Cyber CrimeNorthflank breach exposes AI key theft risk
Cloud platform Northflank told customers to rotate all secrets after a breach led to unauthorized use of AI API keys stored on its systems.
> Search
27 results for “llm”
Cyber Crime
Cyber CrimeAI Agent Tooling Used in Bank Hacks
CrowdStrike attributes a data theft campaign against South Korean financial firms to a Chinese-speaking actor who paired an agentic pentesting tool with multiple LLMs.
AI & MLAI malware cuts humans out of C2 loop
Cisco Talos says it has identified CLOSEDQUORUM, a proof-of-concept implant that uses an LLM panel to drive an attack without a live operator.
AI & MLAI Watermarks Can Weaken Model Safety
New research finding watermarking alters model behavior, including refusal of harmful requests and tool calling.
Cyber Crimenpm stealer tied to bounty hunter
CrowdStrike says a claimed bug bounty hunter likely built the PhantomRaven npm stealer with an LLM, active since November 2022.
AI & MLAI assets become prime targets for attackers
Google's threat intelligence unit reports that espionage and criminal groups are stealing models, prompts, and API keys to power their own AI operations.
SecurityMarimo RCE Hit by Human at Machine Speed
Sysdig says a hand-built toolkit let a human operator exploit a Marimo flaw in eight seconds, without any sign of LLM use.
SecurityLiteLLM's Default Admin Key Exposed
A Wiz Research scan found 294 exposed LiteLLM gateways accepted the example admin key sk-1234, granting access to stored provider credentials.
Cyber CrimeStolen AI Tokens Skirt Login Checks
Okta's review of a 7 GB infostealer dump found unexpired AI service tokens that can be replayed without a password or MFA.
SecurityMalware Lures LLMs Toward Forbidden Prompts
Russia-aligned UAC-0099 embeds nuclear-weapon-style prompts in a VBS script to misdirect AI-assisted malware analysis, ESET says.
SecurityAI agents can be tricked into installing malware via unclaimed code packages
Researchers found 120 unregistered domains in AI documentation that could be hijacked to infect corporate networks.
SecurityHidden HTML Hijacks AI Email Summarizers
Forcepoint shows invisible text can silently change what an AI assistant reads in your email.
SecurityAI-Assisted Malware: Faster to Build, Not Harder to Catch
Unit 42 finds most AI-linked malware never reaches real targets, but speed of development is rising.
SecurityGrok tricked into leaking user data
A researcher found that encrypting malicious instructions lets Grok exfiltrate user chats and personal details.
SecurityAI Threat Modeling Needs a Faster Starting Point
PHANTOM-B framework aims to make AI threat modeling fast enough for busy teams, but experts warn against skipping fundamentals.
AI & MLDefensive AI Startup Targets the Offense Gap
Corma's AI agents stopped a live attack in under 10 minutes while its CEO walked his dog, showing defenders can keep pace.
SecurityWhy 2,500-Org Breach Wasn't What It Seemed
SOCRadar says most orgs hit in LiteLLM attack were earlier Trivy victims, not LiteLLM.
SecurityLiteLLM Attack's True Reach Exposed
CloudSEK details how a Trivy compromise cascaded into LiteLLM, affecting 2,500+ orgs and 434,000 pipelines.
SecurityWhy AI Orchestration Is a Security Risk
New research shows that the choice of AI orchestration framework can increase agent compromise rates by more than 2.6 times.
Architecting Scalable RAG Pipelines
Building retrieval-augmented generation systems requires balancing model latency against data throughput to maintain system integrity.
Cyber CrimeLLM-Assisted Phishing Kits Scale Attacks
Researchers uncovered an exposed server revealing how attackers use AI to industrialize the creation of malware delivery campaigns.
Tech NewsAdobe Adds AI Critics to Mobile Cameras
The experimental Project Indigo app is integrating LLM-based feedback to analyze user photography and suggest real-time edits.
SecurityAI Spam Filters Falter Against Salting
Modern LLM-powered email security is struggling to identify phishing attempts that use decades-old text-hiding techniques.
SecurityGoogle Automates Opt-In for AI Training
Google has updated its search privacy settings to automatically include user media in its large language model training processes.
SecurityThe Hallucination Trap: How AI Browsers Fail Under Pressure
New research reveals that AI browsers can be tricked into ignoring safety guardrails by forcing them into a state of logical delusion.
SecurityDefenders Weaponize Prompt Injection to Neutralize Rogue AI Agents
Researchers at Tracebit are utilizing context bombing to force AI models to trigger their own safety guardrails during attacks.
SecurityHalluSquatting Turns AI Assistants Into Silent Botnet Factories
Researchers discover a critical vulnerability where AI coding agents are tricked into executing malicious code via hallucinations.