AI Agents Compromise 440 PaperCut Instances
A suspected Russian-speaking attacker unleashed a swarm of AI agents to exploit two PaperCut NG/MF flaws, breaching roughly 440 instances at 395 organizations globally. Nearly half of the victims were in education.
Powered by a Codex harness and DeepSeek model, the automated blitz hit 11 organizations in 26 seconds. Attackers swiped credentials from 280 victims, uncovered system or domain secrets at 147, and achieved full domain admin status at 12, including taking over one US high school in seven minutes.
In a classic case of machines ignoring their masters, some agents blatantly bypassed the attacker’s own 28-country geographic exclusion list.
Patch immediately to PaperCut versions 26.0.5, 25.0.13, or 24.1.10; lock down admin interfaces to trusted IPs; rotate exposed credentials; and hunt for suspicious logs, Active Directory changes, and DCSync activity. If compromised, nuke it and rebuild from a clean backup.
Adobe Patches Exploited Commerce Zero-Day
Attackers are exploiting an Adobe Commerce zero-day to remotely execute code and plant persistent malware.
Threat actors can leave malware outside the web root and potentially expose credentials, tokens, payment secrets, and other sensitive information.
I would investigate every credential, integration, and service the compromised system could access before declaring it contained.
Apply Adobe’s hotfix, hunt for malware, and rotate credentials.
Microsoft Fixes Nearly 1,000 Security Flaws
Microsoft’s September Patch Tuesday fixes nearly 1,000 vulnerabilities, including two actively exploited Windows zero-days.
I would prioritize patches based on active exploitation, exposure, and organizational risk rather than letting CVSS scores dictate the queue.
Identify affected systems, prioritize exploited flaws, and verify patches.
ChatGPT Vulnerability Hijacks AI Sessions
Check Point found that malicious instructions could secretly make ChatGPT execute attacker commands and relay data to another account.
The attack could abuse existing session permissions, including Gmail access.
Limit AI assistant connections to necessary data and systems to reduce the reach of successful prompt injection attacks.
Remove unnecessary AI permissions and require confirmation for sensitive actions.
AI Agents Breach Enterprise Network in 10 Hours
A threat actor used AI agents to complete an enterprise intrusion in under 10 hours, compressing roughly two weeks of work.
The AI agents worked in parallel to map systems, find credentials, access secrets, and move across cloud and CI/CD environments.
I think the biggest shift is the attack clock, as AI gives defenders far less time to detect and contain familiar attack techniques.
Automate containment to revoke credentials, terminate sessions, and isolate affected cloud resources.
BigBear Hijacks Microsoft 365 Sessions at 258 Organizations
A phishing service compromised Microsoft 365 accounts at 258 organizations by stealing authenticated sessions to bypass MFA.
BigBear steals sessions and uses residential proxies to appear legitimate.
I think this is another reminder that a successful MFA challenge does not automatically make the resulting session trustworthy.
Use phishing-resistant FIDO2 or WebAuthn and require managed devices through Conditional Access.