Cyberattack forces Ahold Delhaize offline

Hi there, this is your daily ☕️ Cyberpresso.

Cyberattack knocks Ahold Delhaize offline LINK

• Ahold Delhaize, the Dutch grocery owner of US chains Food Lion, Giant Food, and Stop & Shop, pulled many IT systems offline after a cyberattack first disclosed on October 12 that hit most of its brands in Europe and the United States.
• CEO Frans Muller called the shutdown of websites and apps a precaution against "malicious actions of others," and the company said it has found no sign that attackers stole sensitive customer, worker, or partner data, with physical stores staying open.
• Albert Heijn's site and app are back, with home deliveries expected to resume on Friday, while US brands like Hannaford and Martin's should return within days; the company has told Dutch authorities and filed a police report.

Hackers hijack Plex servers to mine crypto LINK

• Hackers are breaking into internet-exposed Plex Media Server systems and installing crypto-mining software, striking just one day after a researcher published a working exploit for a group of four flaws called Plex2Shell.
• Chained together, the bugs (lead flaw CVE-2026-96652) let attackers run code on a vulnerable server without login details; a honeypot caught the first in-the-wild attacks on Saturday, using the public proof-of-concept.
• Plex fixed the flaws and told owners to upgrade back in September, well before the attacks; intruders planted an XMRig miner and later proxy software that could route traffic through home internet connections.

Ransomware crews exploit Palo Alto VPN flaw LINK

• Ransomware crews are exploiting a Palo Alto Networks GlobalProtect VPN flaw to skip multi-factor authentication and break into corporate networks, in one case stealing credentials straight from a domain controller's virtual disk on VMware systems.
• The bug, CVE-2026-0257, affects GlobalProtect portals and gateways in certain PAN-OS and Prisma Access setups that enable authentication override cookies; devices fail to verify who made the cookie, so attackers can forge one and connect without logging in.
• Unit 42 tied the active exploitation to the Qilin and Inc ransomware groups, with investigations running since early June across healthcare, finance, and government; fixes include upgrading affected devices, disabling override, or using a dedicated override certificate.

F5 rushes emergency fix for NGINX flaws LINK

• F5 has shipped emergency, out-of-schedule patches for three critical flaws in its NGINX web traffic software, each scored near the top of the severity scale, that could let attackers crash servers, run code, or knock sites offline.
• The bugs, led by CVE-2026-15176, hit NGINX Plus, Open Source, Ingress Controller, Gateway Fabric, and Instance Manager; F5 found no sign they are being exploited yet and urges customers to upgrade to fixed versions.
• Remote code execution becomes far likelier when ASLR memory protection is off, so F5 suggests keeping it on and offers config workarounds such as disabling asynchronous file I/O and server-side includes, and using named regex captures.

Hackers hijacked AOMEI updates for backdoors LINK

• Hackers tracked as UNC7017 hijacked AOMEI's update system between April and July 2026, briefly swapping download settings to push hidden backdoors to users of the AOMEI Backupper and AnyViewer tools.
• Rather than altering binaries, the attackers edited two config files to point downloads at malware and supply a matching checksum, slipping it past the app's checks without needing any CVE or exploit.
• Victims spanned IT, telecom and university targets across eleven countries, receiving the CROSSROAD downloader and SHINYROAD backdoor; AOMEI says it fixed its servers, but that cleanup does not remove backdoors already installed on infected machines.

Strategies & Tactics

> Bring Your Own Vulnerable Driver: What Detection Teams Should Actually Hunt: Split the vague "EDR killer" label into four distinct attacks and hunt each on its own telemetry, since one driver-load rule leaves three paths silent.

> Frozen Privilege: Remote Code Execution in Snowflake-Managed PostgreSQL: Researchers showed how one minor database permission could be chained into full server control of Snowflake-managed PostgreSQL, which Snowflake has since fixed server-side.

> Rapid7 Analysis: NetScaler intrusions chaining CVE-2026-88771 and CVE-2026-88779: Attackers combine two NetScaler flaws to gain root access, plant hidden web shells and map internal networks, exposing credentials and configuration to theft.

> Leaked API Keys in LLMs: How AI Agents Expose Secrets: Scan an AI coding agent's file reads and tool calls to block API keys before they reach a language model, since leaked keys get memorized and resurface.

Other news you might like

• Documents Reveal Military Plan for Agentic AI That Hacks U.S. Adversaries LINK

• Authentication Bypass, Remote Code Execution, and Database Read in Eptura Archibus (FIXED) LINK

• WordPress SEO plugin Rank Math compromised to backdoor websites LINK

• Siemens Software Center Vulnerability Exploited in ClickFix Malware Attacks LINK

• AI agents broke into 100+ sites for one hacker at $50 each, researcher says LINK

• Feds Suspect Ransom Negotiator Ran ShinyHunters Extortion Site LINK

• OVHCloud CLI repo compromised. Customer reports incident LINK

• Defense contractors in two NATO allies hacked in recent Russia-linked campaign LINK

Trending tools

Harden: installs the Agentic Integrity Foundation free, verifying coding-agent tool calls locally before they run to catch risky actions. LINK

VibeDefend: enforces security and business rules inside AI coding agents like Cursor, Copilot and Codex, catching vulnerabilities and malicious dependencies before the PR. LINK

Google Gemini 3.8 Flash and Cyber: fast, affordable models for agentic coding, multi-step reasoning, autonomous tasks, and cybersecurity work like vulnerability detection across long-horizon workflows. LINK

Replay QA Security Scan: automatically discovers user journeys from a live or localhost URL, runs interactive tests, and records runtime behavior to find bugs. LINK

chat-recall: searches your team's AI assistant conversations across six coding tools, strips passwords locally, and surfaces past work so you stop redoing it. LINK

Koreshield: inspects customer input, retrieved context, and proposed actions in AI support workflows, blocking untrusted content from becoming trusted instructions with recorded decisions. LINK

Trending papers & reports

Hidden backdoor activation shows attackers can hijack an already-deployed AI classifier they can't touch, tricking it into a secret second task like sorting faces by demographic group, using only crafted inputs. LINK

AI agent security catches sneaky attacks that hide malicious instructions across multiple steps, connecting earlier clues to later actions and spotting threats more accurately than existing defenses while adding new detection skills without forgetting old ones. LINK

Safety filter blind spots show that language models often understand a harmful request yet still comply when it's worded more concretely, with refusal rates dropping ~18 percent even after the model grasps the intent. LINK

See you tomorrow for a new dose of ☕️ Cyberpresso!