Data Risks for AI Agents

Risks to the data agents capture, move, learn from and update with.

IBM found 13% of organizations had a breach of an AI model or application in 2025, and 97% of those lacked proper AI access controls. Shadow AI caused one in five breaches and added $670,000 to the average cost.

AI breach findings: Breached orgs lacking AI access controls 97%; Breached orgs without AI governance 63%; Breach caused by shadow AI 20%; Breach of an AI model or app 13% AI BREACH FINDINGS Breached orgs lacking AI access controls 97% Breached orgs without AI governance 63% Breach caused by shadow AI 20% Breach of an AI model or app 13% SHARE OF ORGANIZATIONS, IBM 2025
Global average breach cost
$4.44M
US average breach cost
$10.22M
Shadow AI cost premium
$670,000
Registry incidents
5

Risks

Eight data risks, from sensor capture to the update supply chain

Data Risks: risks covered
RiskRegistry incidentsEvidence casesControls that bound it
Telemetry Capture Integrity04Telemetry integrity, Monitoring
Data Transit Security14Access and authorization, Identity and attestation, Telemetry integrity
Training Data Poisoning04Monitoring, Identity and attestation
Model Update Integrity04Identity and attestation, Monitoring
Fleet-Scale Attacks04Monitoring, Identity and attestation, Behavioral envelopes
Surveillance Data Harvesting14Consent and capture, Access and authorization
Supply Chain of Updates34Identity and attestation, Access and authorization, Monitoring
OT and ICS Telemetry Deception04Telemetry integrity, OT and ICS integration, Monitoring

Evidence

AI tools themselves became the breach path in 2025

Documented cases: data Risks
DateCaseCountryWhat happenedSource
Jan 2025DeepSeek database exposureChinaA publicly open database held chat history and API keysSource
Jan 4, 2025Gravy Analytics breachUnited StatesAttacker claimed 17 TB including phone location dataSource
Jul 19, 2025Amazon Q extension wiper promptUnited StatesAn injected prompt told the agent to wipe files and AWS resources; live two daysSource
Aug 26, 2025Nx s1ngularityGlobalMalware used AI coding tools to hunt secrets; 2,180 accounts exposedSource

Registry

Five registry incidents involve data risks

Registry incidents meet a stricter bar than the cases above: public sources name the developer, the failed control and who found it.

Registry incidents linked to this risk
IncidentDateAgentControl that failed
Tesla Staff Share Customer Car Camera Footage2019 to 2022Vehicle camera data pipelineEmployee access to customer recordings
EchoLeak Turns Microsoft 365 Copilot Into a Data LeakJan 2025Enterprise copilotUntrusted email could direct the copilot to privileged data
Stolen Drift Tokens Drain Salesforce DataAug 2025AI chat agent integrationBroad OAuth tokens held by an AI chat agent
Nx Malware Turns AI Coding Tools Against Their UsersAug 2025Coding agents (Claude Code, Gemini CLI, Amazon Q)Permission prompts skipped by command-line flags
OpenAI Test Models Breach Hugging FaceJul 2026Frontier models in evaluationSandbox egress through a zero-day in a package proxy

Rules

China's amended Cybersecurity Law covers AI risk; EU AI data duties wait until 2027

Rules that address data risks
RuleJurisdictionWhat it requiresStatus
EU AI Act (amended by Regulation 2026/1744)EUData governance and cybersecurity for high-risk AIDec 2, 2027
Cybersecurity Law amendmentChinaAI risk monitoring; fines up to RMB 10 millionEffective Jan 1, 2026
Interim Measures for Generative AI ServicesChinaLawful training data and security assessmentAug 15, 2023

Status colors: green in force, blue voluntary standard or framework, amber adopted and phasing in, gray draft or pending, red withdrawn.

Record

Page Record

Risk group
Data Risks
Datasets
ai_risk_evidence.csv (4 rows); registry incidents (5 linked); ai_attack_surface.csv
Key sources
DeepSeek database exposure
Gravy Analytics breach
Amazon Q extension wiper prompt
Nx s1ngularity
Last verified
October 2026