Tag: LLM
-
Critical Copilot vulnerability allowed hackers to steal 2FA code from users
SearchLeak exploit shows why the industry’s approach to LLM security fails over and over. First seen on arstechnica.com Jump to article: arstechnica.com/security/2026/06/critical-copilot-vulnerability-allowed-hackers-to-seal-2fa-code-from-users/
-
Critical Copilot vulnerability allowed hackers to seal 2FA code from users
SearchLeak exploit shows why the industry’s approach to LLM security fails over and over. First seen on arstechnica.com Jump to article: arstechnica.com/security/2026/06/critical-copilot-vulnerability-allowed-hackers-to-seal-2fa-code-from-users/
-
KI-Sicherheit im Unternehmen: Warum LLMs, KI-Pipelines und Agenten neue Schutzmechanismen brauchen
In vielen Unternehmen läuft generative KI inzwischen im Hintergrund mit, als Chatbot, als Wissensassistent, im Code-Editor und immer öfter als Agent, der eigenständig Systeme anspricht und Daten abruft. Mit jeder dieser Integrationen wächst eine Angriffsfläche, die vorhandene Sicherheitswerkzeuge allein nicht ausreichend abdecken. Firewall, Authentifizierung und Eingabefilter bleiben notwendig, für LLM-, RAG- und agentenbasierte […] First…
-
Openclaw durch Prompt-Injection in Nachrichtenobjekten angreifbar
Die Sicherheitsforscher von Thales haben Sicherheitslücken auf Basis von Prompt-Injection bei Openclaw gefunden. Diese Schwachstellen wurden dem Openclaw-Sicherheitsteam verantwortungsbewusst gemeldet und mit Version 2026.4.23 wurde ein entsprechender Fix bereitgestellt. Dennoch bleiben zwei Herausforderungen bestehen: Prompt-Injection ist ein branchenweit weitgehend ungelöstes Problem. Es gibt keinen Standard, der regelt, wie Messaging-Objekte serialisiert werden, bevor sie ein LLM…
-
Warum LLMs, KI-Pipelines und Agenten neue Schutzmechanismen brauchen
In vielen Unternehmen läuft generative KI inzwischen im Hintergrund mit, als Chatbot, als Wissensassistent, im Code-Editor und immer öfter als Agent, der eigenständig Systeme anspricht und Daten abruft. Mit jeder dieser Integrationen wächst eine Angriffsfläche, die vorhandene Sicherheitswerkzeuge allein nicht ausreichend abdecken. Firewall, Authentifizierung und Eingabefilter bleiben notwendig, für LLM-, RAG- und agentenbasierte […] First…
-
Understanding LLM Coding Personalities Is Now Key to Developer Improvement
Organizations must understand AI coding models’ strengths, weaknesses, and security blind spots to reduce risk. First seen on esecurityplanet.com Jump to article: www.esecurityplanet.com/threats/understanding-llm-coding-personalities-is-now-key-to-developer-improvement/
-
How to defend at machine speed: A post-LLM era playbook
Tags: LLMFirst seen on scworld.com Jump to article: www.scworld.com/native/how-to-defend-at-machine-speed-a-post-llm-era-playbook
-
AI red teaming agents change how LLMs get tested
Adversarial probing of LLMs has piled up a sprawling toolkit over the past three years. Attack techniques with names like Tree of Attacks with Pruning, Crescendo, and Skeleton … First seen on helpnetsecurity.com Jump to article: www.helpnetsecurity.com/2026/05/21/ai-red-teaming-agents-research/
-
The AI backdoor your security stack is not built to see
Enterprises deploying LLMs have spent the past two years building defenses around a reasonable assumption: malicious behavior leaves a trace in the input. Scan for suspicious … First seen on helpnetsecurity.com Jump to article: www.helpnetsecurity.com/2026/05/18/metabackdoor-llm-backdoor-attack/
-
The AI backdoor your security stack is not built to see
Enterprises deploying LLMs have spent the past two years building defenses around a reasonable assumption: malicious behavior leaves a trace in the input. Scan for suspicious … First seen on helpnetsecurity.com Jump to article: www.helpnetsecurity.com/2026/05/18/metabackdoor-llm-backdoor-attack/
-
Okta Studie zeigt: Eingebaute KI-Guardrails reichen nicht aus
In einem Szenario genügte bereits ein simples Webformular auf einer präparierten Website, um ein unzensiertes LLM dazu zu bringen, seinen kompletten Credential-Store offenzulegen. First seen on infopoint-security.de Jump to article: www.infopoint-security.de/okta-studie-zeigt-eingebaute-ki-guardrails-reichen-nicht-aus/a45091/
-
Linux kernel maintainers suggest a ‘kill switch’ to protect systems until a zero-day vulnerability is patched
Tags: access, attack, business, control, cve, cybersecurity, exploit, flaw, group, incident response, infosec, linux, LLM, mitigation, risk, service, strategy, switch, technology, tool, update, vulnerability, zero-day), a logic bug which lets users easily obtain root access, and Dirty Frag, which abuses weaknesses in how the Linux kernel handles fragmented memory pages. The Dirty Frag attack combines two separate vulnerabilities affecting the Linux IPsec Encapsulating Security Payload (ESP) subsystem (CVE-2026-43284) and the RxRPC networking protocol (CVE-2026-43500). The proposal has set off a furious…
-
AI Researchers Target SIEM Migration Bottleneck
System Translates Detection Rules Across Security Platforms. Researchers developed an AI framework that converts threat detection rules between major SIEM platforms including Splunk, Microsoft Sentinel and QRadar. The system uses LLMs and automated validation steps to preserve detection logic during migrations that often require months of manual work. First seen on govinfosecurity.com Jump to article:…
-
AI security is repeating endpoint security’s biggest mistake
Tags: access, ai, api, automation, business, control, data, detection, edr, endpoint, governance, incident response, injection, LLM, monitoring, open-source, radius, risk, saas, sbom, soc, strategy, technology, threat, tool, updateMost AI security is still at the posture phase: Look at where most organizations are with AI security today. Model cards, AI-specific SBOMs, input and output filters, prompt injection guardrails and access controls around model APIs. These are valuable controls, but they reflect a posture-based approach. To truly enhance security, organizations must recognize the importance…
-
1,800+ MCP servers exposed without authentication: How zero trust can secure the AI agent revolution
Tags: ai, attack, authentication, breach, cloud, control, credentials, data, data-breach, defense, exploit, framework, governance, identity, infrastructure, Internet, LLM, malicious, monitoring, network, risk, service, supply-chain, threat, tool, vulnerability, zero-trustThe epistemological chasm: What renders MCP vulnerabilities particularly vexatious is the fundamental asymmetry they exploit between machine cognition and human oversight.Tool poisoning attacks insert malevolent instructions into tool metadata that LLMs process with complete fidelity but that remain utterly invisible to human operators. The machine perceives everything; its ostensible supervisors perceive nothing. We have unwittingly…
-
OpenAI and Anthropic LLMs Used in Critical Infrastructure Cyber-Attack, Warns Dragos
Commercial AI models were used to help plan and conduct cyber-attack against operational technology of a water and drainage facility, say researchers First seen on infosecurity-magazine.com Jump to article: www.infosecurity-magazine.com/news/llm-critical-infrastructure/
-
Poisoned truth: The quiet security threat inside enterprise AI
It takes surprisingly little poison to corrupt: Bad internal data is the immediate problem. But the external supply chain may be even harder to control.Research by Anthropic, the UK AI Security Institute, and the Alan Turing Institute discovered that as few as 250 maliciously crafted documents can poison LLMs of any size.That creates a massive…
-
We Scanned 1 Million Exposed AI Services. Here’s How Bad the Security Actually Is
While the software industry has made genuine strides over the past few decades to deliver products securely, the furious pace of AI adoption is putting that progress at risk. Businesses are moving fast to self-host LLM infrastructure, drawn by the promise of AI as a force multiplier and the pressure to deliver more value faster.…
-
What researchers learned about building an LLM security workflow
Security operations centers are running into the same wall everywhere. Detection tools generate more alerts than analysts can work through, and the early stages of any … First seen on helpnetsecurity.com Jump to article: www.helpnetsecurity.com/2026/05/04/building-llm-security-workflow/
-
Usage-based pricing killing your vibe – here’s how to roll your own local AI coding agents
Take those token limits and shove them by vibe coding with a local LLM First seen on theregister.com Jump to article: www.theregister.com/2026/05/02/local_ai_coding_agents/
-
AI agents can bypass guardrails and put credentials at risk, Okta study finds
Phishing the agent: Why AI guardrails aren’t enough, a report on tests conducted by cloud identity and access management (IAM) company Okta Threat Intelligence, which uncovered all of the problems cited above, and more.Their research focused on OpenClaw, a model-agnostic multi-channel AI assistant which has seen explosive growth inside enterprises since appearing in late 2025.…
-
AI agents can bypass guardrails and put credentials at risk, Okta study finds
Phishing the agent: Why AI guardrails aren’t enough, a report on tests conducted by cloud identity and access management (IAM) company Okta Threat Intelligence, which uncovered all of the problems cited above, and more.Their research focused on OpenClaw, a model-agnostic multi-channel AI assistant which has seen explosive growth inside enterprises since appearing in late 2025.…
-
prompted 2026 Source to Sink: Improving LLM Vuln Discovery
Author, Creator & Presenter: Scott Behrens, Principal Security Engineer At Netflix & Justice Cassel, Application & GenAI Security At Netflix, Our thanks to [un]prompted for publishing their Creators, Authors and Presenter’s outstanding [un]prompted 2026 AI Security Practitioner content on the Organizations’ YouTube Channel. Permalink First seen on securityboulevard.com Jump to article: securityboulevard.com/2026/05/unprompted-2026-source-to-sink-improving-llm-vuln-discovery/
-
AI agents can bypass guardrails and put credentials at risk, Okta study finds
Phishing the agent: Why AI guardrails aren’t enough, a report on tests conducted by cloud identity and access management (IAM) company Okta Threat Intelligence, which uncovered all of the problems cited above, and more.Their research focused on OpenClaw, a model-agnostic multi-channel AI assistant which has seen explosive growth inside enterprises since appearing in late 2025.…
-
prompted 2026 Traditional ML vs. LLMs: Who Can Classifv Better?
Author, Creator & Presenter: Xenia Mountrouidou, Principal Cyber Data Scientist At Expel Our thanks to [un]prompted for publishing their Creators, Authors and Presenter’s outstanding [un]prompted 2026 AI Security Practitioner content on the Organizations’ YouTube Channel. Permalink First seen on securityboulevard.com Jump to article: securityboulevard.com/2026/04/unprompted-2026-traditional-ml-vs-llms-who-can-classifv-better/
-
Bad Bots in the Agentic Age: What the 2026 Thales Bad Bot Report Reveals
Tags: ai, api, application-security, attack, automation, banking, business, container, control, crime, cyber, cybercrime, data, defense, detection, exploit, finance, fraud, identity, infrastructure, intelligence, Internet, LLM, malicious, monitoring, resilience, risk, service, threat, tool, vulnerabilityBad Bots in the Agentic Age: What the 2026 Thales Bad Bot Report Reveals josh.pearson@t“¦ Thu, 04/30/2026 – 07:31 The modern internet is becoming less human by the day. Bot traffic is increasing, and human traffic is shrinking. Malicious automated traffic is getting harder to spot. The Thales 2026 Bad Bot Report, now in it’s…
-
Automated LLM red teaming gets a learning layer
Automated red teaming of large language models has settled into a familiar pattern over the past two years. An attacker model generates jailbreak attempts against a target … First seen on helpnetsecurity.com Jump to article: www.helpnetsecurity.com/2026/04/30/automated-llm-red-teaming-learning-layer/