CERT-AGID identified and disrupted a fraudulent website impersonating ARERA to trick users into revealing personal and financial data through a fake refund tied to the social water bonus. The site used typosquatting and a staged flow that led victims from a phone-number check to a bogus credit card verification page. #ARERA #CERT-AGID #bonussocialeidrico
Category: Threat Research
Sysdig Secure AI uses agentic cloud security to continuously triage vulnerability backlogs, trace 2,731 SLA-breaching findings in a Node.js base image to one fix, and route the remediation through human approval into Jira ticket DEJI-342. The same capability also runs headless in Claude via the Sysdig MCP server, using runtime data to prioritize issues by exploitability, KEV status, and reachability. #SysdigSecureAI #Jira #Claude #SysdigMCPServer #DEJI342
A single PDF factory has produced more than 12,700 structurally similar FakeCaptcha documents on Webflow’s CDN, where they appear in Google search as ordinary “upgrade guides” and feed a traffic-distribution system. The campaign routes qualified visitors to malware, scam, or reseller infrastructure while using rotated lure domains, ww80/wwNN routers, and a custom Elixir/Phoenix gate to keep the operation active. #FakeCaptcha #Webflow #GoogleGemini #Claude #LegionLoader #berapt-medii #yfdpco
A threat actor compromised hundreds of npm packages on August 4, 2026 and used them to distribute a self-propagating backdoor that spreads in ways similar to the Shai-Hulud npm worm. The campaign affected popular packages such as Keyv, Cacheable, and Ecto, and used multiple propagation, exfiltration, and credential-theft paths across npm, GitHub, cloud, and CI environments. #Keyv #Cacheable #Ecto #ShaiHulud
CERT-AGID identified a phishing campaign that abuses the names and visuals of Polizia di Stato and pagoPA to trick victims into paying a fake traffic fine. The fraudulent flow collects vehicle plate data, tax code, email address, and payment card details, while domains containing “poliziadistato” were used to host the fake pages. #PoliziaDiStato #pagoPA #CERTAGID
July featured major security developments including the first agentic ransomware operation attributed to JADEPUFFER, a new US vulnerability coordination body called GOLD EAGLE, and an AI-assisted breach at Hugging Face involving OpenAI agents. The month also saw Azure tenant takeover abuse, ENCFORGE ransomware targeting AI/ML assets, and several high-profile breaches affecting FastJson, Abbott Laboratories, Accenture, and Fairlife. #JADEPUFFER #GOLDEAGLE #HuggingFace #OpenAI #ENCFORGE #FastJson #AbbottLaboratories #Accenture #Fairlife
Elastic built an AI-assisted bug bounty triage system to handle a surge of AI-generated HackerOne reports, using multi-stage analysis, adversarial review, and sandboxed reproduction on ephemeral Google Cloud VMs. The system agrees with human security engineers 85% of the time and uses Elastic-specific triage rules to distinguish real issues from features or out-of-scope reports while keeping a human in the final decision loop. #HackerOne #Elastic #Claude #ElasticWorkflows #GoogleCloudPlatform #Elasticsearch #Kibana
Socket is tracking an active supply chain compromise in the keyv and cacheable npm ecosystems, where trojanized packages used a malicious preinstall hook to download Bun, steal cloud and CI credentials, and republish infected packages through stolen npm tokens. The campaign affected tens of millions of weekly downloads and also planted persistence in developer and AI coding environments via .claude and .vscode hooks. #keyv #cacheable #Jaredwray #Bun #npm #GitHubActions #HashiCorpVault
July 2026 attacks showed how trusted business workflows, legitimate platforms, and built-in tools were abused to drive account takeover, data theft, fraud, and persistent access across the US, Europe, and Brazil. ANY.RUN highlighted campaigns involving Kratos, PhantomEnigma, Kali365, Banana RAT, DARTHVADER Stealer, OVERLORD RAT, DestinyStealer, and fake Zoom event lures that…
Unit 42 found that nearly half of malware samples with command-and-control activity make direct-to-IP connections, bypassing DNS and exposing a major visibility gap in traditional defenses. The report highlights ZT-IP detections for Phorpiex, SectopRAT, Mozi, and Boatnet activity, plus an obfuscated GET exfiltration campaign and hard-coded IP-based infrastructure. #Phorpiex #SectopRAT #Mozi…
Aqua detected a multistage fileless XMRig cryptojacking campaign targeting containerized Node.js applications built with Next.js, exposing malicious activity across in-memory execution, container drift, persistence, and outbound mining communication. The campaign used tactics like memfd-based loading, cron and rc.d persistence, SSH key backdoors, and anti-removal defenses, while Aqua runtime enforcement could block the attack before unauthorized code ran or mining traffic left the workload. #XMRig #Nextjs #Aqua #memfd #Nodejs
Recent attacks on water and wastewater systems in Minnesota, Michigan, Canada, and elsewhere show that exposed PLCs, HMIs, cellular modems, and weak remote-access setups can allow attackers to disrupt physical operations without advanced malware. The incidents are technically similar to an Iranian-affiliated PLC campaign, while separate claims by NoName057(16) and Z-Pentest show that hacktivists are also targeting water infrastructure. #MicroLogix1100 #MicroLogix1400 #NoName05716 #ZPentest #CyberAv3ngers #Unitronics
Talos found that threat actors are rapidly abusing AI for malicious development, operational scaling, and vulnerability research, while weak guardrails are often bypassed with simple claims of ownership, bug bounty labeling, or task decomposition. The report shows AI being used to build DDoS tooling, bulk-mail validation systems, credential-harvesting pipelines, mining operations,…
Elastic built an evidence-based evaluation framework for agentic SOC models that tests real tool use, grounded decisions, and workflow execution inside Elastic Security rather than relying on generic LLM benchmarks. The article also covers parallel suites for Attack Discovery and Automatic Migration, all scored blind to measure quality and reliability across security tasks. #ElasticSecurity #Chrysalis #VirusTotal #Splunk #QRadar #MicrosoftSentinel
Trusted repositories in coding agents can execute attacker-controlled code before the first prompt through project-scoped MCP configuration and environment-based PATH manipulation, even without malicious hooks or shell-command approval. The article also recommends defensive searching and runtime monitoring, and highlights Datadog AI Guard for Coding Agents and Workload Protection as protections against these pre-prompt execution paths. #Codex #ClaudeCode #MCP #DatadogAIGuardForCodingAgents