Verify a member Contact Member sign in
International Initiative for Augmented Security

Research

The IIAS AI Threat Lab analyzes how AI systems are attacked, how AI is used against organizations, and how defenders can respond, and publishes its findings for the security community.

The AI Threat Lab

The AI Threat Lab is the research function of the initiative. It tracks observed and emerging attack techniques against AI systems, and the ways adversaries use AI to scale fraud, impersonation and intrusion.

Each finding is analyzed, mapped to recognized technique taxonomies such as MITRE ATLAS, and paired with practical mitigations and detection guidance that security teams can apply.

Lab findings feed directly into the Body of Knowledge, IIAS training programs and member briefings, so what the initiative teaches reflects what practitioners face.

Threat advisories

Technical analysis of attack techniques, with mapped mitigations and detection guidance. Filter by pillar.

Security from AIAdvisory

Voice-clone attacks on payment-approval workflows

Attack patterns against verbal approval processes, and the out-of-band verification controls that break them.

Read the advisory
Security with AIAdvisory

Human oversight controls for LLM-assisted alert triage

Designing review thresholds, override logging and performance monitoring when LLMs participate in SOC decisions.

Read the advisory
Security for AIAdvisory

Provenance verification for open-weight models

Risks in downloading and deploying third-party model weights, and the integrity checks to apply before production use.

Read the advisory
Security from AIBriefing

Synthetic identities in remote onboarding

Use of AI-generated documents and faces against remote identity proofing, and layered verification measures.

Read the briefing

What every advisory contains

Advisories follow a consistent structure, so security teams can assess relevance quickly and act on the findings.

Advisory structure
Summary
The technique, who it affects and why it matters, in a few sentences.
Affected systems
The types of models, applications, agents or processes exposed.
Technique mapping
References to MITRE ATLAS techniques and the OWASP Top 10 for LLM Applications where applicable.
Attack walkthrough
How the attack works in practice, described at a level that supports defense.
Mitigations
Prioritized controls across design, configuration and process.
Detection guidance
Signals, logs and indicators that help teams identify attempts.
Body of Knowledge reference
The domains and core topics the advisory relates to.

Publications

The Lab publishes three types of output, each serving a different need.

Threat advisories

Focused analysis of a specific technique or campaign, published when a finding is ready to act on.

For security engineers, analysts and architects

Technical briefings

Shorter analyses of emerging issues and live sessions with Lab researchers. Briefing attendance is recorded in CE hours.

For practitioners and members

Annual AI Threat Report

A yearly review of how attacks on and with AI have developed, with trends and priorities for the year ahead.

For security leaders and policymakers

Current research focus

Research priorities are organized by the three pillars of practice and reviewed as the threat landscape changes.

Security for AI

  • Prompt injection in agents and retrieval systems
  • Tool permissions and excessive agency
  • Model and dataset supply-chain integrity

Security with AI

  • Reliability of LLM-assisted triage
  • Human oversight and override design
  • Securing the SOC's own AI tooling

Security from AI

  • Voice-clone and video deepfake fraud
  • Synthetic identities in onboarding
  • AI-scaled phishing and social engineering

Report a technique or vulnerability

If you have identified a new attack technique against AI systems, or a vulnerability in an AI product, the Lab can help coordinate disclosure with affected parties before publication.

  • Reports are handled confidentially
  • Affected vendors are notified before any publication
  • Reporters are credited with their consent
Coordinated disclosure policy

Contribute to research

Members can join technical working groups that support Lab research on specific domains. Researchers and organizations can propose collaborations on shared problems.

  • Member working groups by domain
  • Collaboration with academic and industry researchers
  • Review of draft advisories by practitioners
Propose a collaboration

Stay ahead of AI-enabled threats

Receive each advisory on publication, with invitations to technical briefings. Members receive advisories before public release.