Skip to main content
Mallory
← Back to News
ai-platform-security

SentinelLABS Releases VerifiedIDA to Make LLM Reverse Engineering Auditable

Emerging Technology Risk · 6 sources · Updated 17d ago · First seen Nov 4, 2026

SentinelLABS Releases VerifiedIDA to Make LLM Reverse Engineering Auditable

SentinelLABS released VerifiedIDA, an open-source reference implementation for IDA Pro designed to make LLM-driven reverse engineering inspectable, resumable, and reusable. Its evaluation found recurring inaccurate labels, unrecorded findings, lost unfinished work, and premature declarations of completion. The researchers proposed a disassembler-independent standard requiring attributable evidence, verified database changes, preserved investigation state, and enforceable scope with independent review. VerifiedIDA mediates database access, verifies edits, and maintains persistent investigation records, but does not guarantee analytical correctness (SentinelLABS).

Case studies using the Turla-associated ComRAT backdoor showed that models recovered useful structures and behavior while also producing misleading annotations and missing capabilities. In targeted reruns, an experimental call-flow coverage mode prompted identification of forwarded-export resolution that six earlier runs had overlooked. ComRAT is a consequential test case: ESET previously documented government espionage involving its Gmail-based command channel and document theft (ESET). API semantics also matter when validating findings: Microsoft specifies that Windows PROCESS_INFORMATION contains process and primary-thread handles and identifiers, and that callers must close both handles after use to allow proper cleanup (Microsoft). Security teams adopting autonomous analysis should require evidence-backed findings, persistent handoffs, and independent validation rather than treating completed annotations as proof of comprehensive coverage.

New surface, new adversariesTrack how attackers are adapting to this technology.
Track this risk
Event timeline

How this story unfolded

Events from the most recent confirmed update back to the earliest known activity.

13 events · 2007 – Jan 2020
  1. Reported Nov 4, 202618d ago

    SentinelLABS released VerifiedIDA as an open-source reference implementation

    SentinelLABS released VerifiedIDA for IDA Pro with an interface specification and user guide. The framework mediates model access, verifies edits against recorded evidence, preserves investigation state, and supports independent review, without guaranteeing analytical correctness.

    VerifiedIDA | Making LLM-Driven Reverse Engineering Inspectable, Continuable and Reusable
  2. Reported Nov 4, 202618d ago

    VerifiedIDA coverage tests recovered previously missed ComRAT behavior

    Two subsequent runs using VerifiedIDA's experimental call-flow coverage examined all 20 direct internal callees and identified forwarded-export resolution missed by the earlier runs. A whole-project coverage attempt generated 265 required nodes for eight edits, leading researchers to reserve the mode for selected claims.

    VerifiedIDA | Making LLM-Driven Reverse Engineering Inspectable, Continuable and Reusable
  3. Reported Nov 4, 202618d ago

    SentinelLABS proposed requirements for verifiable LLM reverse engineering

    SentinelLABS proposed a disassembler-independent standard requiring attributable evidence, verified database changes, preserved project state, and enforceable scope with independent review. The proposal requires unresolved review findings and pending investigative work to remain persistently tracked.

    VerifiedIDA | Making LLM-Driven Reverse Engineering Inspectable, Continuable and Reusable
  4. Reported Nov 4, 202618d ago

    Six ComRAT analysis runs missed forwarded-export resolution

    None of six earlier GPT-5.6 Sol runs inspected ComRAT function sub_180081450, which recursively resolves forwarded exports. All four runs that annotated its parent function omitted that capability.

    VerifiedIDA | Making LLM-Driven Reverse Engineering Inspectable, Continuable and Reusable
  5. Reported Nov 4, 202618d ago

    ComRAT case study exposed misleading model-generated annotations

    GPT-5.6 Sol recovered a ComRAT structure and correctly identified a PROCESS_INFORMATION field, but produced misleading or incomplete annotations elsewhere. These included an ambiguous label for an unused field, a misleading process-creation function name, and a generic pointer annotation that omitted recognized logging semantics.

    VerifiedIDA | Making LLM-Driven Reverse Engineering Inspectable, Continuable and Reusable
  6. Reported Nov 4, 202618d ago

    SentinelLABS evaluated autonomous LLM reverse engineering

    SentinelLABS tested proprietary and open-source models analyzing binaries and annotating IDA Pro databases without analytical feedback. Researchers found recurring inaccurate labels, unrecorded findings, lost unfinished work, and premature termination of analysis.

    VerifiedIDA | Making LLM-Driven Reverse Engineering Inspectable, Continuable and Reusable
  7. Jan 2020

    ComRAT v4 remained in operational use

    ESET observed ComRAT v4 in use as recently as January 2020, establishing continued operation beyond the compilation dates of its known samples.

    From Agent.BTZ to ComRAT v4: A ten-year journey
  8. Nov 2019

    Latest ComRAT v4 sample in ESET's investigation compiled

    The most recent ComRAT v4 sample observed in ESET's investigation was compiled in November 2019.

    From Agent.BTZ to ComRAT v4: A ten-year journey
  9. Apr 2017

    First known ComRAT v4 sample compiled

    The earliest ComRAT v4 sample known to ESET was compiled in April 2017. This generation introduced a substantially different, more complex C++ code base.

    From Agent.BTZ to ComRAT v4: A ten-year journey
  10. 2008

    ComRAT used in U.S. military breach

    ComRAT was used in the 2008 breach of the U.S. military that brought the malware widespread attention.

    From Agent.BTZ to ComRAT v4: A ten-year journey
  11. 2007

    First ComRAT version likely released

    The first version of ComRAT, also known as Agent.BTZ and Chinch, was likely released in 2007. It could spread through removable drives.

    From Agent.BTZ to ComRAT v4: A ten-year journey
  12. Date unknown

    ESET disclosed ComRAT v4's Gmail-based command channel and Turla attribution

    ESET documented ComRAT v4's Gmail web-interface command channel, legacy HTTP channel, and Virtual FAT16 filesystem. Researchers assessed that Turla exclusively used the backdoor, citing victimology, shared protocols and infrastructure, and relationships with other Turla tools.

    From Agent.BTZ to ComRAT v4: A ten-year journey
  13. Date unknown

    ESET identified government targets and document theft using ComRAT v4

    ESET identified at least three targets: two Ministries of Foreign Affairs and a national parliament. Operators stole sensitive documents, including through a custom .NET tool querying a central SQL Server repository, and used public cloud services for exfiltration.

    From Agent.BTZ to ComRAT v4: A ten-year journey
Linked entities

Related entities

Vulnerabilities, threat actors, malware, products, organizations, breaches, and observables Mallory has linked to this story. Indicator values are masked here and available in full in the app.

Open in app →

Observables

1 linked

Indicator values are masked on this page. See the values in Mallory Domains, IPs, hashes, and URLs are exportable to your SIEM.

Do you run this?Follow how adversaries are adapting to this technology, and where it touches your stack today.
Track This Risk
Mallory platform

See the full picture, correlated to your attack surface.

This page covers what’s public. Mallory adds the parts that aren’t: which of your assets are affected, which threat actors are using it right now, and what to deploy next.

Exposure mappingMap indicators from this story to your assets and identify affected systems in minutes.
Threat actor evidenceEvery observed campaign, victim, and pivot linked to actors named in this story.
Associated malwareMalware, exploits, and IOCs connected to the activity described here.
Detection signaturesYARA, Sigma, and Snort rules deployed to your SIEM as soon as they’re published.
Scheduled alertsGet matching new stories delivered to your team as they break, not the next morning.
AI threadsAsk questions about this story and take action on the answers.