GET THE BRIEFING

AI THREAT INTELLIGENCE

AI changes the pace.
Judgment sets the priority.

Track how adversaries use AI, how AI systems become targets, and which controls reduce exposure now.

VIEW THE CURRENT THREAT →Evidence reviewed Sep 9, 2026
AI THREAT RADAR
01 -Attacker Accelerator
02 -AI Systems as Targets
03 -AI Supply Chain
04 -Human Identity
PLACEMENT KEY Inner: active now Middle: expanding Outer: emerging

Distance from the center shows operational maturity based on the cited reporting. Angle only separates categories; it is not a severity score. Select the matching surface below for signals and defensive priorities.

OBSERVED ACCOUNTS832Analyzed after enforcement
MALWARE WRITING67.3%560 of 832 accounts
LATERAL MOVEMENT6.5%54 of 832 accounts
MEDIUM+ RISK SHARE33% → 56%First half vs second half

Scope: Anthropic accounts with sufficient detail, March 2025–March 2026. These figures describe that investigated dataset and are not estimates of all cyber activity.

2026 OPERATING PICTURE

What defenders should separate.

Current reporting shows routine AI assistance at scale and early agentic experimentation. The distinction matters when setting controls and urgency.

OBSERVED NOW EMERGING
OBSERVED NOWHIGH CONFIDENCE

AI is already a force multiplier.

  • Phishing, translation, impersonation, and social engineering content
  • Malware and script development, debugging, and adaptation
  • Reconnaissance, stolen-data analysis, and operational research
  • Identity fabrication and support for fraudulent remote-worker operations

Human operators still commonly choose objectives, targets, and deployment.

EMERGINGWATCH CLOSELY

Agentic workflows are moving deeper.

  • Autonomous vulnerability discovery and exploit development
  • Tool-driven reconnaissance and post-compromise actions
  • Theft or misuse of proprietary models, research, and training data
  • Compromised cloud environments used to run unauthorized AI infrastructure

Reporting shows experimentation and adoption, but capability and scale vary by actor.

AI THREAT SURFACE

Four surfaces. One control plan.

Select a surface to see the immediate defensive priority.

DEFENSIVE PRIORITIES

Control the agent. Protect the decision.

Start with access, visibility, and containment. Apply these actions to sanctioned AI and shadow use.

  1. 01

    Inventory AI use

    Identify applications, models, agents, plugins, data sources, owners, and business decisions they influence.

  2. 02

    Constrain privileges

    Use least privilege, short-lived credentials, narrow tool access, and approval gates for consequential actions.

  3. 03

    Treat inputs as untrusted

    Assume prompts, retrieved documents, websites, and tool results can contain hostile instructions or poisoned data.

  4. 04

    Log the full chain

    Record prompts, retrievals, model responses, tool calls, identity context, and final actions for investigation.

  5. 05

    Protect models and data

    Limit training-data exposure, restrict model downloads, scan dependencies, and monitor unusual cloud compute or egress.

  6. 06

    Practice containment

    Test how to suspend an agent, revoke credentials, isolate connected tools, preserve evidence, and restore safely.

CURRENT EVIDENCE

2026 reporting, with scope intact.

Direct links open the original research. Vendor observations are attributed and should be evaluated alongside your own telemetry.

MAR 06 · MICROSOFT

AI as tradecraft

Observed use across phishing, malware support, research, translation, and identity fraud; agentic use remains early.

READ SOURCE ↗ JUN 03 · ANTHROPIC

AI use mapped to ATT&CK

Analysis of 832 enforced accounts shows deeper use across malware writing and post-compromise activity.

READ SOURCE ↗ MAY 11 · GOOGLE GTIG

AI and initial access

Reporting on the integration of AI into adversarial workflows, including suspected use of agentic tools.

READ SOURCE ↗ SEP 07 · GOOGLE GTIG

From prompting to autonomy

Tracks model and research theft, cloud abuse, and the evolution of adversarial AI operations.

READ SOURCE ↗ APR 16 · MANDIANT

Enterprise AI vulnerabilities

Explains how capable models change vulnerability discovery and why organizations need a prepared defense.

READ SOURCE ↗ APR 22 · MICROSOFT

AI-accelerated defense

Documents how AI can find weaknesses, chain issues, and compress the disclosure-to-exploitation window.

READ SOURCE ↗
JATECK AI THREAT BRIEF

Prioritize what can act, access, and decide.

Use this page with the Zero-Day Clock and Priority Action Board to connect emerging capability to immediate defensive work.

ZERO-DAY CLOCK →PRIORITY BOARD →