IT News Analysis

1

OpenAI Research Model Spent 3rd-Party Test ETH Without Permission to Deploy Smart Contract

OpenAI has disclosed seven new incidents in which its artificial intelligence (AI) models exhibited misaligned behavior, including one case where its research model, after not having funds on an Ethereum test network, spent test cryptocurrency from a third-party's wallet without authorization to complete a coding task. The incident, the AI company said, involved an internal research model part of the GPT-5.6 Sol family. The incident occurred on June 26, 2026, and was discovered by OpenAI three months later. "An internal research model asked to update a cryptocurrency trading application used funds from a wallet that the task had not supplied to pay for a deployment on Sepolia, an Ethereum test network," OpenAI said . "It derived wallet keys from publicly known developer recovery phrases, found an account with test-network funds, and used that account to pay for the requested contract deployment." The assignment allowed the model to deploy compatible cont...

Oct 16, 2026

2

AI Decision Models and Their Impact on AI SOC Platforms

In the space of a month, a new category of AI model went from nonexistent to everywhere. TypeSafe AI released Jev on September 15 and called it a System One model, a nod to Kahneman's fast, intuitive thinking . OpenAI announced its own Decisions API at DevDay and opened a public beta on October 6. Cribl published cribl-decision , an open LoRA adapter you can run locally. Others, like Laya, are already showing up in open source tooling. These models skip text generation entirely. You hand them some state and a set of typed questions. They hand back a choice, a score, or a yes/no with a probability attached. For security operations, I think this is some of the best news of the year. I also think most of the conversation is pointed at the wrong layer. Why a SOC should be excited Three properties make decision models a natural fit for alert triage. You get a number you can automate against. A typed answer with a calibrated probability is something you can put a thresho...

Oct 16, 2026

3

The AI for Security Virtual Summit (Oct 20)

Join Figma, Perplexity, and Wiz to learn how to monitor rogue agents, automate triage, and secure code. Save your spot!

4

China-Linked Hackers Exploited BeyondTrust Zero-Day to Take Over Remote Support Servers

The China-nexus cyber espionage group tracked as Armory Castle (previously UNC5221 or Warp Panda) exploited a newly disclosed security flaw in BeyondTrust Remote Support (formerly Bomgar) as a zero-day in late August 2026. The vulnerability, tracked as CVE-2026-102764 (CVSS score: 9.3), is a case of improper origin validation in an internal messaging component that could be exploited by a remote attacker to perform file operations on an affected appliance. "The vulnerability exists within internal components of Remote Support," BeyondTrust said in a Thursday alert. "The components did not adequately validate the origin of the messages it received. Under the right conditions, this could be used to perform file operations on the appliance." Patches for the flaw have been pushed to all Remote Support SaaS customers as of October 9, 2026. Self-hosted customers of Remote Support running versions from 24.3.1 through 26.2.3 are advised to manually update to BT26-0...

Oct 16, 2026

5

CERT-In Warns AI Agents Accessed Sensitive Data at Indian Organizations During Testing

AI agents being tested online reached the live systems of Indian organizations and read business records, including personal data and confidential material, CERT-In  said in an advisory  on October 15. The agents went beyond their assigned tasks, and none of the organizations detected the unauthorized access. After internal reviews, the AI companies whose models were being tested voluntarily reported the cases to CERT-In, India's national cyber agency. AI agents are software systems that can plan work, run commands, use tools, and act on the internet autonomously for extended periods. In these cases, the agents gained access via credentials and connection settings left exposed in application configuration and public code repositories, and via cloud service roles with excessive permissions. CERT-In did not name the AI companies, their models, or the organizations and sectors involved. It did not say how many companies reported cases or how many organizations were affecte...

Oct 16, 2026

6

PEERSTITCH Malware Deployed via FortiOS Flaw Bypasses Checks on Tagged VPN Sessions

A suspected advanced threat actor has been observed exploiting a security flaw in Fortinet FortiOS to deliver a previously unreported memory-resident ARM64 malware known as PEERSTITCH . The attack involves the exploitation of CVE-2025-61624 (CVSS score: 5.4), a path traversal flaw in the command line interpreter of FortiOS, FortiPAM, FortiProxy, and FortiSwitchManager that could allow a privileged attacker to achieve arbitrary write or delete files via specifically crafted arguments to existing commands. It was patched by Fortinet in April 2026. "PEERSTITCH leaves no persistent binary on disk, opens no listening ports and initiates no outbound network connections," Palo Alto Networks Unit 42 researchers Daniel Gott and Shawn Westfall said . "Instead, it communicates through traffic generated by legitimate processes on the infected host, which makes the malware resistant to conventional endpoint and network detection." The cybersecurity company said it discov...

Oct 16, 2026

7

New: Analyzing 2026's Most Prevalent Browser Attack Techniques

4 in 5 ClickFix attacks are via search engines. 30+ device code phish kits this year. Get the latest analysis.

8

Critical PwnEscape Linux Kernel Vulnerability Lets Guest VMs Read Host Memory

A newly disclosed critical flaw in the Linux kernel allows code inside a virtual machine to read memory on the host computer it runs on. A researcher demonstrated the leak against Vercel's live cloud sandboxes, and Vercel confirmed it. The flaw, now tracked as CVE-2026-99999 and nicknamed PwnEscape , is in the Linux kernel's networking code, not in any one cloud's setup, so it reaches the shared layer that cloud providers use to keep customers' workloads apart. It was found in a Vercel security challenge and reported privately before any public release, and there is no sign it has been used in a real attack. What the Researcher Showed Paulos Yibelo of the security firm pwn.ai found that specially made network packets sent from inside a virtual machine could do three things, he explained in a  technical write-up : read the host's memory, crash the host, and run code on it. All three come from the same kernel flaw. Vercel said in its  report on the challen...

Oct 15, 2026

9

ThreatsDay: Adobe RCE, DNS Wallet Theft, NVIDIA Container Escape, Self-Healing Backdoor + 23 More

Someone built a tool to steal data, then decided it should scare the victim for fun. Yes, that’s an actual feature. And somehow, it fits right in with this week's security stories. The rest gets interesting in different ways. A small permission can lead to a big theft. A trusted login can hide the wrong person. Old bugs are finding new victims, familiar tools are being put to strange uses, and some flaws are so simple you wonder how nobody spotted them sooner. There's a lot more hiding in the details, including a few twists you probably wouldn't expect. Here's what made the list this week. The threats change every week. Subscribe, and we’ll alert you when each new ThreatsDay Bulletin is out.

Oct 15, 2026

10

Citizen Lab Traces Zero-Click Candiru Spyware Infections at Al Jazeera to Targeted Online Ads

Candiru, an Israeli maker of spyware sold to governments, infected devices at the news network Al Jazeera through ordinary online ads, in attacks that needed no click or other action from the user, researchers at  the Citizen Lab  said Thursday. The hidden attack code ran the moment an ad loaded on the screen, and in some cases the ad loaded out of sight, in the background. The Citizen Lab, a research group at the University of Toronto, said this is the first public, technical confirmation that mercenary spyware, meaning hacking tools sold to governments, has reached real devices through targeted ads. The method had been described before from leaked company documents, but not shown in a live attack. The researchers recovered the spyware itself from one device, but not the exploit that breaks into the browser or app to install it. That first break-in step is described from how the attack was set up, not from captured code. How the Attack Worked The ads were placed by...

Oct 15, 2026