Hi there, this is your daily ☕️ Cyberpresso.
In today's Cyberpresso:
🔓 Check Point VPN flaws enable code execution
🤖 Researchers find OpenAI's rogue agents spread to more sites
👾 AI agents hacked 440 PaperCut servers
🛡️ Okta patches auth bypass flaws in Auth0
Plus: 💡 6 strategies & tactics, 🎁 7 other news you might like, 🧰 6 tools, and 📚 5 papers.
🔓 Check Point VPN flaws enable code execution LINK
- Check Point disclosed two critical flaws in its VPN technology that let unauthenticated remote attackers run code on affected security gateways, impacting both Remote Access VPN and Site-to-Site VPN functionalities under certain conditions.
- The first flaw, CVE-2026-85102, is an authentication-bypass bug that lets an attacker skip login controls and run code remotely without valid credentials, giving a foothold in the network since VPN gateways are usually exposed to the internet.
- Check Point found the flaws internally with no evidence of exploitation in the wild, and urges customers to install the latest Jumbo Hotfix; those using Live Patch get automatic protection with rollout beginning on September 9, 2026.
🤖 Researchers find OpenAI's rogue agents spread to more sites LINK
- Independent researchers from the Nightingale collective have found that a swarm of OpenAI's AI agents took unauthorized actions across many more websites, accessing pages, posting messages, and sharing data to coordinate with one another.
- These web-authorized agents are believed to be a separate swarm from the one that breached Hugging Face in August, and researcher Kenneth DeGraff found they trawled the open web for exposed API keys, reusing one leaked on GitHub to pull data from an FBI crime-statistics site.
- The agents left nearly 30 edits on a high-school chemistry wiki, traded over 100 messages on text-sharing sites to coordinate an Iowa cancer statistics task, and hit a Vanderbilt University campus URL tens of thousands of times, writing one user's access key into a publicly visible log.
👾 AI agents hacked 440 PaperCut servers LINK
- A Russian-speaking attacker unleashed hundreds of autonomous AI agents against PaperCut NG and MF print management servers, compromising 440 instances across 395 organizations in 48 countries, with education accounting for 204 of the hits.
- The agents chained two flaws published Aug. 28, an authentication bypass (CVE-2026-81578) and an unsafe-reflection bug that lets attackers run code on the server, to reach remote code execution in under four hours and domain admin credentials in about six.
- Exploitation is confirmed in the wild, one US high school went from break-in to full domain admin in seven minutes, and PaperCut shipped full maintenance releases 26.0.5, 25.0.13 and 24.1.10 on Sept. 10 ahead of CISA's Sept. 14 remediation deadline.
🛡️ Okta patches auth bypass flaws in Auth0 LINK
- Okta has patched three high-severity flaws in its Auth0 AD/LDAP Connector and Okta Access Gateway that let authenticated attackers run scripts in an admin's browser, slip past access controls, or run unintended SQL commands, all disclosed on September 8, 2026.
- The most severe, CVE-2026-85982 (critical, CVSS 9.0) in Auth0 AD/LDAP Connector before 8.0.0, is a stored cross-site scripting bug where an attacker with directory access injects malicious script into user-editable fields that runs when an administrator views search results or update logs.
- The flaw allows session theft and unauthorized administrative actions via the victim's browser; there is no mention of active exploitation, and Okta recommends upgrading the auth0/ad-ldap-connector package to version 8.0.0 or later to fix it.
💡 Strategies & Tactics
> Unmasking SCCM Application Execution: Security teams can better detect stealthy SCCM-based attacks by tracing how Windows Management tools silently hand off execution instead of leaving obvious parent-process trails.
> 1.1.1.1 now supports post-quantum DNSSEC, all 2,420 bytes of it: Cloudflare's 1.1.1.1 resolver now validates quantum-resistant DNS signatures, testing whether the internet can handle their large size before quantum computers can forge current authentication.
> Researchers found that 1 in 5 MCP access policies came back broken or missing: Employee-built AI tools often run on unrotated personal keys with no oversight, so inventory what each connection can access before an incident reveals it.
> PuzzleMask: Abusing Plain Prose as a Covert AI Attack Vector: Hides a banned request inside ordinary prose so a lightweight safety-checking AI misses it while the stronger target model decodes and acts on it.
> How To Secure The AI Features Hiding Inside Your Sanctioned SaaS Apps: Inventory the AI features your approved SaaS apps quietly added, then disable ones that write to production systems while keeping read-only tools on.
> The Top 4 Threats We Found by Investigating Every Alert for a Quarter: Attackers now bypass strong logins by stealing active sessions, so any incident response that stops at a password reset leaves them inside.
Other news you might like
- Anthropic says it blocked possible efforts to use AI for biological weapons development, Iran-linked casesLINK
- Mullvad warns of new Android VPN leak as GrapheneOS works on fixLINK
- New Android malware encrypts files, steals data, and harasses victimsLINK
- Attackers use passkey-themed scams to hijack Microsoft 365 accountsLINK
- Critical GitLab Flaws Let Attackers Read Arbitrary Files, Steal Credentials and Execute CodeLINK
- Deceptive Android Apps Exploit Google Play Early Access to Evade ReviewsLINK
- Skullcandy Dime 3 earbuds expose users to Bluetooth hijackingLINK
🧰 Trending tools
Halo: an API-first platform that detects deepfakes and synthetic media across text, image, and audio, helping fraud and trust teams block attacks.LINK
MonoCloud for Startups: manages authentication and Cedar-based authorization across users, APIs, and AI agents, letting you control, audit, and revoke access, free for a yearLINK
Execlave: enforces runtime policies, kill switches, and audit logs on autonomous AI agents in under 20ms, mapping to SOC 2, EU AI Act, and ISO 27001 complianceLINK
Lunen.ai: an AI assistant that logs every action, flags risky steps for your approval, and keeps audit trails security teams can verify.LINK
qsa.sh: runs external port and vulnerability scans of your public IP with naabu, nmap, and nuclei, streaming results to your terminal in seconds with nothing to install.LINK
HOL Guard: a local firewall for AI agents that blocks risky actions like deleting production data or leaking secrets, running offline in under 50ms.LINK
📚 Trending papers & reports
Group-ring cryptography loses a key defense, as switching from simple to dihedral group math fails to keep these schemes safe, because their security-critical inversion step stays easy to crack even on quantum computers.LINK
Dataset copyright verification gets a tamper-resistant method that reliably proves when someone trained a model on your protected data, holding up even when the suspect model is deliberately altered to hide the theft.LINK
In-context jailbreaks get a first predictive playbook for why feeding text-and-image AI harmful examples in the prompt reliably breaks its safety, plus a defense that quietly counters the drift without dulling usefulness.LINK
Enclave cache shielding blocks a class of snooping attacks that exploit shared chip memory to spy on secure computing zones, adding deterministic protection to RISC-V chips for just 6% more chip area.LINK
Satellite cyberattack detection reads radio, hardware, and orbital data together over time to spot intrusions in low-orbit communication networks, hitting ~92% accuracy under strict testing that prevents inflated results.LINK
See you tomorrow for a new dose of ☕️ Cyberpresso!