Patch, watch and understand: up to eight stories from the past 24 hours.
Latest included publication: .
Automatically selected from published coverage, prioritising known exploitation, critical severity and source coverage. Selection is not a human review. Each story links to its evidence and available actions.
A high-severity flaw tracked as CVE-2026-114713 in UnrealIRCd◆ allows attackers to execute commands with server and IRCOp privileges. The article reports that the vulnerability is already being exploited in attacks. It does not give affected versions or a fixed release.
Huntress◼ has confirmed exploitation of CVE-2026-84869 in ScreenConnect against MSP technicians, allowing malicious files to be run on their accounts. The excerpt does not give affected versions, a patch status or the full attack chain. Huntress has published the attack chain, indicators of compromise and protective guidance.
Honeypots at the SANS Internet Storm Center saw requests to /mcp-rest/test/tools/list that appear to exploit CVE-2026-42271 and CVE-2026-48710, which Horizon3.ai◼ says chain to remote code execution without authentication. The root cause is that Starlette◆ versions before 1.0.1 do not validate the Host header, so LiteLLM◆'s access control can be bypassed using crafted Host values.
Arion Kurtaj, 21, from Oxford, has pleaded guilty to a string of online offences, including the 2022 hack in which he stole dozens of unfinished Grand Theft Auto 6◆ clips and tried to extort the makers. He was originally sentenced to an indefinite stay in a secure hospital and was released before entering his plea.
The FBI◼ alleges a campaign that distributed eight games containing malware, which collected credentials, browser cookies, tokens and autofill data, then used them to drain crypto wallets. The complaint, filed in July, names Zyaire Dontaevious Zamarion Wilkins, and a further charge was filed on 9 November; the FBI's victim notice links the games to Steam◆. Promotion ran through Discord, Telegram, X and LinkedIn, with bots targeting users holding large crypto balances.
The Google Threat Intelligence Group reports that attackers are deploying autonomous multi-agent frameworks such as CAIRN and PentAGI to run multi-stage intrusions faster, and some are automating patch reverse engineering to weaponise n-day flaws. The techniques are familiar, relying on stolen credentials and unpatched systems. Mandiant◼ has handled LLMjacking incidents where fraudulent charges reached $1 million in the worst case before cleanup.
Google Threat Intelligence Group◼ and Mandiant report that threat actors are using open-source multi-agent hacking frameworks such as ARTEX, CAIRN and PentAGI for reconnaissance, privilege escalation and lateral movement. GTIG also found an exposed framework that found, wrote and ran exploits on its own, and GreyNoise◼ reported an AI-assisted campaign against PaperCut flaws that reached domain admin within minutes. Separately, attackers are stealing AI keys and cloud access for resale, and fake charges have reached very high sums.
The OWASP Top 10 Neocloud and AI Data Center Security Risks is a new incubator project led by Michael Katchinskiy and Yakir Kadkoda of Lava◼. It ranks infrastructure risks such as firmware tampering, weak GPU interconnects, poor tenant isolation and insecure out-of-band management, with the first five rated critical. The authors cite exposed IPMI interfaces and the Supermicro◼ BMC patching problems as real examples.