Digital forensics and incident response (DFIR) teams are facing a dilemma: Attackers are moving faster and generating more alerts, but investigators are being given less time to establish what happened.
Agentic artificial intelligence may offer a solution. It promises to accelerate investigations by automating repetitive forensic tasks while preserving human control over evidence and conclusions.
During a recent SC Media webcast sponsored by Exterro, Robert Bond, senior product marketing manager at Exterro, and Daniel Gugliotta, technical solutions engineer at Exterro, explained how agentic AI is changing investigative workflows.
From AI assistance to agentic investigation
Generative AI can summarize evidence investigators have already collected, but the investigators need to feed it the data. Agentic AI goes further by taking action on its own, initiating authorized searches, examining results and coordinating subsequent investigative steps.
In other words, AI agents can save analysts the trouble of manually correlating endpoint logs, process activity, user accounts and suspicious files across multiple systems.
Instead of navigating separate tools, investigators can now describe a goal in natural language, specify the endpoint and time window, and authorize AI-assisted forensic workflows.
Connecting AI to forensic capabilities
Exterro's ARMOUR for FTK uses the Model Context Protocol (MCP) to connect supported AI models to FTK Central's forensic capabilities.
An investigator submits a request through an approved AI environment. A governed MCP server communicates with FTK Central, which directs endpoint agents to execute authorized forensic commands and return results.
The system complements existing EDR and XDR tools by providing deeper forensic analysis beyond security alerts and telemetry.
Yet investigators retain ultimate control over scope, permitted actions and evidence collection. The architecture also provides job trails and auditing information to document investigative activity.
"The keyword is 'governed' here," explained Gugliotta.
Accelerating triage, collection and reporting
Consider an unusual PowerShell execution on a finance employee's laptop. The investigator will need to establish who was logged in, what launched PowerShell, whether sensitive files were accessed and whether those files left the organization.
ARMOUR can coordinate searches of Windows event logs, registry artifacts, process histories, browser activity and removable-storage records. Investigators can also search for indicators of compromise across endpoints.
During the webcast, Gugliotta demonstrated AI-assisted workflows that identified PowerShell activity, investigated possible exfiltration, collected evidence and generated reports.
In one example discussed by Bond, a document investigation that previously required five days to reach an initial write-up reached that milestone in 90 minutes using ARMOUR.
Such acceleration can help investigators preserve evidence before logs disappear and meet demanding incident-response deadlines.
Preserving forensic rigor
Speed cannot replace evidentiary standards. An AI-generated timeline showing that a confidential document was opened shortly before a USB drive was connected does not establish that the document was copied.
Investigators must verify underlying records, examine file hashes and provenance, preserve chain of custody and distinguish confirmed findings from unresolved suspicions.
"I want to be able to look at that and verify that and report on that because that's going to ultimately come down to me, and I can't just defer to the AI in that regard," said Gugliotta.
Organizations should establish approved investigative workflows, test automated findings against known cases and require human authorization when investigations expand beyond their original scope.
Agentic AI's value is not autonomous judgment. It is reducing the time spent finding evidence so qualified examiners can spend more time validating it and making defensible decisions.