Attackers are using AI at machine speed, and defenders cannot match that pace without equally powerful agents for threat detection, proactive product security reviews, and always-on evidence collection to maintain top-level security certifications and standards. At World of Workato 2026, Hans Gustavson, Workato’s CISO, and two members of his team, Sid Yadav (Senior Manager, Security Operations and Engineering) and Sal Hernandez (Director, Governance, Compliance and Risk), showed what they have built in response: agentic workflows for security operations, audit evidence, and product security reviews.
If you run integration, enterprise architecture, or business systems, this is worth your time. The building blocks on their slides are ones IT teams will recognize: Agents, Skills built as recipes, Knowledge Bases, event streams, and APIM. Hans pointed out that Workato has historically been seen as “really focused on the enterprise architect, and the IT leader,” and that the security domain is a strong fit as well.
The pressure: AI works for both sides
The session opened with three external data points, as cited on the team’s slides:
- 87% of security leaders flagged AI-related vulnerabilities as the fastest-growing cyber risk (World Economic Forum, Global Cybersecurity Outlook 2026).
- Attack speed is up 65% from 2024, with the fastest observed breakout at 27 seconds (CrowdStrike, 2026 Global Threat Report).
- 80 to 90% of tactics executed by threat actors in one AI-orchestrated campaign ran without human intervention (Anthropic Research, November 2025).
Hans drew a blunt conclusion for security teams:
“The notion of a human SOC is, I believe, truly dead. In order to protect yourself from the AI attacks and threats, we have to operate at the same speed.”
Hans Gustavson, CISO, Workato
The CISO framed it as a choice: AI expands what defenders can do and what attackers can do, and “the gap between the two is a choice, not a given.” Hans said the team’s work moved from AI-assisted use cases, to single-agent experiments last year, to solutions now running in Workato’s production environment.
1. Agentic SOC: triage every alert, not just the loud ones
Security tools are disconnected, and each new tool adds alerts, so an engineer looking at an endpoint alert may never see the cloud activity it connects to. Sid cited MDR (Managed Detection and Response) studies showing 40% of alerts are never triaged or analyzed.
The SecOps Agent works in stages:
- Ingest and normalize. Alerts from many platforms land in one place and are normalized to the OCSF (Open Cyber Security Framework) format. The slides list sources such as SentinelOne, Wiz, and AWS GuardDuty.
- Enrich. The enrichment agent correlates indicators of compromise against threat intel platforms and internal sources. It also adds context: who the identity is, what permissions it has, and what lateral movement an attacker could achieve from it.
- Triage. The triage agent builds an investigation plan on the fly, runs it, and calls the alert a true or false positive, with its confidence and reasoning.
- Respond. If the alert is a true positive and the agent has a remediation skill, such as quarantine, it acts. If not, it hands the alert to a human analyst. Containment workflows include human-approval gates.
SecOps Agentic SOC Workflow
Sid said, investigations that used to take a couple of hours to days now take action within seconds. About 80% of alerts are auto-triaged by the agents, much of it false-positive noise, and only 20% go on to a human. Sid said the team no longer needs a team investigating alerts around the clock, and threats are being mitigated within their SLAs.
2. Agentic GRC: collect evidence once, answer every audit
Sal shared typical pain around doing audit evidence collections the old way, which was manual and reactive. Auditors send long request lists, and the GRC (Governance Risk and Compliance) team chases other teams for screenshots. His summary of that experience: “You become extremely unpopular in about five minutes.” Workato holds 11 certifications, and by the time evidence reaches an auditor, a control gap is already too late to fix quietly.
The team built two pieces:
- An Evidence Collector. On a schedule you configure (monthly or weekly), it pulls evidence from systems like Okta and compares it to baseline parameters and the last evidence collected. Sal’s example: a password policy says 15 characters, the collector finds 10, and the control owner is notified right away. The owner either opens a ticket to remediate or corrects the policy, so the finding rarely reaches the auditor. Evidence is stored in a central repository that GRC controls.
- A GRC Audit Agent When the external auditor’s request list arrives (IRL, PBC, or spreadsheet), it is uploaded, mapped to control IDs, and matched to evidence in the repository. GRC reviews and approves any identified delta before the evidence goes to the auditor.
GRC Agentic Workflow for Audit Evidence Collection
The mapping covers about 1,300 controls across 23 frameworks. Sal said the team trusts it because it has been through Workato’s own 11 audits, using the most restrictive framework for evidence. One piece of evidence maps to many controls, so, in Sal’s words, “We get evidence once, and it complies with everything.” He estimated teams can save about 67%, even 80%, of the time.
Sal, a former auditor himself, was candid about the human side. Auditors have to learn to trust you, so be ready to show the automation end to end the first time. Hans added that chain of custody matters. Workato logs the activity today and wants to sign evidence objects, which he said is not available yet. The team also plans to produce machine-readable JSON output alongside screenshots, for frameworks that require it.
3. Security Review Agents: streamlines product security design
Hans described a familiar bottleneck. Developers filed a Jira ticket, a security engineer reached out, and with a globally distributed team, scheduling a design review was slow. The team was not meeting the bar for design reviews and threat models.
Now a developer starts a conversation with an angent in Slack, attaches requirements, and links Confluence, Google Drive, or PMO tickets. The slides note the intake is kept to a short exchange (three messages at most). If a GitHub link is provided, a code analysis agent runs static analysis, secrets scanning, and dependency scans. The agent applies STRIDE-based threat modeling and a three-tier decision: approve, conditional, or escalate. Small, low-risk changes can be approved with no human review. Larger ones are escalated with a prioritized, pre-populated Jira ticket that includes risk rationale and a compliance mapping, so “security isn’t starting from a blank page.” The developer confirms the agent’s summary before anything is filed, and a report lands in Confluence.
Product Security Design Review Agents
Hans said developer feedback has been “very, very positive,” because the team is helping developers do their jobs “and not trying to be a barrier” to shipping secure products. A PMO (Product Management Office) agent that watches Jira for new features and offers feedback proactively is something the Workato GRC team is currently working on for internal use.
What Security and IT teams can take from this
- Start where the signals already flow. All three workflows connect systems that already hold the data: alert platforms, identity systems, ticketing, code repositories, and collaboration tools. Connectivity came first, and the agents sit on top of it.
- Put humans at the decisions that matter. Containment approvals, a developer confirming a summary, and GRC reviewing deltas are designed in, not bolted on.
- Treat trust as a requirement. Audit trails mapped to MITRE ATT&CK and NIST, activity logging, and a plan for signed evidence all came up because auditors and analysts will ask.
- Engage your security counterparts early. The session slides made the case that enterprise security is a team sport across SecOps, Product Security, and GRC.
If you are thinking about agentic security in your own environment, the Workato Security and GRC team’s templates, best practices, and accelerators will be a great place to begin.
