Practice, at length.
A Speak to It™ term tells you what something is. This is where the same subject is worked through properly: what good looks like, what to require, how to evidence it, and where most teams get it wrong.
289 written, 1405 more commissioned. Free to read, because a common professional vocabulary should not depend on ability to pay.
AI API Key Exposure
Developer Portal. Jupyter Notebook. GitHub. Public Repository. Bot. Six Weeks.
6 min read · 15 Sep 2026
AI Adversarial Attacks
97% Benchmark Accuracy. Adversarial Perturbation: Invisible to Humans. Model Confidence on Wrong Classification: 97.3%.
4 min read · 13 Sep 2026
AI Attack Surface Expansion
Three AI Vendors. Each Assessed Independently. Aggregate Data Access: Complete Customer Profile. Aggregate Assessment: Not Conducted.
5 min read · 10 Sep 2026
AI Bias in Risk Decisions
Objective Risk Score. Training Data: Historical Assessments. More Intensive Assessment: More Findings. Score: Circular.
6 min read · 8 Sep 2026
AI Compliance Enforcement
AI Policy: Eighteen Months Active. Reviewed Systems: Seven. Unreviewed Production Systems: Four. Enforcement: Absent.
5 min read · 5 Sep 2026
AI Compliance vs Security Gap
GDPR: Confirmed. SOC 2: Confirmed. Article 22 Right to Explanation: Model Cannot Provide It.
6 min read · 3 Sep 2026
AI Continuity and Exit Planning
Model Sunset: 90 Days. Training Data: Not Maintained. Feature Engineering: Undocumented. Alternatives: Not Evaluated. Continuity Plan: None.
5 min read · 31 Aug 2026
AI Data Lineage Issues
AI Detected at 2:17am. Response Started 9:15am. Seven Hours. Attacker: Still Present.
6 min read · 29 Aug 2026
AI Data Retention Risks
M&A Targets Uploaded for Summarisation. Retained Thirty Days. Zero Retention: Not Configured.
6 min read · 26 Aug 2026
AI-Driven Automation Risks
847 Renewal Offers. 312 Sent to Customers Who Had Cancelled. Automation: Correct. Integration: Missing.
5 min read · 24 Aug 2026
AI Explainability Challenges
SHAP Values: What the Model Used. Why Those Features Are Valid Credit Risk Indicators: Not Answerable from Outputs.
4 min read · 21 Aug 2026
AI Governance Frameworks
AI Ethics Policy: Confirmed. Model Review Process: Described. Implementation Evidence: Not Requested.
5 min read · 19 Aug 2026
AI Governance Gaps
AI Ethics Committee: Reviewed and Approved. EU AI Act High-Risk Classification: Not Assessed.
7 min read · 16 Aug 2026
AI Hallucination Risk in Decisions
Credit Denied. Reason: Fourteen Features. Dominant Feature: Payment Timing Meets Geographic Mobility. Appeal: Dismissed.
6 min read · 14 Aug 2026
AI Incident Response Gaps
AI Decision Incident. IR Playbook: Data Breach, Ransomware, Outage. AI Incident: Not Covered.
5 min read · 11 Aug 2026
AI Inference Data Leakage
Zero Retention Configured. In-Session Context Window: Customer Data Accessible to Crafted Prompts.
6 min read · 9 Aug 2026
AI Logging and Traceability
Audit Log: API Called. Decision Returned. Model Version: Not Captured. Input Features: Not Captured. Decision: Unreconstructable.
6 min read · 6 Aug 2026
AI Misuse by Vendors
Support AI. Historical Training Labels. Historical Bias Encoded as Normal. Production Outage: Low Priority.
5 min read · 4 Aug 2026
AI Misuse Scenarios
AI Sales Tool: Approved. Personalised Outreach: Generated. Prospect Research: Autonomous, Undisclosed, Unconsented.
5 min read · 1 Aug 2026
AI Model Supply Chain Risk
The Model Works Perfectly. The Risk Is in What It Was Trained On and Who Has Access to It.
8 min read · 30 Jul 2026
AI Model Version Control
Foundation Model: Eleven Months Outdated. Security Patches: Three. Privilege Escalation: Unpatched. Validation Cycle: Three to Four Months.
6 min read · 27 Jul 2026
AI Operational Risks
94% Accuracy. 6% Missed. Missed Records: Systematically the Most Complex. Human Review: Assumed Comprehensive.
5 min read · 25 Jul 2026
AI Output Data Leakage
AI Draft. Confident Statements. Factually Incorrect. Published as Analysis.
6 min read · 22 Jul 2026
AI Plugin Vulnerabilities
94% Detection Rate. 56 Hours of False Positive Investigation. 23 Missed Anomalies. None of It in the Dashboard.
7 min read · 20 Jul 2026
AI Risk Scoring Reliability
Risk Score: 73. Historical Accuracy at This Range: Unknown. Confidence Interval: None. Trend: Not Available.
6 min read · 17 Jul 2026
AI Security Monitoring
Weekly Reports: Alert Volumes, Resolution Rates, Coverage Statistics. Gaps, Limitations, Human Supplement Needed: Not Shown.
6 min read · 15 Jul 2026
AI Security Testing Gaps
SOC 2: Infrastructure Covered. Pentest: Web App Covered. Code Review: Application Covered. AI Model: Not Tested.
6 min read · 12 Jul 2026
AI Supply Chain Dependencies
Four AI Pipeline Libraries. Three Well-Audited. One: Single Maintainer, Eight Months Inactive, Arbitrary Code Execution.
6 min read · 10 Jul 2026
AI Threat Detection Gaps
Detection Dashboard: All Green. Novel Technique Category: Not on the Dashboard. Attacker: Undetected.
6 min read · 7 Jul 2026
Supply Chain AI Trust Boundaries
Finance AI Trusted to Answer Policy Questions. Attacker with Compromised Credentials: Also Trusted. AI Cannot Tell the Difference.
5 min read · 5 Jul 2026
AI Vendor Lock-In Risks
API Deprecation Notice: 30 Days. Migration Plan: None. Alternative Evaluated: None. Calibration Reproducible: No.
5 min read · 2 Jul 2026
API-Based AI Exposure
Vendor Confirmed Data Security. Data Was Flowing to OpenAI. Enterprise Had Never Assessed OpenAI.
6 min read · 30 Jun 2026
LLM Integration Risks
LLM Connected to 140,000 Documents. Service Account Access Broader Than User Permissions. Context Window: Exfiltration Channel.
6 min read · 27 Jun 2026
Model Access Control
Authenticated API. Rate Limiting: None. Query Volume: 500,000. Model: Reconstructed by Competitor.
6 min read · 25 Jun 2026
Model Drift Risks
Metrics: Stable. Detection Rate: Good. Behaviour: Quietly Changed Over Eight Months.
6 min read · 22 Jun 2026
Model Theft Risks
Model Secured. Vendor API Secured. Enterprise Customer Application: Not Secured. Model: Accessed Through the Customer.
6 min read · 20 Jun 2026
Prompt Injection via Vendors
Customer Service Chatbot. Attacker Sends an Inquiry. LLM Follows the Attacker's Instructions.
7 min read · 17 Jun 2026
Third-Party AI Integrations
Salesforce Einstein. HubSpot AI. Workday AI. Enabled by Default. Assessed as SaaS. Never as AI.
6 min read · 15 Jun 2026
Training Data Poisoning
The Model Learned from Three Years of Data. One Year Was Labelled Wrong. The Model Learned That Too.
7 min read · 13 Jun 2026
Vendor AI Usage Transparency
AI Features: Disclosed. AI Used to Score, Classify, and Route the Enterprise: Not Disclosed.
6 min read · 11 Jun 2026