Hackers exploit an Apple flaw to spread malware

Hi there, this is your daily ☕️ Cyberpresso.

Hackers exploit Apple flaw to spread malware LINK

• Hackers tied to Russia are abusing an Apple authentication flaw to push data-stealing malware, prompting CISA to add CVE-2024-27834 to its Known Exploited Vulnerabilities list after reports of active attacks in the wild.
• The medium-severity bug lets an attacker bypass Pointer Authentication, and Apple fixed it back in May 2024 across iOS, iPadOS, macOS Sonoma, tvOS, watchOS, and Safari after it was shown as a zero-day at Pwn2Own Vancouver.
• Researchers at Sekoia link the campaign to APT29, which used hacked sites, fake lookalike pages, and a supply chain attack to run malicious JavaScript, and federal agencies must patch by November 18.

DreamHost breach exposed FTP logins and webmail sessions LINK

• Web host DreamHost says attackers broke into its web-based File Manager and spent two months stealing customers' FTP and SFTP usernames and passwords before the intrusion was found on October 23, 2026.
• The thief exploited a flaw in the software running the File Manager, reaching login details for anyone who signed in from August 2026 onward; DreamHost has since blocked the access and patched the hole.
• Webmail systems may also have been touched, exposing session data and credentials for some users, though DreamHost found no sign that control panel logins or billing details were reached, and is offering a year of Equifax credit monitoring.

OpenAI's AI automates mass vulnerability discovery LINK

• OpenAI showed off an automated pipeline that used an upcoming cyber-focused AI model to find, confirm, and write patches for 948 local privilege escalation flaws in a widely used Linux distribution, costing $1.2 million in API fees.
• The five-stage agentic system reproduced over 1,700 vulnerabilities within 30 hours and built working proof-of-concept exploits for the privilege escalation bugs plus hundreds of information leaks before human researchers reviewed each result.
• OpenAI said the agents came up with at least 10 new Linux attack techniques, and argued patch generation must keep pace so that an outdated low-privilege account effectively hands a resourced attacker root access.

SonicWall won't patch retired tool LINK

• SonicWall says it will not fix two security flaws in its Global Management System software, version 9.5.2 and earlier, because the product has reached end-of-life, and is urging customers to move to its newer Network Security Manager.
• The more serious bug, CVE-2026-109593, rated 9.1, lets an attacker on the network read internal GMS files without logging in; the second flaw needs an account and allows database tampering through SQL injection.
• Neither flaw is reported as exploited, but a separate maximum-severity SonicWall bug patched last month, CVE-2026-102255, is under active attack, with Previdian counting about 2,540 attempts since early October, including dozens on Tuesday.

APT flaw lets attackers swap packages LINK

• A newly found flaw in APT, the package manager behind Debian and Ubuntu, lets an attacker who can tamper with HTTP traffic swap in an unverified package and run code with root privileges.
• Tracked as CVE-2026-109518, the bug abuses HTTP redirects and download retries to overwrite hard links with a chosen payload; Debian disclosed it yesterday, crediting researcher Max Justicz, and Ubuntu rated it High at 7.7.
• Both distributions have shipped fixes, including APT 3.0.4 for Debian 13 and a separate patched version for Debian 12; Ubuntu advises updating, briefly turning off retries, and downloading packages over HTTPS.

Strategies & Tactics

> Mi Scusi, My SCSI: Abusing iSCSI Target Server: Shows how skipping optional authentication on Windows iSCSI shared storage lets attackers impersonate clients and steal files, including domain controller credentials, bypassing file permissions entirely.

> The SOC Files 5: Ghost in the pipeline, or a fileless SharePoint breach by Storm-2603: A fileless attack ran a SharePoint webshell entirely inside the web server's memory, hiding over the Christmas holiday to steal data and plant stealthy backdoor accounts.

> AI agent detection signals that still work when agents look human: Catch disguised AI agents by watching behavioral tells like click timing and session swarms, since fingerprint and metadata signals are easily spoofed and failing fast.

> Standing Access: The Security Risk in Your Cloud: Replace leftover permanent cloud permissions with access granted only when a task needs it and removed when done, since forgotten credentials are attackers' easiest route in.

> Building incident playbooks for machine identity breaches: Build incident response plans specifically for stolen machine credentials, since human-focused playbooks fail when there's no user to ask and revoking a key can crash live services.

> The Quiet Phase: Why Active Directory Group Enumeration…: Monitor who reads Active Directory admin-group membership, not just who changes it, because attackers map privileged accounts weeks before escalating, and decoy groups make such reconnaissance a reliable alarm.

Other news you might like

• GhostMap Android flaw lets malicious app trigger persistent boot loop LINK

• French bank confirms cyber heist, hackers suspected to have stolen an estimated $133 million LINK

• India flags Chinese-linked AI tools in hacking campaign against financial firms LINK

• NIKKE PC game launcher hacked to spread remote access, data-stealing malware LINK

• DigiCert Confirms Data Breach Exposing Customer Details and Sensitive Identity Documents LINK

Trending tools

CtrlOps: a local-first desktop app for managing Linux servers with an AI-assisted terminal, real-time monitoring, and agentless SSH access across all your infrastructure. LINK

Fraud Blocker: filters bots, click farms, and fake clicks from your ad campaigns so you only pay for real traffic and genuine leads. LINK

Axari: an AI security teammate that handles recurring tasks and goals inside Slack or Teams, cutting time spent managing security workflows. LINK

face2social: reverse image search that matches a photo across Instagram, Facebook, LinkedIn, TikTok, and X to verify identities and spot scammers before trusting them. LINK

Plango: monitors AI coding agents on your Mac, guarding secrets and files, blocking suspicious network activity, and scanning projects for vulnerabilities locally. LINK

chat-recall: unifies AI assistant histories from Claude Code, Codex, Cursor, and OpenCode into one searchable log, redacting secrets and flagging leaked keys locally. LINK

Trending papers & reports

Private AI watermarks let companies measure how often language models are being used across a population without tracking any single person's text, resolving the privacy fight that has made watermarking hard to deploy. LINK

Backdoor cleanup spots hidden sabotage triggers planted in language models and scrubs them out without the costly full retraining or spare clean model that current defenses demand, keeping compromised systems usable. LINK

Local security bots can turn a bug description into a crash-triggering test using an open model small enough to run on one consumer GPU, with even a tiny version handling most easier cases and a 27B model needed for the hardest. LINK

Repository context for AI coders turns out to barely improve code security even as it boosts functionality, with coding assistants spotting a project's safeguards in just ~27 to 36% of attempts. LINK

Fragile safety guardrails in chatbots often hinge on a few vulnerable spots, so this training method spreads safety across the whole system, cutting some attack success rates by up to 90% while keeping capabilities intact. LINK

See you tomorrow for a new dose of ☕️ Cyberpresso!