TL;DR
HeliosAI Investigation, the newest capability in the HeliosAI Agent and part of the Helios AIDR (AI Detection and Response) operating model, turns Tier 2 and Tier 3 analysts into faster experts, not just faster typists. Instead of answering only the questions you know to ask, it guides you through a cross-cloud and SaaS investigation the way a senior analyst would, shows the exact evidence and confidence behind every verdict, and turns hundreds of scattered forensic events into one clear narrative, all grounded in the same full-fidelity, normalized data lake that already cuts alert triage time by 90%. Three features are available now: HeliosAI Investigation, Natural-Language Log Search, and AI Summary for Workbench Events.
Security teams have an evidence problem. Not a headcount problem.
Every new AI assistant on the market promises to “investigate like an analyst.” In practice, their scope is much more narrow. They only answer the question you know to ask. That's not the same as investigating. A real investigation follows the case wherever it leads, checking the next lead before you'd have thought to ask about it. And the hardest part of an investigation is the question you didn't know you had. For a Tier 2 or Tier 3 analyst deciding whether to escalate a cross-cloud incident at 2 AM, a tool that only answers what you already suspected isn't investigating alongside you. It's just responding to you.
That's the problem HeliosAI Investigation solves. It's the newest capability in the HeliosAI Agent, powered by the same engine that already cuts alert triage time by 90%, and it's now generally available to all Mitiga customers. It extends the agent from triage into full investigation, grounded in real forensic evidence at every step.
Every verdict shows its evidence
Most AI assistants for security are built on whatever uncorrelated telemetry happens to be available, with raw logs, unstructured alerts, and uncontextualized data that was never designed to be reasoned over by a language model. Some vendors even market this as a feature, promising no data migration and no normalization. That's exactly backwards. The less structured the underlying data, the more room a model has to fill the gaps with a confident-sounding guess. As one Splunk report put it, generative AI can be “like that overconfident colleague who will never say they don't know.”
HeliosAI Investigation starts from the opposite premise that an AI is only as trustworthy as the evidence it's grounded in. For every conclusion it reaches, it shows exactly how it got there. You see the question it asked, the forensic evidence it pulled to answer it, and what specifically raised or lowered its confidence. Maybe that's the OAuth scope that made this look more suspicious, the IP reputation that made it look less so, or the baseline from the last 30 days that tipped the balance. This is cross-cloud identity and access forensics rather than a code-to-cloud posture score, the kind of evidence an incident responder actually needs to decide whether to escalate, and grounded in the same full-fidelity, normalized Cloud Security Data Lake behind that 90% cut in alert triage time.
And starting an investigation takes exactly one action. Just pick an alert with no prompt to write and no query to construct.
Our latest release adds three features to the HeliosAI Agent: HeliosAI Investigation, natural-language log search, and AI Summary for Workbench Events.
HeliosAI Investigation: the question you didn't know to ask
HeliosAI Investigation turns the agent from something that answers your questions into something that runs the case. It walks you through a case the way your most experienced analyst would walk a junior analyst through one, surfacing the next lead, connecting an identity's activity across services, checking whether this entity's behavior falls outside its normal baseline, and flagging the OAuth grant that lines up with yesterday's S3 access before you'd have thought to look for it. It reasons across AWS, GCP, Azure, Microsoft 365, Okta, and the SaaS applications your environment uses as one continuous investigation rather than five consoles you have to stitch together yourself.
Natural-language log search: when you already know the question
Guided investigation is the default experience, but not every question needs a full investigation. Sometimes, it's a specific IP, a specific entity, or a specific time window, and you just need the raw data, fast. That's what natural-language log search is for. Ask in plain English, with no query syntax and no schema to memorize, and get straight to the logs behind it. It's the fast lane for when you already know the question, sitting right next to the tool that helps you when you don't.
AI Summary for Workbench Events: the whole chain in one narrative
A real cloud incident rarely shows up as a single alert. It's dozens to hundreds of forensic events scattered across services and time. Point AI Summary for Workbench Events at a group of those events and it reads the full chain, then gives you a clear narrative covering what happened, in what order, and why it matters. Every summary is generated from Mitiga's forensic data lake, so it stays connected to the raw log entries and detectors behind it. If you want to see the underlying evidence yourself, it's one click away.
An answer you can act on
Every serious vendor in this category now calls itself an “agentic analyst.” Very few can show you the evidence behind an answer or exactly what made them more or less confident in it. The HeliosAI Agent can because it's built on the same full-fidelity, normalized Cloud Security Data Lake that delivers a 90% reduction in alert triage time, and it's available via API for teams building their own agentic SOC workflows.
For Tier 2 and Tier 3 analysts investigating incidents that span a dozen platforms, that's the difference between an AI that sounds right and one you can actually act on.
The HeliosAI Agent is generally available today to all Mitiga customers.