M365 Copilot Failed Authentication Patterns: endpointWeb ApplicationNoneversion:4
Detects M365 Copilot users with failed authentication attempts, MFA failures, or multi-location access patterns indicating potential credential attacks or account compromise. The detection aggregates M365 Copilot Graph API authentication events per user, calculating metrics like distinct cities/countries accessed, unique IP addresses and browsers, failed login attempts (status containing "fail" or "error"), and MFA failures (error code 50074). Users are flagged when they access Copilot from multiple cities (cities_count > 1), experience any authentication failures (failed_attempts > 0), or encounter MFA errors (mfa_failures > 0), which are indicators of credential stuffing, brute force attacks, or compromised accounts attempting to bypass multi-factor authentication.
M365 Copilot Session Origin Anomalies: accessWeb ApplicationNoneversion:4
Detects M365 Copilot users accessing from multiple geographic locations to identify potential account compromise, credential sharing, or impossible travel patterns. The detection aggregates M365 Copilot Graph API events per user, calculating distinct cities and countries accessed, unique IP addresses, and the observation timeframe to compute a locations-per-day metric that measures geographic mobility. Users accessing Copilot from more than one city (cities_count > 1) are flagged and sorted by country and city diversity, surfacing accounts exhibiting anomalous geographic patterns that suggest compromised credentials being used from distributed locations or simultaneous access from impossible travel distances.
M365 Copilot Jailbreak Attempts: endpointWeb ApplicationNoneversion:6
Detects M365 Copilot jailbreak attempts through prompt injection techniques including rule manipulation, system bypass commands, and AI impersonation requests that attempt to circumvent built-in safety controls. The detection searches exported eDiscovery prompt logs for jailbreak keywords like "pretend you are," "act as," "rules=," "ignore," "bypass," and "override" in the Subject_Title field, assigning severity scores based on the manipulation type (score of 4 for amoral impersonation or explicit rule injection, score of 3 for entity roleplay or bypass commands). Prompts with a jailbreak score of 2 or higher are flagged, prioritizing the most severe attempts to override AI safety mechanisms through direct instruction injection or unauthorized persona adoption.
M365 Copilot Information Extraction Jailbreak Attack: endpointWeb ApplicationNoneversion:5
Detects M365 Copilot information extraction jailbreak attacks that attempt to obtain sensitive, classified, or comprehensive data through various social engineering techniques including fictional entity impersonation, bulk data requests, and privacy bypass attempts. The detection searches exported eDiscovery prompt logs for extraction keywords like "transcendent," "tell me everything," "confidential," "dump," "extract," "reveal," and "bypass" in the Subject_Title field, categorizing each attempt by extraction type and assigning severity levels (CRITICAL for classified/proprietary data, HIGH for bulk extraction or privacy bypass). Prompts are further analyzed for compound risk patterns such as "Confidential+Extraction" or "Bulk_Request+Bypass," filtering out low-severity cases to surface the most dangerous attempts to exfiltrate sensitive organizational information through AI manipulation.
M365 Copilot Impersonation Jailbreak Attack: endpointWeb ProxyNoneversion:6
Detects M365 Copilot impersonation and roleplay jailbreak attempts where users try to manipulate the AI into adopting alternate personas, behaving as unrestricted entities, or impersonating malicious AI systems to bypass safety controls. The detection searches exported eDiscovery prompt logs for roleplay keywords like "pretend you are," "act as," "you are now," "amoral," and "roleplay as" in the Subject_Title field. Prompts are categorized into specific impersonation types (AI_Impersonation, Malicious_AI_Persona, Unrestricted_AI_Persona, etc.) to identify attempts to override the AI's safety guardrails through persona injection attacks.
M365 Copilot Non Compliant Devices Accessing M365 Copilot: endpointWeb ApplicationNoneversion:6
Detects M365 Copilot access from non-compliant or unmanaged devices that violate corporate security policies, indicating potential shadow IT usage, BYOD policy violations, or compromised endpoint access. The detection filters M365 Copilot Graph API events where deviceDetail.isCompliant=false or deviceDetail.isManaged=false, then aggregates by user, operating system, and browser to calculate metrics including event counts, unique IPs and locations, and compliance/management status over time. Users accessing Copilot from non-compliant or unmanaged devices are flagged and sorted by activity volume and geographic spread, enabling security teams to identify unauthorized endpoints that may lack proper security controls, encryption, or MDM enrollment.
M365 Copilot Application Usage Pattern Anomalies: endpointWeb ApplicationNoneversion:4
Detects M365 Copilot users exhibiting suspicious application usage patterns including multi-location access, abnormally high activity volumes, or access to multiple Copilot applications that may indicate account compromise or automated abuse. The detection aggregates M365 Copilot Graph API events per user, calculating metrics like distinct cities/countries accessed, unique IP addresses, number of different Copilot apps used, and average events per day over the observation period. Users are flagged when they access Copilot from multiple cities (cities_count > 1), generate excessive daily activity (events_per_day > 100), or use more than two different Copilot applications (app_count > 2), which are anomalous patterns suggesting credential compromise or bot-driven abuse.
M365 Copilot Agentic Jailbreak Attack: endpointWeb ApplicationNoneversion:5
Detects agentic AI jailbreak attempts that try to establish persistent control over M365 Copilot through rule injection, universal triggers, response automation, system overrides, and persona establishment techniques. The detection analyzes the PromptText field for keywords like "from now on," "always respond," "ignore previous," "new rule," "override," and role-playing commands (e.g., "act as," "you are now") that attempt to inject persistent instructions. The search computes risk by counting distinct jailbreak indicators per user session, flagging coordinated manipulation attempts.
