shell in OpenJDK allows unauthenticated remote code execution in Jenkins, Jira, Confluence, and Keycloak.

    BleepingComputercvercejavaopenjdk
    Feed excerpt

    A critical vulnerability dubbed Java4Shell in OpenJDK, the open-source implementation of Java, can allow unauthenticated attackers to remotely execute code on popular enterprise applications, including Jenkins, Atlassian Jira, Confluence, and Keycloak. [...]

  • Researchers disclose Java4Shell, a remote code execution flaw in OpenJDK's Base64 processing that can crash the JVM.

    Hacker News (front page)cvejavarceopenjdk
    Feed excerpt

    Article URL: https://calif.io/java4shell/ Comments URL: https://news.ycombinator.com/item?id=50538240 Points: 399 # Comments: 106

  • U.S. intelligence reportedly assesses with low confidence that Iran was responsible for the attacks on French fridges.

    Schneier on Securityattributioniraniotnation-state
    Feed excerpt

    Interesting: U.S. intelligence agencies have recently reached a “”ow confidence” assessment that Iran was responsible for the attacks in France, according to three U.S. officials who spoke on the condition of anonymity to discuss sensitive intelligence. A “low confidence” assessment generally means that there is some intelligence that points in a certain direction but is fragmented, and that the assessment could change as more information is collected. The agencies’ judgment on the French incidents linked those attacks to Iran generally and not to a specific group such as the Revolutionary Guards, two of the officials said. They added that Iran was also seen as probably responsible for other similar attacks around Europe, but that those investigations were in their early stages. Alternate link.

  • Researchers show an unauthenticated network attacker can replace the cloud LLM system prompt and personality on a Pollen Reachy Mini Lite robot.

    arXiv cs.CR (Cryptography & Security)iotprompt-injectionroboticsconfiguration
    Feed excerpt

    arXiv:2611.22761v1 Announce Type: new Abstract: Large language models (LLMs) in embodied robots create security dependencies beyond inference. We investigate a network-exposed configuration attack on the Pollen Robotics Reachy Mini Lite, allowing an unauthenticated attacker to replace personality settings and the cloud LLM's system prompt through a management interface, without adversarial conversational input. Eleven prompt-injection attempts across five categories were reported as resisted, whereas both configuration-layer attack paths produced observable behavioural changes on physical hardware. We demonstrate local and network-adjacent access, mapping the weakness to CWE-306 (Missing Authentication for a Critical Function) and CWE-1327 (Binding to an Unrestricted IP Address), with a reported CVSS v3.1 score of 6.5 (Medium). A configuration endpoint also accepts cross-origin requests without a CORS preflight (CWE-352; 6.5 Medium), while the persistent-memory sink stores content without enforcing its stated restrictions. A source review of 59 public repositories identifies missing authentication in 10 of 11 source-verifiable genuine applications, or 11 of 11 including a framework…

  • Hundreds of websites are serving a malware bundle that delivers cryptocurrency theft payloads and iOS exploits to visitors.

    Ars Technicamalwarewebios-exploit
    Feed excerpt

    Malware cocktail delivers cryptocurrency theft payloads and iOS exploits, among other things.

  • NebuSec researchers demonstrated two RCE exploits using malicious bot widgets that bypass Telegram's new Android security filter.

    The Hacker Newsrcetelegramandroidexploit
    Feed excerpt

    A malicious widget can bypass the security filter that Telegram added to its Android app last week and run code outside it, according to NebuSec, a security company. Widgets are small interactive blocks that bots send inside chat messages, and their JavaScript runs on the user's phone. The researchers ran their demonstration on a Telegram test server, because only approved bots can send

  • The Tor Project released an emergency update while responding to an ongoing DDoS attack on its network.

    BleepingComputertorddospatch
    Feed excerpt

    [...]

  • A supply chain attack on Optimizely delivered malware to websites of CNN, Fox News, BBC, and Salesforce.

    Hacker News (front page)supply-chainmalwareweb
    Feed excerpt

    Article URL: https://www.bleepingcomputer.com/news/security/optimizely-supply-chain-attack-delivers-malware-to-cnn-fox-news-bbc-salesforce-sites/ Comments URL: https://news.ycombinator.com/item?id=50534664 Points: 43 # Comments: 6

  • Uber says attackers used stolen credentials to access an AI vendor and steal driver's license data.

    The Registerbreachcredentialsuber
    Feed excerpt

    Ride-hailing biz says stolen credentials allowed crooks to help themselves to driver's license data

  • Oracle will patch 11 remotely exploitable Java SE flaws with max CVSS 8.1, as reports warn of a possible critical OpenJDK RCE.

    The Hacker Newscvejavapatch
    Feed excerpt

    Oracle is preparing to release security patches for 11 vulnerabilities in Java SE on Tuesday, November 17, 2026, all of which may be exploited remotely without authentication, as separate reports warn of a potentially critical remote code execution (RCE) flaw affecting OpenJDK. The 11 vulnerabilities carry a maximum CVSS 3.1 severity score of 8.1, according to Oracle's advance security notice.

  • The Trigram campaign impersonates about 50 AI and blockchain brands to trick developers into running infostealer malware, per Stormshield.

    The Hacker Newsinfostealermalwaresupply-chain
    Feed excerpt

    Artificial intelligence (AI), blockchain, and logistics sectors across Europe and the U.S. have emerged as the target of a large-scale, industrialized campaign dubbed Trigram. "Trigram is a large-scale campaign that impersonates about 50 brands in the blockchain and AI sectors, to trick developers into executing an infostealer themselves," Stormshield said in a deep-dive technical report

  • Alliance for Creativity and Entertainment announces shutdown of Dizigom, a major Turkish piracy streaming site, via settlement.

    BleepingComputerpiracyenforcement
    Feed excerpt

    The Alliance for Creativity and Entertainment (ACE) has announced the shutdown of Dizigom, one of Turkey's largest piracy sites, after a settlement agreement with the site's operators. [...]

  • A UK member of the Scattered Spider group was sentenced in the US to six years in prison for crypto theft and hacking a dozen companies.

    BleepingComputerscattered-spidersentencingcybercrime
    Feed excerpt

    A leader of the Scattered Spider cybercrime collective has been sentenced to six years in prison in the United States for stealing millions in cryptocurrency and conspiring to hack a dozen companies. [...]

  • A paper surveys security risks of vibe coding, where AI agents autonomously write, test, and deploy software.

    arXiv cs.CR (Cryptography & Security)ai-agentsvibe-codingthreat-model
    Feed excerpt

    arXiv:2611.21822v1 Announce Type: new Abstract: Vibe coding is reshaping software development, enabling faster implementation and allowing individuals with limited programming experience to describe requirements in natural language while AI agents write, test, and deploy the code. As these agents evolve into autonomous systems that execute commands, manage dependencies, access cloud resources, and interact with other agents, their security risks extend beyond incorrect code to actions with direct consequences for software systems. Although research on these risks is expanding, it remains fragmented across areas including prompt injection, insecure code generation, supply-chain attacks, and human factors, with limited consensus on terminology, threat assumptions, and evaluation. This paper consolidates emerging research through the lens of the vibe coding workflow. We establish a reference architecture, characterise six levels of agent autonomy, and develop a threat model centred on five trust boundaries spanning developers, AI models, tools, cloud infrastructure, and the wider software ecosystem. Building on this foundation, we propose a taxonomy of twenty-one threats across nine c…

  • GapLeak paper shows LLM web agents can escalate to unauthorized sources and leak sensitive information when a site requests unavailable fields.

    arXiv cs.CR (Cryptography & Security)llm-agentsprivacyvulnerability
    Feed excerpt

    arXiv:2611.22223v1 Announce Type: new Abstract: Large Language Model (LLM)-based web agents rely on data minimization and least privilege to protect user privacy. We show that these principles are necessary but insufficient: they constrain authorized sources, but not higher-sensitivity sources that remain reachable. A website can request a field unavailable from authorized sources, creating an information gap. To complete the task, a goal-driven agent may escalate to reachable but unauthorized sources and disclose accumulated sensitive information. We formalize this vulnerability as GapLeak, where an adversary exploits a legitimate web form without prompt injection or other malicious instructions. We evaluate GapLeak on three web-agent frameworks and six LLMs from Anthropic, Google, and OpenAI across recruitment, e-commerce, travel, and healthcare. GapLeak induces an average unauthorized disclosure of approximately 37%. The information gap is the primary driver: eliminating it reduces disclosure by more than 50%. Existing instruction-level defenses based on least privilege partially mitigate the vulnerability but often reduce task success. We propose submission-time verification, w…

  • VTBA proposes a black-box jailbreak for multimodal LLMs that exploits visual trust bias using naturally occurring harmful image-text pairs.

    arXiv cs.CR (Cryptography & Security)jailbreakmllmprompt-injection
    Feed excerpt

    arXiv:2611.22292v1 Announce Type: new Abstract: Multimodal large language models (MLLMs) are rapidly advancing, yet their safety remains a critical concern, especially under jailbreak attempts. Existing MLLM jailbreaks either inject fragile adversarial perturbations into irrelevant images or synthesize toxic visual cues to achieve text-driven attack intents. In contrast, a more practical MLLM jailbreak emphasizes enhancing attack efficacy directly on naturally occurring harmful image-text pairs. We address this problem by proposing Visual Trust Bias Attack (VTBA), a new jailbreak framework that exploits a previously overlooked attack surface: MLLMs exhibit an intrinsic trust bias toward authoritative visual features, which enables transferable and stealthy jailbreaks. VTBA constructs a bank of authoritative templates, matches each input pair to the most relevant one, and integrates the harmful intent into the selected layout, thereby leveraging credible visual cues to override safety refusals while preserving naturalness. We further construct a dataset of 2,500 naturally occurring image-text pairs to evaluate jailbreaks in this practical multimodal setting. Extensive experiments sh…

  • Analysis of 3,758 vulnerability IDs finds 98% have disclosure dates but only 15.7% have patch evidence, using Vulnerability-Lookup.

    arXiv cs.CR (Cryptography & Security)cvevulnerability-managementkev
    Feed excerpt

    arXiv:2611.22479v1 Announce Type: new Abstract: -05-X-03), Known Exploited Vulnerabilities (KEV) catalogs, and exploitation sightings. We introduce an evidence model implemented in Vulnerability-Lookup that selects the earliest evidence for each lifecycle event, distinguishes vulnerability-specific records, preserves date precision, and ensures traceability to original sources. Among 3,758 vulnerability identifiers with exploitation evidence, 98% have disclosure dates, while only 15.7% have patch evidence. Vendor-side dates required by VTEM are entirely absent. Most exploitation dates originate from catalog listings, including 1,041 entries associated with bulk catalog imports. After excluding these entries and ambiguous same-day observations, 59 vulnerabilities exhibit patch evidence following exploitation evidence, with a median interval of 7.6 days, predominantly between catalog listings and distributor fixes. Evidence traceability also exposes systematic measurement errors: one collector incorrectly classified exploit publications as active exploitation, accounting for 40% of initial exploitation dates, while another catalog assigned vulnerability publication timestamps to 11% …

  • Evolution Under Fire co-evolves agent safety harnesses and adaptive injection stressors to defend tool-using agents against prompt injection.

    arXiv cs.CR (Cryptography & Security)prompt-injectionagentsdefense
    Feed excerpt

    arXiv:2611.22583v1 Announce Type: new Abstract: Tool-using language-model agents must act on information from untrusted emails, webpages, and documents. This creates a central security challenge: malicious instructions and legitimate task directives can arrive through the same interaction channel. Existing defenses are largely hand-designed and static, making further improvements dependent on repeated manual redesign. We introduce Evolution Under Fire, a staged framework that couples safety-harness evolution with adaptive stressor evolution. Utility-constrained, Pareto-scheduled tree search evolves complete executable safety harnesses from execution evidence while keeping the task model, benchmark, tools, and evaluator fixed. Adaptive stressor evolution then discovers utility-preserving attacks that guide a subsequent stage of harness evolution. On AgentDyn with GPT-4o-mini, an evolved harness reduces attack success from 50.4% to 3.0%, improves clean and attacked utility over the initial harness, and offers a stronger security-utility trade-off than reproduced defenses. On Agent-SafetyBench with Qwen2.5-7B-Instruct, harness evolution reduces attack success from 82.6% to 62.0% while…

  • Re-evaluation shows published adversarial face-privacy methods lose much of their claimed protection when a realistic detection and alignment front end is used.

    arXiv cs.CR (Cryptography & Security)adversarialfacial-recognitionprivacy
    Feed excerpt

    arXiv:2611.21923v1 Announce Type: cross Abstract: Published face-privacy methods report protection above 90% against black-box recognition models. That number comes from an evaluation protocol with two independent defects, reported in a measure that cannot size either one. The input is wrong: methods published since 2022 feed the network the whole photograph, or a detector's box, where deployed systems detect, align to a landmark template and crop. Re-scoring the same files through a realistic front end costs a standard transfer attack 56% of its dodging protection and a budget-allocating attack 7%, with no image regenerated. The measure is wrong: that loss is a difference of success rates, so it saturates, moving by 8.2x over a budget sweep while the effect beneath it moves by 1.34x. We therefore report the paired cosine shift, in which both thresholds cancel algebraically, so the input defect holds without committing to a calibration. Measured that way it is 2.2x and 10.7x larger on three unseen recognizers, and at matched budget it reverses the L-infinity transfer ranking: the one surveyed method that aligns at evaluation gains 0.18 where MI-FGSM loses 0.05. The cause is a resam…

  • GuardAgent introduces a memory-augmented agent framework to detect adversarial promotions manipulating LLM-based recommender systems.

    arXiv cs.CR (Cryptography & Security)llmrecommendationprompt-injection
    Feed excerpt

    arXiv:2611.22274v1 Announce Type: cross Abstract: In traditional recommender systems (RSs), textual information typically serves as auxiliary signals for representation learning. In contrast, in LLM-based RSs, textual information becomes the primary medium for recommendation generation through semantic understanding and reasoning. This shift fundamentally changes the attack surface, making promotion-driven manipulations more feasible, effective, and harder to detect. However, effectively detecting such attacks remains largely unexplored. To fill this gap, we present the first systematic study on detecting promotion-driven attacks against LLM-based RSs, aiming to safeguard next-generation recommendation platforms. Specifically, we propose GuardAgent, a novel LLM-powered agentic detection framework that operates through a loop of Perception, Action, Reflection and Memory. In contrast to conventional LLM-based detectors that function in a static, one-off, and opaque fashion, GuardAgent continually learns and self-evolves to adapt to unseen threat patterns while generating explainable reasoning traces and outcomes. Extensive experiments on real-world datasets demonstrate that GuardAgen…

  • ConfGuard proposes a physics-grounded reference monitor defending LLM predictive-maintenance agents against prompt injection and sensor perturbation attacks.

    arXiv cs.CR (Cryptography & Security)llm-agentsprompt-injectionindustrial
    Feed excerpt

    arXiv:2611.22288v1 Announce Type: cross Abstract: Large-language-model agents now close the industrial predictive-maintenance loop, reading sensor windows, retrieving procedures and dispatching work orders. Two attack channels meet there: injected text in a retrieved document can hijack the decision, and a bounded sensor perturbation can corrupt the remaining-useful-life estimate behind it. Recent agent defenses police data provenance and tool privilege but cannot tell whether a prognostic quantity is physically plausible. ConfGuard is a reference monitor with a physics-grounded policy. Its conformal monitor tests each prediction, anchored to the asset's admitted history, against degradation physics: capped remaining life never rises and falls at most one cycle per cycle. Calibrating its threshold over whole asset trajectories lets the operator set the expected share of a clean asset's windows the running monitor escalates. A physics-policy validator checks the agent's maintenance-urgency decision against the decision a deterministic policy derives from the same interval. On all 10,196 C-MAPSS test windows the monitor cuts a covert attacker's reach on healthy assets from 0.529 to 0…

  • Researchers reverse-engineer GPU load/store units and expose security risks from contention and timing side channels.

    arXiv cs.CR (Cryptography & Security)gpuside-channelhardware
    Feed excerpt

    arXiv:2611.23022v1 Announce Type: cross Abstract: GPUs are the main platform for large-scale machine learning and power the largest supercomputers. Their load/store units (LSUs) often determine kernel performance and have been exploited in prior attacks. Yet what is known about the LSU is only a few isolated mechanisms and parameters, and even a single one has been enough for an attack. In particular, it is still unknown what a load costs, where loads contend, and even whether a warp can starve due to contention. Until each is measured, neither its security risk nor a defense is known. We reverse-engineer how the LSU serves load instructions on recent NVIDIA GPUs with microbenchmarks and analyze the security implications. We give a cycle-accurate latency formula for a global L1 hit or shared memory access from the addresses it touches, whose address-dependent terms are bank conflicts and the L1 tag lookup; we show that loads contend at both, and that when the LSU pipeline is full a warp can starve a warp of a later-launched block. We then show that the LSU's attack surface is far broader than prior attacks revealed: through channels never exploited before, it can leak information, …

  • Introduces PSM, a black-box framework using LLM-guided optimization to harden system prompts against extraction attacks.

    arXiv cs.CR (Cryptography & Security)prompt-extractionllmdefense
    Feed excerpt

    arXiv:2511.16209v3 Announce Type: replace Abstract: System prompts are critical for guiding the behavior of Large Language Models (LLMs), yet they often contain proprietary logic or sensitive information, making them a prime target for extraction attacks. Adversarial queries can successfully elicit these hidden instructions, posing significant security and privacy risks. Existing defense mechanisms frequently rely on heuristics, incur substantial computational overhead, or are inapplicable to models accessed via black-box APIs. This paper introduces a novel framework for hardening system prompts through shield appending, a lightweight approach that adds a protective textual layer to the original prompt. Our core contribution is the formalization of prompt hardening as a utility-constrained optimization problem. We leverage an LLM-as-optimizer to search the space of possible SHIELDs, seeking to minimize a leakage metric derived from a suite of adversarial attacks, while simultaneously preserving task utility above a specified threshold, measured by semantic fidelity to baseline outputs. This black-box, optimization-driven methodology is lightweight and practical, requiring only API …

  • Proposes MACE, an MCTS-guided adversarial co-evolution method that improves LLM safety alignment against jailbreak attackers.

    arXiv cs.CR (Cryptography & Security)jailbreakllmalignment
    Feed excerpt

    arXiv:2511.19218v3 Announce Type: replace Abstract: Large Language Models (LLMs) increasingly power web-facing applications, including conversational assistants and content services, where open user interaction exposes them to diverse safety threats and adaptive attacks. In particular, jailbreak attackers repeatedly reformulate harmful requests to bypass safety alignment. Although existing studies have concentrated on defending dynamic jailbreaks, they often retain discovered attacks as flat samples and discard information about how those attacks were found. In this paper, we introduce MACE, an MCTS-guided adversarial co-evolution framework that turns search into policy-learning signals for both an attacker and a defender. Its search component, GS-MCTS, explores multi-step compositions of rewriting operators and associates discovered attacks with empirical success frequency and path depth. Its optimization component, AC-TGPO, uses these statistics for difficulty-aware group-relative credit assignment and constructs a temporally asymmetric curriculum from current interactions, pairings between the latest attacker and preceding defender, and attacks that survive a defense update. Thu…

  • Describes a two-stage post-training recipe turning a small local model into a Linux privilege escalation security agent.

    arXiv cs.CR (Cryptography & Security)privilege-escalationlinuxllm-agent
    Feed excerpt

    arXiv:2603.17673v3 Announce Type: replace Abstract: LLM agents are increasingly important in security, yet leading systems are typically closed-source and cloud-hosted, limiting reproducibility and their use with sensitive data. This creates a need for small, local models that can perform security tasks under strict resource constraints, though effective methods for developing them remain underexplored. In this paper, we address this gap by proposing a two-stage post-training recipe that turns a small local language model into a security agent. To this end, we focus on Linux privilege escalation as a representative setting to systematically study the training of local models, as the task is automatically verifiable and requires multi-step interactive reasoning. Using an experimental setup that mitigates benchmark leakage, we post-train a small 4B model in two stages: supervised fine-tuning on traces from procedural privilege-escalation environments, followed by reinforcement learning with verifiable rewards. On a held-out benchmark of 12 Linux privilege-escalation scenarios, supervised fine-tuning nearly doubles the baseline success rate under a tight budget of 20 interaction round…

  • Argues that LLM misuse safeguards rely on untrusted user-claimed identities, creating a widespread high-risk bypass vulnerability.

    arXiv cs.CR (Cryptography & Security)llmsafeguardsjailbreak
    Feed excerpt

    arXiv:2607.27951v2 Announce Type: replace Abstract: Large language models now help professionals in fields such as cybersecurity and biology, yet the same answers would help an attacker. Deployers rely on misuse safeguards to keep such answers from attackers. We find that these safeguards share a high-risk vulnerability: they decide whether to release an answer from what users write, including the professional identity or purpose they claim, and anyone can write such a claim. Threat reports from model vendors and security firms show that attackers who fabricate a professional identity or purpose already exploit this vulnerability to cause serious harm. The vulnerability is also widespread across the state of the art: every defense proposed in 129 papers from the latest editions of 12 top venues has it. Safeguard research studies the wrong object: it treats harm as a property of the text of a request, although the same text can come from a professional or an attacker. Current safeguard benchmarks therefore cannot reveal the vulnerability: a safeguard can score high on them and still give dangerous answers to attackers. Because the vulnerability lies in what users write, no improveme…

  • Finds AI vulnerability-analysis agents produce dramatically different results on functionally identical code after natural code transformations.

    arXiv cs.CR (Cryptography & Security)ai-agentsvulnerability-analysisllm
    Feed excerpt

    arXiv:2610.14350v3 Announce Type: replace Abstract: Having demonstrated a strong potential for automating vulnerability discovery, AI agents can also be used to attack software systems at scale. Today's software systems need mechanisms to level the playing field by slowing down potential attacks, giving them time to patch vulnerabilities and strengthen software. Along this line, our study of agent trajectories in vulnerability analysis shows a surprising finding: agents produce dramatically different results across real-world code commits that carry the same functionality and the same vulnerability. This phenomenon appears consistently in multiple code domains and vulnerability categories. It motivates us to study natural and functionality-preserving code transformations derived from real-world software development, and their potential as a defense against attacker agents. We build a corpus of natural transformations by mining recurring development patterns from real-world code commits, and apply them to evaluate impact on agentic vulnerability analysis. Our results show that even a single form of transformation can effectively "confuse" frontier agents, e.g., Claude Mythos 5.2 wit…

  • OverServe is a multi-agent framework showing malicious SaaS tenants can prompt agents to extend execution and waste provider resources.

    arXiv cs.CR (Cryptography & Security)llmagentssaasabuse
    Feed excerpt

    arXiv:2611.14808v2 Announce Type: replace Abstract: Advances in large language models (LLMs) are enabling software-as-a-service (SaaS) providers to automate longer business workflows with agents. In subscription-based SaaS deployments, model and tool calls consume provider resources without additional customer payments. Because the agent autonomously decides when to stop, a malicious tenant can send a prompt to the agent, exploiting its completion decisions to extend execution and waste provider resources. In this paper, we present OverServe, a multi-agent framework for constructing such attacks against SaaS agents. OverServe co-designs Model Context Protocol (MCP) status tools and completion-confirmation prompts for the attacker. Each tool call returns accurate status and terminates normally, yet the agent may repeatedly check an unfinished condition to satisfy the prompt. This consumes the SaaS provider's resources even after the original business task is completed. OverServe constructs service-specific tool--prompt pairs and uses separate tool and prompt reviews to guide candidate pair selection. Tested tool specifications trigger no alerts in the evaluated MCP scanners. We eval…

  • ProcReverser is a multi-agent system that automates black-box microarchitectural reverse engineering to aid processor vulnerability analysis.

    arXiv cs.CR (Cryptography & Security)hardware-securitymicroarchitecturellm-agents
    Feed excerpt

    arXiv:2611.20827v2 Announce Type: replace Abstract: When processor implementation details are unavailable, black-box microarchitectural reverse engineering helps security researchers identify hardware conditions related to security vulnerabilities. However, this process requires substantial manual effort to design experiments, develop test programs, and analyze results for each processor. We present ProcReverser, a multi-agent system that automates this process. ProcReverser uses execution feedback to improve experimental plans and test programs and identifies microarchitectural properties from measurements. We develop a comprehensive suite of property identification tasks focused on hardware mechanisms relevant to Spectre-class attacks, covering speculative execution, branch prediction, memory dependence, and cache behavior. We evaluate ProcReverser on the XuanTie C910, the gem5 RISC-V O3 CPU model, and the SiFive P550. ProcReverser achieves Pass@1 and Pass@5 scores of 81.3% and 93.8%, respectively, on C910, and 83.0% and 100.0% on gem5 O3. On the SiFive P550, ProcReverser further characterizes microarchitectural properties covered by the proposed task suite. These findings highli…

  • Phrack released issue 73, its latest edition of the long-running hacking and security zine.

    Hacker News (front page)zinehacking
    Feed excerpt

    Article URL: https://phrack.org/issues/73/introduction_md Comments URL: https://news.ycombinator.com/item?id=50535992 Points: 121 # Comments: 6

  • Vals testing found leading AI models could be manipulated into misidentifying civilians as military targets when told it is a game.

    TechCrunchllmjailbreakevaluation
    Feed excerpt

    Vals, a San Francisco-based AI evaluation startup, tested whether leading AI models could be manipulated into participating in autonomous warfare. They usually compiled.

Written by claude-sonnet-5-5 from feed titles and summaries. The model can misstate details; every claim links to its source.