Abnormal AI

Abnormal AI email threat investigation agents

Abnormal AI agents investigate difficult email threats with sandboxed code and draft detection heuristics from misclassification patterns.

Reported results

  • Tens of thousands of difficult cases per day
  • About 100 batch jobs per week

As stated by AWS. Not independently verified.

Platform
AWS Bedrock
Type
Company deployment
Source
AWS (aws.amazon.com)
Added
Oct 28, 2026

What the agent does

Abnormal AI routes the hardest email classification cases to inline agents that analyze threat intelligence and write sandboxed code before deciding whether to block a message. A separate analyst agent reviews misclassifications, identifies trends, drafts candidate heuristics for high-volume detection, and proposes improvements to machine-learning models. The live system handles tens of thousands of difficult cases per day, while the analyst workflow runs about 100 batch jobs per week in a no-egress sandbox.

Read the source on aws.amazon.com

Stack

Amazon Bedrock AgentCoreAgentCore Code InterpreterMachine LearningSandboxed Python

Related cases

Sumo Logic

Sumo Logic Mobot security analysis agent

Sumo Logic runs Mobot and specialist agents to interpret security telemetry, combine alerts, and suggest investigation steps.

20% latency improvement24% cost reduction
Security · Software & SaaS · AWS Bedrock
Druva

Druva DruAI cyber resilience agents

Druva built DruAI as a set of specialized agents for incident response, investigation, and cyber-resilience recovery workflows.

68% of support issues solved without human interventionMore than 3,000 users in four months
Security · Software & SaaS · AWS Bedrock
LoanPro

LoanPro loan servicing assistant

LoanPro servicing teams use a multi-agent assistant to summarize borrower accounts and recommend next actions during calls.

Up to 15% reduction in customer call timesInternal ticket volumes decreased by nearly 8%
Customer support · Software & SaaS · AWS Bedrock
MoEngage

MoEngage Merlin AI marketing agents

MoEngage runs Merlin AI agents to draft, analyze, adjust, and decide customer campaigns using live tenant-scoped platform data.

Campaign go-live 50% faster38% uplift in push notification click-through rates
Sales & marketing · Software & SaaS · AWS Bedrock
Cogent Security

Cogent Security cyber remediation agents

Cogent Security runs agents that investigate enterprise vulnerabilities, execute permitted fixes, and verify remediation.

75% lower mean time to remediate critical vulnerabilities5x vulnerability-triage throughput
Security · Software & SaaS · OpenAI
Greenlight Financial

Greenlight Financial Luna security operations agent

Greenlight Financial uses Luna to investigate security and financial-risk alerts, resolve routine cases, and escalate evidence to analysts.

More than 90% autonomous resolution rate3,000-4,000 cases processed per month
Security · Financial services · AWS Bedrock