Intelligence & Research
See AI risk before it becomes AI headlines
Market, regulatory, and threat intelligence, industry benchmarking, and original research on AI security, governance, and the risks of emerging AI.
Research programmes
Four standing programmes
Research at Deep Heuristics exists to improve the quality of our assessments. It is published because findings that stay private do not raise the standard of the field.
Flagship publication
The 2026 State of AI Trust Report
Our annual assessment for boards, regulators, and enterprise AI leaders. Fourteen chapters, 25 defining statistics, more than 40 documented case studies, and five original frameworks — synthesised from Stanford HAI, McKinsey, Deloitte, PwC, IBM, Edelman, Gartner, NIST, the EU AI Office, Singapore’s IMDA, the UK AI Security Institute, OWASP, MITRE, the AI Incident Database, and court records.
- Governance and board oversight
- Risk taxonomy and heatmap
- Security, injection, and red teaming
- Assurance and certification
- Regulatory divergence
- Incidents and case studies
How we publish
Research standards we hold ourselves to
- Method disclosed
- Sample, period, and method are stated. A finding without a method is an opinion with a chart.
- Limits stated
- Where a sample is small, unrepresentative, or self-selected, we say so in the finding rather than the appendix.
- No client attribution
- Client engagements never appear as case material without written consent, and never in identifiable form without it.
- Coordinated disclosure
- Security findings affecting identifiable products follow our responsible disclosure policy before publication.
- Corrections published
- Corrections are issued against the original publication and recorded, not silently edited in.
Deploy AI with confidence.
Start with an independent assessment of your highest-stakes AI system.