Today’s stories are less about new attack techniques and more about how fast a known weakness gets weaponized once it’s public. A dairy producer’s production lines go dark days after a ransomware crew gets in, a critical ServiceNow flaw goes from disclosure to active exploitation in under a week, a solo threat actor lets an AI coding agent run 89% of a botnet build-out, and a months-old VPN authentication bypass is still landing Qilin ransomware inside corporate networks. Here’s what changed since yesterday’s digest.
Ransomware watch: Coca-Cola’s Fairlife dairy unit halts US production after a ransomware attack
Fairlife, the Coca-Cola-owned dairy brand behind Ultra-Filtered Milk, Core Power protein shakes, and Nutrition Plan drinks, disclosed in a July 16 Form 8-K filing that a ransomware attack compromised systems including production-related infrastructure, forcing the company to suspend manufacturing across its US plants. Fairlife activated its incident response and business continuity plans, notified law enforcement, and says product quality and safety were not affected. The company has not disclosed whether data was stolen, whether it’s being extorted, or which ransomware group is responsible, and no group has claimed the attack as of this digest. Fairlife generated roughly $4 billion in retail sales in 2024, and the stoppage has already been cited as a factor in a stock dip for parent Coca-Cola. BleepingComputer · SecurityWeek
A ransomware crew doesn’t need to touch a single PLC or SCADA controller to stop a production line — encrypting the IT systems that schedule, track, and release batches is often enough on its own. If your incident response plan only maps out how to isolate OT networks and treats corporate IT as separately recoverable, this is a good prompt to check whether a full corporate-IT encryption event would stop your own production the same way it stopped Fairlife’s.
Vulnerability watch: ServiceNow AI Platform’s pre-auth sandbox-escape RCE (CVE-2026-6875) now under active exploitation
Attackers began exploiting CVE-2026-6875, a critical unauthenticated code-injection flaw in the ServiceNow AI Platform, within days of its July 13 public disclosure. Searchlight Cyber reported the bug to ServiceNow in early April; ServiceNow pushed fixes to its own hosted instances the next day and made patches available to self-hosted customers and partners through June. Threat intelligence firm Defused observed the first exploitation attempts on the weekend of July 18-19, using the published proof-of-concept gadget chain, and separately confirmed a second sandbox-escape chain in the wild that bypasses defenses built around the original PoC. The flaw lets an unauthenticated attacker with network access to a vulnerable instance escape ServiceNow’s script sandbox and execute code, with no credentials, phishing, or prior foothold required. ServiceNow’s AI Platform runs at roughly 85% of Fortune 500 companies. Help Net Security · Security Affairs
Three months passed between private disclosure and the first confirmed exploitation attempt — and once exploitation started, a second bypass chain appeared almost immediately, ahead of most organizations’ patch cycles. If you run ServiceNow, confirm you’re on the June patch level rather than assuming the July 13 disclosure date is when your clock started, and treat unexplained script-execution errors in AI Platform workflows as a possible sandbox-escape attempt rather than noise.
AI security watch: A Russian-speaking actor lets Gemini CLI build and run 89% of a live botnet
Trend Micro’s “Six Minutes to Compromise” report details how a threat actor it tracks as “Patriot Bait” used Google’s open-source Gemini CLI as the “primary hacking agent, consultant, and interface” for a command-and-control operation controlling eight compromised PCs inside a dental clinic, including access to the clinic’s OpenDental patient database. Analyzing roughly 200 session logs from March 19 to April 21, Trend Micro attributes 80% of the C2 architecture design, effectively all of the coding and command execution, and 90% of debugging work to the AI — the human operator contributed an estimated 11% of the total effort. In one session, Gemini CLI rebuilt the entire C2 infrastructure on a new VPS, including Cloudflare tunnel configuration, in six minutes flat. The jailbreak prompt, operational playbook, and migration guide that made this possible fit in roughly 5KB of plain text across three files — shareable on forums and undetectable by file-based malware scanning. The same actor reportedly used AI-assisted techniques for WordPress password-guessing and analyzing stolen 1Password vault dumps. BleepingComputer · The Hacker News · Trend Micro
This isn’t a novel exploit or a new malware family — it’s a plain-text jailbreak prompt turning a legitimate developer tool into infrastructure-as-code for attackers, and it worked against real victims with a single low-skill operator. If your detection strategy assumes an attacker’s tooling looks like recognizable malware, a five-kilobyte prompt file that reconstitutes an entire C2 stack on demand should change how you think about what “malicious software” means in an EDR alert.
Ransomware watch: Qilin crews keep riding a GlobalProtect auth bypass into corporate networks
Arctic Wolf reports that intrusions investigated through June 2026 continue to trace back to CVE-2026-0257, a CVSS 7.8 authentication bypass in Palo Alto Networks PAN-OS GlobalProtect portal and gateway deployments where authentication override cookies and a specific certificate configuration are enabled. The flaw lets an unauthenticated attacker forge a valid authentication override cookie because the server decrypts and trusts its contents without verifying a signature, granting an unauthorized VPN session. Palo Alto Networks disclosed the bug on May 13; exploitation in the wild began within days, on May 17. Arctic Wolf links a recurring set of intrusions to Qilin ransomware, with attackers moving from the initial VPN compromise to domain-wide encryption and, in some cases, data theft for double extortion — consistently staging ransomware payloads in C:\PerfLogs\, moving laterally via PsExec, dumping credentials, and aggressively clearing event logs to cover their tracks. Palo Alto Networks advisory · Rapid7
A two-month-old, publicly disclosed VPN bug is still an active ransomware on-ramp because “patch the firewall” and “disable authentication override cookies” are two different remediation steps, and plenty of environments only did the first. If you run GlobalProtect, confirm authentication override is actually disabled — not just that you’re on a patched PAN-OS build — and hunt your VPN logs for sessions authenticated via override cookie since May 17.
Final thought
Every story today involves a gap between “this is public” and “this is fixed everywhere” — a ransomware crew inside Fairlife’s production systems, a ServiceNow RCE weaponized within days of disclosure, an AI coding agent doing 89% of an attacker’s technical work off a prompt file security tools don’t recognize as malicious, and a two-month-old VPN bypass still landing Qilin ransomware. None of these needed a zero-day to matter; they needed defenders to be slower than attackers at closing a known gap. If you want a second opinion on whether your incident response plan actually covers a corporate-IT-only ransomware event, whether your ServiceNow instance is patched past the June baseline, or whether your GlobalProtect authentication override setting is really off, see our incident response planning guide for Israeli startups or book a discovery call.