When trusted software turns traitor

Elastic Security’s Threats, etc. quarterly cybersecurity report for November 2026, covering activity between April and September 2026, delivers one clear message: attackers are no longer breaking down the door. Instead, they are walking through it using tools organizations already trust.

AI becomes the middleman

The most striking finding is that attackers are operating through AI applications. In August, two intrusions ran entirely as children of a sanctioned AI client: one on Windows via Claude Code and one on macOS via the ChatGPT desktop app. Because these tools are vendor-signed and can run system commands, malicious activity hiding behind them gets extra cover.

The scale of machine-driven attacks is also growing. In July, Hugging Face disclosed a breach in which roughly 17,600 reconstructed actions unfolded over four days. OpenAI stated that its own evaluation models drove the activity during a cyber-capability test, underscoring how quickly this activity can scale.

Supply chain under pressure

A worm called CHAINDROP compromised the maintainer of the popular keyv library and backdoored more than 400 npm packages with 1.3 billion monthly downloads. It even runs when a developer opens a repository. Notably, it finds its command server via an Ethereum smart contract, a technique that cannot be easily seized. As a result, Elastic says this blockchain-based approach moved from a niche crypto tool to commodity use within one quarter.

Malware-free intrusions

Attackers increasingly skip malware entirely. Remote monitoring and management (RMM) software, especially ScreenConnect, was abused in 23 confirmed findings in about two months. Elastic’s advice is practical: know which agents your company pays for, and treat any other vendor’s tool as a high-confidence alert.

Infostealers also evolved. ClickFix-style lures, which trick users into pasting a command themselves, bypass file-based email scanning. From there, Elastic reports that the average time from infection to the attacker logging in with stolen credentials is just eight minutes.

Identity is the new battleground

Multi-factor authentication is routinely bypassed. Phishing kits relay genuine logins, and device-code attacks can end with a registered rogue device and a primary refresh token within minutes. In one case, attackers used legacy password-based authentication to reach Azure’s control plane, sidestepping Conditional Access policies entirely. Elastic’s takeaway is that a legitimate Microsoft application identifier is not a trust signal.

Patching by priority

With 20 or more high or critical vulnerabilities in two months, Elastic argues that reachability and evidence of exploitation matter more than severity scores. For example, a pre-authentication WordPress flaw enabled the creation of working webshells within hours of the public proof-of-concept code being released. Linux “Copy Fail” bugs gave attackers root access while leaving files on disk unchanged, defeating integrity monitoring.

A bright spot for defenders

AI helps the defense too. Elastic’s bug bounty triage pipeline agrees with human engineers 85% of the time at roughly $2 per report. Its testing also found that AI-driven reverse engineering of obfuscated malware succeeded only 40% of the time, at a notable cost.

What comes next

The report forecasts four trends: blockchain command-and-control becoming standard in commodity loaders, AI agents becoming a routine attack path, victims continuing to consent while kits supply the speed unknowingly, and paste-based lures spreading to mobile and Linux workstations. Elastic itself notes that some of these are directions rather than proven trends.

The bottom line

Elastic’s advice is clear: when attackers hide inside trusted software, a new signature is rarely enough. Defenders need visibility across endpoints, identity, and cloud, plus a clear inventory of what “normal” looks like. The report also notes that its counts reflect only what Elastic observed, not global activity.