Claude, Codex, and Hermes installed unowned code inside corporate networks

2026-08-27T20:51:38Zb06d5cbbf2a3078acf4b556c6a4ada891569062697aac6e9e267aaf32a7ad501
AI securityBMCLLM prompt injectionMicrosoft ExchangeZoomaccount takeoveractive exploitationbrowser fingerprintingcorporate networkscredential theftcybercrimedata exfiltrationmacOSmalwarenation-state activityphishingremote accessserver securitysupply-chain attacksurveillancezero-day

What happened

Ars Technica security feed covering active exploitation, supply-chain compromise, credential exfiltration, AI-enabled attacks and prompt-injection failures, device and server takeover vulnerabilities, browser fingerprinting, phishing, surveillance, and account-takeover defenses. The most urgent reports involve an actively exploited macOS screen-sharing flaw, a maximum-severity Microsoft Exchange vulnerability exploited by Kremlin-linked actors, compromised AI packages leaking terabytes of credentials, and vulnerabilities enabling remote device or server control.

Why it matters

A reviewed impact interpretation has not been published for this record.

Evidence and limitations

Source ID
arstechnica_security
Record identifier
b06d5cbbf2a3078acf4b556c6a4ada891569062697aac6e9e267aaf32a7ad501
Enrichment time
2026-08-27T20:51:38Z
AI-assisted enrichment
Yes

This record may overlap with other records. Its enrichment can be incomplete or wrong, and machine assistance was used. Validate consequential decisions against the linked source and your own environment.