Hvězdička u CVE znamená, že jsem to číslo vytáhl z širšího textu článku. Neukazuju u něj proto KEV, EPSS ani CVSS, a to preventivně, protože článek se na něj mohl jen odkazovat, třeba jako na starší kauzu.
Employees using personal ChatGPT accounts at work can expose sensitive company data, even when their organization provides an approved Enterprise workspace. NordLayer explains how IT teams can investigate unauthorized AI use and block sensitive information before it reaches personal accounts. [...]
US
The National Institute of Standards and Technology (NIST), National Security Agency’s Super Intelligence Security Center (NSA SISC) and Cybersecurity and Infrastructure Security Agency (CISA) published Key Practices for Incident Readiness and Response in Super Intelligence Training and Evaluation. The product outlines ten voluntary core practices to help organizations that train and evaluate highly capable super intelligence (SI) agents—particularly pre-deployment, cutting edge SI agents with…
US
AI can help you send more job applications. But when you lose track of where you’ve applied, a fake recruiter’s message can start to look like the reply you’ve been waiting for. Job hunting can feel like shouting into the void. You send applications, rewrite cover letters, and wait for replies that may never arrive. Then a message lands on your phone: someone wants to talk about a job. But was it one of the jobs you applied for? It is hard to keep track when you’ve sent out a lot of them. Two…
US
OpenAI is facing separate legal challenges over how ChatGPT interacts with children. The Guardian reports that eight US states accuse the company of misleading parents about the chatbot’s safety and exposing vulnerable users to dangerous conversations. Separately, CNET reports that a mother suing on behalf of her two children alleges that OpenAI collected and used children’s sensitive personal information without valid parental consent. The cases raise serious questions about what happens when…
OpenAI US
Classification: Critical, Solution: Official Fix, Exploit Maturity: Proof-of-Concept, CVSSv3.1: 9.9, CVEs: CVE-2026-105740, Summary: Any authenticated Langflow user can achieve Remote Code Execution (RCE) on the server by adding an MCP server with the "Stdio" transport. The user-supplied command field is passed directly to bash -c "exec {command}" with zero validation, no allowlisting, and no sandboxing. The command executes immediately when the server list is fetched. Additionally, the env…
EPSS 0.01 CVSS 9.9 CVE-2026-105740 Langflow FI
I believe the cybersecurity industry may be approaching AI security the wrong way. AI agents are becoming more powerful every day. Systems like ChatGPT, Claude, and Grok can write and execute code, use command shells, browse the web, call APIs, interact with MCP servers, and access enterprise data. Yet much of the industry is taking these extraordinarily powerful, general-purpose agents and trying to make them safe by building cages around them. We add containers, policy layers, monitoring,…
Check Point US
Our analysis an AI-powered espionage campaign targeting government Linux infrastructure with custom malware, web shells, and advanced credential theft. The post AI-Powered Espionage Campaign Targets Government Linux Infrastructure appeared first on Unit 42.
Google announced new AI-powered defenses against scam messages and call-forwarding abuse on Android, currently rolling out to eligible devices. [...]
AI pentesting can uncover vulnerabilities missed by human testers, but it also has limitations in consistency, risk assessment, and understanding business context. Intruder compares the strengths and weaknesses of both approaches and explains where AI agents can improve security testing. [...]
US
Photos you take with your phone and some digital cameras can contain hidden information, including where they were taken. You can strip that data out before sharing a photo with someone, which would help to protect your location. But the image itself may still give you away. A company called Graylark says its Raven AI tool can match an indoor photo to a specific property using visual clues alone. No GPS data or view out of the window required. We’ve seen this company before. Its tool, GeoSpy,…
Your streaming box might not look like much of a computer. But connect enough inexpensive devices together, and they can run an AI model. Researchers at Anthropic demonstrated this by running a Google Gemma model across 15 off-brand Android TV boxes. They warn that small models could soon spread across vulnerable internet-of-things (IoT) devices, including smart ovens and TV boxes, stating: “Our experiments suggest that small AI models may be months away from being capable enough to self…
Texas HB 3834 and HB 3512 require annual cybersecurity and AI awareness training. Learn who’s in scope, what’s required, and how to comply fast.
Learn the latest endpoint security trends, from AI-powered defense to zero trust and human-led threat hunting that help under-resourced teams stop threats faster.
US
AI-powered penetration testing can uncover complex attack paths, but safety restrictions and network limitations leave many assets and vulnerabilities untested. Picus Security explains how combining agentic pentesting with exposure validation and breach simulation can help close these gaps. [...]
TruConfidence: The AI-Powered Remediation Intelligence & Orchestration engine powering TruRisk Eliminate 2.0 delivers autonomous remediation at scale with proof at every step. Executive Summary Vulnerabilities are now discovered faster than enterprises can remediate them. Patch automation has been around for years, but vulnerability backlogs are still growing. Automation isn’t the bottleneck. Confidence is: without evidence of how a change will behave, Security and IT teams hold back, and the…
Qualys US
Police tip hotlines already have their fair share of crank callers. Apparently they now have to contend with crank AI agents as well. AI agents have already been caught hacking software companies, breaching an Australian government portal, and trying to hack a Canadian government website. Now Anthropic’s Claude has added a made-up murder tip to the list. Claude’s Haiku 4.5 model filed a false tip in an unsolved Philadelphia homicide via the police department’s online public tip form, Anthropic…
Classification: Severe, Solution: Official Fix, Exploit Maturity: Not Defined, CVSSv3.1: 10.0, CVEs: CVE-2026-65180, CVE-2026-65173, CVE-2026-65172, CVE-2026-65170, CVE-2026-65164, CVE-2026-65162, CVE-2026-65161, CVE-2026-65131, CVE-2026-65110, CVE-2026-65104, CVE-2026-65103, CVE-2026-65101, CVE-2026-67456, CVE-2026-65150, CVE-2026-65149, CVE-2026-65148, CVE-2026-65147, CVE-2026-65141, Summary: CVE-2026-65180 7.8 - NVIDIA Megatron-LM contains a vulnerability in checkpoint loading where an…
EPSS 0.01 CVSS 10.0 a dalších 1 CVE-2026-65103 CVE-2026-65104 CVE-2026-65110 CVE-2026-65131 CVE-2026-65141 CVE-2026-65147 CVE-2026-65148 CVE-2026-65149 CVE-2026-65150 CVE-2026-65161 CVE-2026-65162 CVE-2026-65164 CVE-2026-65170 CVE-2026-65172 CVE-2026-65173 CVE-2026-65180 CVE-2026-67456 NVIDIA FI
Anthropic is warning that small AI models could soon be able to self-replicate across consumer IoT devices, from Android TV boxes to smart ovens, creating distributed botnets capable of running AI workloads without relying on data centers. [...]
US
The new research, published by AI security startup AIR, found that a network filtering flaw in Vercel Sandbox — a tool used to run code in isolated environments — could allow agents to exploit certificate validation processes to transmit data outside approved networks.
Vercel US
TL;DR; Our security assessment of LiteLLM (v1.77.7), a widely used open-source LLM gateway, uncovered multiple high-impact vulnerabilities (tracked as CVE-2026-35029, CVE-2026-59823) which have been resolved as of LiteLLM version 1.83.9. The most significant issues stemmed from a missing authorization check on the /config/update endpoint, which allowed any authenticated user (including low-privileged accounts) to modify the proxy’s runtime configuration. This single flaw enabled arbitrary file…
EPSS 0.39 CVE-2026-35029 CVE-2026-59823 LiteLLM IL
Critical infrastructure faces mounting threats from nation-state groups and AI-driven attacks. In the latest installment of our Cybersecurity Awareness Month blog series, learn how leading cybersecurity agencies, such as CISA, are advising critical infrastructure organizations on ways to protect themselves and the essential services they provide. Key takeaways Critical infrastructure organizations often struggle with asset visibility gaps that create hidden attack vectors. Many older OT systems…
<p>This article was first published <a href="https://www.linkedin.com/pulse/building-intent-based-access-control-without-reading-minds-borges-dxg6e/" target="_self">on LinkedIn</a>.</p> Categories: AI Research, Threat Research Tags: AI, AI RESEARCH, Threat Research
GB
Anthropic announced that its internal AI models undergoing "high-risk and cybersecurity evaluations" no longer have internet access following incidents involving U.S. government websites at the federal, state, and local levels. [...]
V pondělí 5. řína proběhlo setkání Pracovní skupiny CSIRT.CZ jíž se účastnilo více než 100 účastnic a účastníků z bezpečnostních týmů napříč sektory. Pracovní skupinu otevíral Ondřej Filip. Následovala přednáška Andrei Kropáčové z CESNETu, která navázala na úvodní řeč Ondřeje Filipa. Andrea ocenila zvládnutou kontinuitu činností, která vyplynula nejen z této řeči 🙂, ale byla především tématem první přednášky. Andrea připomněla dvě věci, které znějí jednoduše, ale v krizové situaci rozhodují:…
CZ
US
Selling to banks or government? Aikido now offers CREST-accredited penetration testing, plus AI pentesting between audits. Category: Compliance
BE
You ignored the suspicious email. You didn’t open the attachment, click the link, or reply to the sender. Then you asked your AI assistant to summarize your inbox. Could the message you wisely avoided find its way back onto your to-do list—only this time presented as an urgent recommendation from a tool you trust? That’s the risk behind phishing emails containing hidden instructions aimed at AI assistants. Instead of relying only on persuading you, attackers are also trying to influence the…
US
The cyberattacks that shook the South Korean financial sector earlier this month were launched by a Chinese hacker using the ARTEX AI penetration testing suite and Claude agents. [...]
finance US
Traditional attack surface management helps organizations discover exposed assets, but visibility alone is not enough to address threats. Criminal IP introduces AITEM, an AI-powered approach that connects exposure discovery with investigation, risk prioritization, and response. [...]
Criminal IP US
AI agents can use valid credentials to perform actions beyond their assigned permissions, creating risks that traditional access controls may not prevent. Token Security explains how organizations can enforce agent-specific policies without sacrificing autonomy. [...]
US
Agentic AI is the breakthrough of the moment, in security as everywhere else: agents that are threat-aware, active, and proactive in investigation. But it is also the technology behind a recent documented autonomous AI intrusion. Both facts are true, and the distance between them is what this article is about.
SG
Welcome to this week’s edition of the Threat Source newsletter. My name is Pierre Cadieux, and I’ll be helping contribute to these newsletters. A little about me: I’ve been working in the cybersecurity industry in many roles over the past 20+ years, first focusing on endpoint security, policies, and firewalls, then moving to risk management and compliance, disaster recovery, and investigations. I spent about 15 years as a consultant working across many well-known consultancy firms and…
KEV ✓ EPSS 0.11 CVSS 8.7 CVE-2026-88779 Cisco Citrix Google US CA FI GB FR IT
We just did a major update to FOR577 and added a lot of new material on day 5 about investigating AI usage in incident response. In the new material we dicsuss 8 of the most popular AI coding assistants and agents including Claude Code, Codex, Gemini CLI, Cursor, Copilot, Warp, Windsurf, and Qwen Code. I've been using Claude Code and a little bit of Codex, but I also have recently been playing with OpenCode and am setting up Hermes. I decided to figure out where the chat history was in OpenCode…
US
Community-built AI agents, skills, and MCP servers are landing in SOC workflows fast. Here’s what the Exchange Inspector tests before a listing earns its vetted tag on the CyberAgents Exchange. Three tools have already passed. Key takeaways Every Inspector-vetted listing clears three gates: an automated check, a frontier model assessment, and human verification. Tenable uses Tenable One AI Exposure to screen for prompt injection and exposed secrets, and OpenAI GPT Cyber models to assess the…
Microsoft will soon introduce support for third-party deepfake detection solutions and impersonation protection in Teams meetings. [...]
Microsoft US
If you thought that having companies trawling your social media, browsing history, and TV habits for behavioral clues was bad, sit tight. Meta is just getting started. Its Muse personal AI agent is taking surveillance to the next level. TIME magazine analyzed the software’s internal instructions and found that it maintains constantly updated dossiers on users and the people they know. Meta released Muse last month, positioning it as a digital assistant that can handle different parts of your…
Meta US
Day Three of Pwn2Own Ireland 2026 wrapsup with 17 attempts scheduled, roughly $1.33 million on the board. The headliner is the Google Pixel 10, which draws three separate attempts: Xint at 11:15 AM, Ikotas Labs at 2:00 PM, and the Valsamaras/Ken Gannon (Djini.ai)/Valsamara team closing things out at 6:30 PM — by far the single biggest payday in play all week. The Samsung Galaxy S26 gets one more run too, from BunkyoWesterns at 4:15 PM. Smart Home stays busy with three attempts on the Philips…
Cisco Talos identified an advanced persistent threat (APT) spear-phishing campaign against individuals affiliated with Taiwan research organizations. The operation leveraged legitimate public event themes and impersonated reputable academic and policy institutions to establish credibility. The phishing emails exhibited highly consistent structure, rhetoric, and personalization patterns, suggesting the threat actor likely used AI-assisted content generation to rapidly customize invitation lures…
“AI-analysis evasion” encapsulates the real-world techniques malware authors are developing in attempt to obstruct or defeat any layers of automated AI analysis. This technique is cheap to add but inconsistently impactful — the best techniques steered the outcome in the attacker’s favor in about 35% of test runs. Further, it must always be plaintext and therefore is always detectable. The operators are not wrong to assume AI tools are in the analysis pipeline, but the answer is not to remove…
US
US
For the last several months, we’ve been busy talking to security teams and building answers to the AI-accelerated challenges that you’ve told us you’re facing. Among these are the overwhelming volume and velocity of alerts, exacerbated by attackers equipped with AI. The Hugging Face incident demonstrated another core challenge in the security operations center (SOC). Even if all of the signals are detected, that’s not enough, unless something connects them as part of a single attack. Combined,…
Elastic US
In this article Shift 1. From frontier capability to scale Shift 2. From token consumption to reasoning economics Shift 3. Validation and remediation as a continuous learning loop Transferring security research to help open source projects Open-source vulnerability reporting snapshot Measured costs of automated validation FORGE OSS Bug Hunt Party: Human coordination at scale (Hackathon) Akrites: Coordinated effort to remediate vulnerabilities What’s next for FORGE Lab? The mission of Microsoft…
Microsoft US
Rilevata una vulnerabilità con gravità "alta" in llama.cpp, noto motore di inferenza open source che consente di eseguire LLM in locale. Tale vulnerabilità, qualora sfruttata, potrebbe consentire a un attaccante remoto di compromettere la disponibilità del servizio e, potenzialmente, di eseguire codice arbitrario sui sistemi interessati.
EPSS 0.00 CVE-2026-107183 llama.cpp IT
A cryptomining campaign targeting exposed AI services is using PoeLLM malware to turn compromised servers into scanners and exploit launchpads. [...]
A North Carolina musician was sentenced to 18 months in prison for collecting more than $10 million in royalties from Spotify, Apple Music, Amazon Music, and YouTube Music in a massive streaming royalty fraud scheme. [...]
For some time now, the cybersecurity community has seen examples of autonomous agents, built inside AI labs, attacking public infrastructure (to name a few, Hugging Face, DSEWiki, and RubyGems). Of course, frontier labs have built-in security to prevent these attacks from occurring, but every now and then, the training or prompting appears to be insufficient — especially when the agents themselves attempt to use logic to probe and bypass the restrictions placed on them. The question that…
US
Day Two of Pwn2Own Ireland 2026 has 24 attempts on the docket, with roughly $780,000 up for grabs, running 9:30 AM to 7:30 PM IST. Samsung's Galaxy S26 will be attempted three separate times — by Valsamaras/Ken Gannon (using Djini.ai)/Valsamara at 1:00 PM, Kyeongmin Kim at 3:15 PM, and PetoWorks at 5:30 PM — worth watching since only the first successful demo claims the cash. Home Assistant Green is the day's most-attempted device, with five researchers scheduled to take a run at it. AI…
Rilevate 24 nuove vulnerabilità di sicurezza, di cui 2 con gravità “critica” e 18 con gravità “alta”, che interessano il software Langflow, nota piattaforma open-source che permette di costruire, testare e distribuire applicazioni e agenti basati su intelligenza artificiale.
EPSS 0.15 a dalších 3 CVE-2026-104335 CVE-2026-88962 CVE-2026-93443 CVE-2026-93445 CVE-2026-93447 CVE-2026-93449 CVE-2026-93674 CVE-2026-93675 CVE-2026-93677 CVE-2026-93678 CVE-2026-97655 CVE-2026-97673 CVE-2026-97674 CVE-2026-97676 CVE-2026-97678 CVE-2026-97679 CVE-2026-97680 Langflow IT
finance US
Serial number: AV26-1005 Date: October 6, 2026 As of October 6, 2026, Progress Software is affected by a vulnerability in the following product: Progress Software Autonomous REST Connector GenAI Agents ARCGenAI-Generator Prior to 2.1 The Cyber Centre encourages users and administrators to review the provided web links and apply any necessary updates as they become available. Progress DataDirect Critical Security Alert Bulletin – September 2026 – CVE-2026-91140 Progress Trust Center
EPSS 0.02 CVE-2026-91140 Progress CA
Most of what has been written about AI and vulnerability management focuses on speed: how much faster frontier AI models can scan code, find weaknesses, design patches, and build exploits than any human team. That part is true, and it matters to how we remediate vulnerabilities. The more challenging question is what happens after the scan? Frontier AI models are about to hand every security team a far larger set of findings than they have ever had to work through. The real test for chief…
Microsoft US
A new campaign targeting ad account managers uses fake ChatGPT, Gemini, Claude, and Perplexity sites that steal login credentials and multi-factor authentication (MFA) codes through browser-in-browser attacks. [...]
US
AI tools let non-technical employees build workplace apps in minutes with natural language prompts. While these AI-generated apps boost productivity, they can create severe security and data risks. As Cybersecurity Awareness Month kicks off, we’re sharing how Tenable adopted a structured governance framework that allows our “citizen coders” to build secure and compliant apps with AI. Key takeaways Unsanctioned applications that non-technical staff build using AI tools often bypass quality…
US
As organizations deploy autonomous AI agents, security teams face a significant shift as non-human non-human entities making decisions, invoking tools, and delegating tasks to other agents without human intervention. Security architectures built around human users, static APIs, and distinct endpoints break down when AI agents dynamically collaborate across an environment. As these interactions become more common, securing agent-to-agent communication without blocking adoption will require…
US
Aikido has achieved FedRAMP Moderate authorization, bringing AI-native, automated vulnerability remediation to U.S. federal agencies. Category: Company Updates
The Wikimedia Foundation says rogue OpenAI agents made unauthorized Wikipedia edits and may have been partially responsible for a May outage. [...]
OpenAI US
Disponibili Proof of Concept (PoC) per lo sfruttamento di 4 nuove vulnerabilità di sicurezza, di cui 2 con gravità “critica” e 2 con gravità “alta”, che interessano il software Langflow, nota piattaforma open-source che permette di costruire, testare e distribuire applicazioni e agenti basati su intelligenza artificiale.
EPSS 0.01 CVE-2026-105697 CVE-2026-105699 CVE-2026-105740 CVE-2026-105741 Langflow IT
US
Fáilte! Pwn2Own Ireland 2026 has arrived in Cork, and the world's top security researchers are ready. We've got more phone entries than we've ever had, and more than 60 entries total. Earlier today, we held the random draw to determine attempt order. Below is the official schedule. All times are Irish Standard Time (IST) and may change as the competition progresses. You can find our live leaderboard here. This will be updated throughout the day. Jump to: Day One Day Two Day Three DAY ONE…
Companies like Google and Microsoft are find much bigger numbers of vulnerabilities in their own products as a result of AI. But the same technology is leading to public reporting programs becoming overwhelmed by the mass submission of speculative, duplicated, or hallucinated findings. Now, Google’s announced it has temporarily stopped accepting submissions to its open source bug bounty program, OSS VRP. “Why is this happening? This pause is due to a significant rise in automated submissions,…
Google US