Tag: LLM
-
Im Kontext der Datensicherheit sind LLMs als Menschen zu betrachten
Große Sprachmodelle und die Frage der Data Security Sicherheitsfragen rund um LLMs. Angesichts der rasanten KI-Entwicklung wird immer deutlicher, dass die grundlegenden Leitplanken, Plausibilitätsprüfungen und prompt-basierten Sicherheitsmaßnahmen, die derzeit gelten, durchlässig und unzureichend sind. Bei der Entwicklung von Strategien zur Verbesserung der Datensicherheit in KI-Workloads ist es entscheidend, die Perspektive zu ändern und… First seen…
-
New LLM jailbreak uses models’ evaluation skills against them
First seen on scworld.com Jump to article: www.scworld.com/news/new-llm-jailbreak-uses-models-evaluation-skills-against-them
-
‘Bad Likert Judge’ Jailbreak Bypasses Guardrails of OpenAI, Other Top LLMs
A novel technique to stump artificial intelligence (AI) text-based systems increases the likelihood of a successful cyberattack by 60%. First seen on darkreading.com Jump to article: www.darkreading.com/cyberattacks-data-breaches/bad-likert-judge-jailbreak-bypasses-guardrails-openai-other-llms
-
LLMs Crack the Code: 95% Success Rate in Hacking Challenge
A recent study demonstrates the transformative potential of large language models (LLMs) in offensive cybersecurity tasks. Researchers Rustem Turtayev, Artem Petrov, Dmitrii Volkov, and Denis Volk have achieved a record-breaking... First seen on securityonline.info Jump to article: securityonline.info/llms-crack-the-code-95-success-rate-in-hacking-challenge/
-
SECURITY AFFAIRS MALWARE NEWSLETTER ROUND 26
Security Affairs Malware newsletter includes a collection of the best articles and research on malware in the international landscape. Now You See Me, Now You Don’t: Using LLMs to Obfuscate Malicious JavaScript Analyzing Malicious Intent in Python Code: A Case Study DigiEver Fix That IoT Thing! Botnets Continue to Target Aging D-Link Vulnerabilities OtterCookie, […]…
-
AI Could Generate 10,000 Malware Variants, Evading Detection in 88% of Case
Cybersecurity researchers have found that it’s possible to use large language models (LLMs) to generate new variants of malicious JavaScript code at scale in a manner that can better evade detection.”Although LLMs struggle to create malware from scratch, criminals can easily use them to rewrite or obfuscate existing malware, making it harder to detect,” Palo…
-
Die 10 häufigsten LLM-Schwachstellen
Tags: access, ai, api, application-security, awareness, breach, cloud, control, cyberattack, data, detection, dos, encryption, injection, least-privilege, LLM, ml, monitoring, privacy, RedTeam, remote-code-execution, risk, service, tool, update, vulnerability, zero-trust -
Sophos stellt Tuning-Tool für große Sprachmodelle als Open-Source-Programm zur Verfügung
Große Sprachmodelle (Large-Language-Models, LLMs) haben das Potenzial, die Arbeitslast zu automatisieren und zu reduzieren, einschließlich der von Cybersicherheitsanalysten und Incident-Respondern. Generischen LLMs fehlt jedoch das domänenspezifische Wissen, um diese Aufgaben gut zu bewältigen. Auch wenn sie mit Trainingsdaten erstellt wurden, die Cybersicherheitsressourcen enthalten, reicht dies oft nicht aus, um spezialisiertere Aufgaben zu übernehmen, die aktuelles…
-
Sophos stellt neues Trainings-Framework zur Optimierung der LLMs zur Verfügung
Durch den Einsatz von DeepSpeed wird die Skalierung großer Trainingsaufgaben ermöglicht, unter anderem durch parallele Datenverarbeitung. First seen on infopoint-security.de Jump to article: www.infopoint-security.de/sophos-stellt-neues-trainingsframework-zur-optimierung-der-llms-zur-verfuegung/a39320/
-
SophosAI-Team stellt Open-Source-Tuning-Tool für LLMs bereit
Large-Language-Modelle (LLMs) haben das Potenzial, die Arbeitslast zu automatisieren und zu reduzieren, einschließlich der von Cybersicherheitsanalysten und Incident Respondern. Generischen LLMs fehlt jedoch das domänenspezifische Wissen, um diese Aufgaben gut zu bewältigen. Auch wenn sie mit Trainingsdaten erstellt wurden, die Cybersicherheitsressourcen enthalten, reicht dies oft nicht aus, um spezialisiertere Aufgaben zu übernehmen, die aktuelles und…
-
OWASP Top 10 Risk Mitigations for LLMs and Gen AI Apps 2025
The rapid advancement of AI, particularly in large language models (LLMs), has led to transformative capabilities in numerous industries. However, with great power comes significant security challenges. The OWASP Top… First seen on securityboulevard.com Jump to article: securityboulevard.com/2024/12/owasp-top-10-risk-mitigations-for-llms-and-gen-ai-apps-2025/
-
Platforms are the Problem
Tags: ai, breach, business, chatgpt, cloud, cyber, cybercrime, cybersecurity, data, defense, detection, finance, firewall, fraud, infrastructure, intelligence, LLM, network, saas, service, technology, threat, toolA better path forward for cybersecurity Why is it that cybersecurity is struggling to keep pace with the rapidly evolving threat landscape? We spend more and more, tighten our perimeters, and still there are trillions of dollars being lost to cybercrime and cyber attacks. Setting aside the direct costs to individuals and businesses, and the…
-
FuzzyAI: Open-source tool for automated LLM fuzzing
FuzzyAI is an open-source framework that helps organizations identify and address AI model vulnerabilities in cloud-hosted and in-house AI models, like guardrail bypassing and … First seen on helpnetsecurity.com Jump to article: www.helpnetsecurity.com/2024/12/13/fuzzyai-automated-llm-fuzzing/
-
AI Slop is Hurting Security, LLMs are Dumb and People are Dim
Artificial stupidity: Large language models are terrible if you need reasoning or actual understanding. First seen on securityboulevard.com Jump to article: securityboulevard.com/2024/12/ai-slop-security-reports-richixbw/
-
IT-Trends 2025 Video-Statements von Experten der Netzpalaver-Community
Alle Märkte der ITK-Branche sind im Umbruch. Grund: die künstliche Intelligenz. Optimierungen, Effizienz, Automatisierung, Einsparungen, Customer-Experience, Fachkräftemangel etc. LLMs, KI-Agenten und Embedded-Systeme revolutionieren die Datacenter, die Cybersicherheit, Unified-Communications, die Programmierung und sämtliche Aspekte der IT-Infrastruktur. Das Jahr 2025 steht vor etlichen Herausforderungen in Bezug auf die künstliche Intelligenz und vielen weiteren Trends. Welche das sind,…
-
The imperative for governments to leverage genAI in cyber defense
Tags: ai, attack, cyber, cyberattack, cybersecurity, dark-web, data, deep-fake, defense, detection, email, endpoint, gartner, government, incident response, infrastructure, intelligence, LLM, malicious, malware, microsoft, strategy, tactics, threat, tool, training, vulnerabilityIn an era where cyber threats are evolving at an unprecedented pace, the need for robust cyber defense mechanisms has never been more critical. Sixty-two percent of all cyberattacks focus on public sector organizations directly and indirectly. Nation-state actors, equipped with generative artificial intelligence (genAI) sophisticated tools and techniques, pose significant threats to national security,…
-
Black Hat: Latest news and insights
The infosecurity world decamps to London this week, with research on vulnerabilities in AI systems at the fore of the latest edition of Black Hat Europe.The four-day program runs from Dec. 9-12, with two-and four-day options of hands-on trainings, but the main event at ExCeL London occurs on Dec. 11 and 12 featuring the latest research, developments,…
-
Black Hat Europe preview: Cryptographic protocol attacks and AI in the spotlight
Tags: access, ai, application-security, attack, authentication, backdoor, best-practice, computer, conference, control, cybercrime, cybersecurity, data, dns, encryption, exploit, finance, github, government, hacker, healthcare, identity, injection, Internet, LLM, malicious, microsoft, mitigation, office, open-source, radius, RedTeam, risk, service, sophos, technology, tool, training, vulnerability, vulnerability-management, windowsThis week in London Black Hat Europe will feature a diverse range of talks and presentations covering the latest developments in cybersecurity.The opening keynote on Wednesday will be delivered by Frédérick Douzet, a professor of geopolitics at the University of Paris 8, and director of the French Institute of Geopolitics research team. No preview is…
-
Microsoft Challenge Will Test LLM Defenses Against Prompt Injections
Microsoft is calling out to researchers to participate in a competition that is aimed at testing the latest protections in LLMs against prompt injection attacks, which OWASP is calling the top security risk facing the AI models as the industry rolls into 2025. First seen on securityboulevard.com Jump to article: securityboulevard.com/2024/12/microsoft-challenge-will-test-llm-defenses-against-prompt-injections/
-
Microsoft Bets $10,000 on Prompt Injection Protections of LLM Email Client
Microsoft offers $10,000 in rewards to researchers who can manipulate a realistic simulated LLM-integrated email client. The post Microsoft Bets $10,000 on Prompt Injection Protections of LLM Email Client appeared first on SecurityWeek. First seen on securityweek.com Jump to article: www.securityweek.com/microsoft-bets-10000-on-prompt-injection-protections-of-llm-email-client/
-
Hackers Can Hijack Your Terminal Via Prompt Injection using LLM-powered Apps
Researchers have uncovered that Large Language Models (LLMs) can generate and manipulate ANSI escape codes, potentially creating new security vulnerabilities in terminal-based applications. ANSI escape sequences are a standardized set of control characters used by terminal emulators to manipulate the appearance and behavior of text displays. They enable features such as text color changes, cursor movement, blinking text, and more. Terminal emulators interpret these sequences…
-
>>Hack<< this LLM-powered service and get paid
Microsoft, in collaboration with the Institute of Science and Technology Australia and ETH Zurich, has announced the LLMail-Inject Challenge, a competition to test and improve … First seen on helpnetsecurity.com Jump to article: www.helpnetsecurity.com/2024/12/09/llm-prompt-injection-attacks-challenge/
-
CrowdStrike hilft bei der Sicherung des EndEnd-KI-Ökosystems, das auf AWS aufbaut
Die erweiterte Integration bietet End-to-End-Transparenz und Schutz für KI-Innovationen, von LLMs bis hin zu Anwendungen, durch verbesserte Amazon SageMaker-Unterstützung, KI-Container-Scanning und AWS IAM Identity Center-Integration. Da Unternehmen ihre Innovationen in der Cloud und die Einführung von KI beschleunigen, ist die Sicherung von KI-Workloads und -Identitäten von entscheidender Bedeutung. Fehlkonfigurationen, Schwachstellen und identitätsbasierte Bedrohungen setzen… First…
-
LLMs Raise Efficiency, Productivity of Cybersecurity Teams
AI-powered tools are making cybersecurity tasks easier to solve, as well as easier for the team to handle. First seen on darkreading.com Jump to article: www.darkreading.com/vulnerabilities-threats/llms-raise-efficiency-productivity-of-cybersecurity-teams