Your AI agent’s system prompt is not a security control

2026-09-03T08:51:40Z23d370a58155383332a14820a4a859387ddfcd2e13a604570f38b9e5aafb1b47
CVE-2026-62911CVE-2026-83548CVE-2026-83549CVE-2026-9586AI securityMicrosoft ExchangeOAuth phishingSQL injectionSSRFSangoma SwitchvoxSonicWall SMA 1000active exploitationauthentication bypassmemory integritythreat intelligencevulnerability managementzero-day

What happened

Help Net Security coverage highlights actively exploited and widely exposed vulnerabilities, including zero-day attacks against SonicWall SMA 1000 appliances, exploitation of a Sangoma Switchvox SQL injection flaw, and nearly 22,000 unpatched Microsoft Exchange servers affected by an authentication-bypass vulnerability. Additional articles discuss OAuth consent phishing, AI-agent authorization risks, Windows memory-integrity changes, and threat-intelligence ingestion practices.

Why it matters

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

Evidence and limitations

Source ID
helpnetsecurity
Record identifier
23d370a58155383332a14820a4a859387ddfcd2e13a604570f38b9e5aafb1b47
Enrichment time
2026-09-03T08:51:40Z
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.