Back to today’s briefing

Story perspectives

AI Defense Confidence Lags as Cyber Threats Surge

9/26/2025

50 4 Full Breakdown

Summary

AI Defense Insights

A LevelBlue report: 51% confident in AI defense; 32% prepared for 2025 AI threats.
Verizon: AI use in cyberattacks has doubled; state-sponsored actors employ AI for phishing and malware development.
Australia: 30% of data breaches result from human error.
Secure Code Warrior launches Trust Agent to oversee AI coding tools.
SciGuard researchers aim to curb AI misuse in chemical sciences.

Summary

Noma Secures Salesforce

Noma Security uncovered ForcedLeak, a vulnerability in Salesforce's AI system Agentforce, scoring 9.4 on CVSS (Common Vulnerability Scoring System) and affecting Web-to-Lead users.
Attackers exploit indirect prompt injection by embedding malicious instructions in the Description field of the Web-to-Lead form.
Salesforce fixed the issue by enforcing a Trusted URL allowlist and re-securing an expired domain; users should audit lead data.

History

Historical Context

The rise in cyberattacks attributed to state-sponsored actors, particularly from China, has escalated since 2010, when Google reported significant breaches. By 2016, the U.S. Department of Homeland Security noted a 300% increase in such attacks. As organizations adapt, 51% now express confidence in AI defenses, highlighting the urgent need for robust cybersecurity measures amid evolving threats like the recent ForcedLeak incident.

1 of 3

AI Defense Insights

A LevelBlue report: 51% confident in AI defense; 32% prepared for 2025 AI threats.
Verizon: AI use in cyberattacks has doubled; state-sponsored actors employ AI for phishing and malware development.
Australia: 30% of data breaches result from human error.
Secure Code Warrior launches Trust Agent to oversee AI coding tools.
SciGuard researchers aim to curb AI misuse in chemical sciences.

1 / 3