Arnold Chan
@slazTrusted contributorCompletionist
2 followers2,099 downloads4,340 copies51 likes25,914 views
7,555 detections
Filters
Last updated
All Time
Detection languages
5,650
1,006
394
169
98
Categories
2,356
1,807
1,233
1,013
924
Platforms
4,191
1,622
1,250
676
676
Products / Services
2,820
1,003
991
720
485
MITRE Techniques
2,087
1,520
1,413
1,070
536
CVEs
38
33
24
22
21
IDS Classtypes
477
179
104
79
52
IDS Protocols
554
156
144
86
18
Detects phishing messages that deliberately mix writing systems/scripts/homoglyphs in the Subject line as ONE corroborating signal within a broader, multi-signal phishing pattern. Hardened against false positives: brand/product tokens (e.g. Microsoft, PayPal) are stripped before the Latin-script check so a lone brand mention in non-Latin correspondence doesn't count as mixed-script; a weighted score (script-mixing=2, homoglyph=3, high-confidence phishing vocabulary=2, threshold=6) means no two of the three signal categories can reach the threshold alone — all three must co-occur before the rule fires.
Detects phishing messages that deliberately mix writing systems/scripts/homoglyphs in the Subject line as ONE corroborating signal within a broader, multi-signal phishing pattern. Hardened against false positives: brand/product tokens (e.g. Microsoft, PayPal) are stripped before the Latin-script check so a lone brand mention in non-Latin correspondence doesn't count as mixed-script; a weighted score (script-mixing=2, homoglyph=3, high-confidence phishing vocabulary=2, threshold=6) means no two of the three signal categories can reach the threshold alone — all three must co-occur before the rule fires.
Detects phishing emails that mix writing systems or scripts in the Subject line, combined with phishing vocabulary.
Detects a multi-stage attack chain where an AI-agent's configuration file (e.g., mcp.json) is modified, followed immediately by the execution of a tool process (node/python) spawned by the agent, the subsequent spawning of a shell process, and finally an outbound network connection to a rare, non-standard domain. This sequence is indicative of AI agent tool poisoning, where malicious tool definitions are introduced to execute arbitrary commands and exfiltrate data. Coverts T1195.002, T1565.001, T1059
Detects anomalous, high-frequency, multi-channel probing of AI agent input surfaces. The rule identifies external actors sending instruction-injection style keywords across different communication channels (e.g., email, Teams, calendar) within a short window, which may indicate an attempt to identify and manipulate an organization's autonomous agents. Covers T1595.002
Detects phishing messages that deliberately mix writing systems/scripts/homoglyphs in the Subject line as ONE corroborating signal within a broader, multi-signal phishing pattern. Hardened against false positives: brand/product tokens (e.g. Microsoft, PayPal) are stripped before the Latin-script check so a lone brand mention in non-Latin correspondence doesn't count as mixed-script; the suspicious-URL signal requires a high-confidence credential/urgency term (not generic business vocabulary); sender risk requires the domain to be new to both the recipient and the organization; and at least three independent signal categories (not two) must corroborate the language-confusion signal before the rule fires.
This rule detects potential adversarial reconnaissance activities targeting AI agents via multiple communication channels (e.g., email, collaboration tools). It identifies patterns where a sender transmits messages containing common 'jailbreak' or 'instruction-override' phrases (e.g., 'ignore previous instructions', 'bypass approval') across at least three distinct channels within a 24-hour window, indicating a concerted effort to manipulate or probe agent logic.
This rule detects potential adversarial reconnaissance activities targeting AI agents via multiple communication channels (e.g., email, collaboration tools). It identifies patterns where a sender transmits messages containing common 'jailbreak' or 'instruction-override' phrases (e.g., 'ignore previous instructions', 'bypass approval') across at least three distinct channels within a 24-hour window, indicating a concerted effort to manipulate or probe agent logic.
This rule detects potential adversarial reconnaissance activities targeting AI agents via multiple communication channels (e.g., email, collaboration tools). It identifies patterns where a sender transmits messages containing common 'jailbreak' or 'instruction-override' phrases (e.g., 'ignore previous instructions', 'bypass approval') across at least three distinct channels within a 24-hour window, indicating a concerted effort to manipulate or probe agent logic.
Detects autonomous behavior where AI-driven processes attempt multiple network connections (SMB, SSH, HTTP, RDP, WinRM) to diverse targets following failed access attempts. The rule correlates initial failure activity with subsequent credential-related events and new lateral movement traffic from the same agent process within a short window, suggesting adaptive adversary pivoting.
Page 120 of 756