The Next Input — Issue #196

In partnership with

⚡ The Briefing — 60 sec

• Anthropic, OpenAI and SpaceX are bigger than the last 25 years of tech exits Umm… WHAT? Three private companies have apparently looked at decades of tech exits and said, “Cute.” Whatever happens from here, this concentration of capital and power is absolutely fucking staggering.

• Spotting AI deepfakes is becoming a cat-and-mouse game Remember Will Smith eating spaghetti all those years ago? Pffft. We’ve now reached the point where, given four images, your actual face might be the one people identify as fake.

• Sovereign AI is critical for national innovation and independence This theme is steadily emerging. Models, compute, data, infrastructure and jurisdiction are becoming national strategy—not merely technology procurement. Interesting to see how it all shapes up.

🛠️ The Playbook — Digital Provenance Engine

Mission
Build a verification workflow that establishes whether high-risk digital content is authentic before your organisation trusts or distributes it.

Difficulty
Advanced

Build time
4–6 hours

ROI
Reduces fraud, misinformation and reputational exposure while creating a reliable chain of trust for digital evidence.

0) Why This Matters

AI power is concentrating while synthetic media is becoming harder to distinguish from reality.

That creates an uncomfortable combination: extraordinarily capable systems, increasingly convincing fabrications and critical infrastructure controlled by a small number of providers and jurisdictions.

A generic “AI detector” will not save you. Organisations need layered verification built around provenance, source confirmation, metadata, internal records and human escalation.

Trust now requires infrastructure.

1) Architecture

2) Workflow

1. Route suspicious media, documents or claims through one controlled intake channel.

2. Preserve the original file and record who submitted it, where it came from and what it supposedly proves.

3. Extract metadata and compare it with the stated timeline, device, location and source.

4. Cross-check names, events, documents and imagery against trusted internal or public records.

5. Use AI to identify inconsistencies and prioritise investigation—not to issue the final verdict.

6. Escalate ambiguous or consequential cases to a named human reviewer and retain the full decision trail.

3) Example Prompts

Evidence Triage

You are a digital evidence triage analyst.

Review the supplied content and supporting context.

Identify:
- internal inconsistencies
- missing provenance
- suspicious timestamps or chronology
- visual or linguistic anomalies
- claims requiring independent confirmation
- potential manipulation indicators

Do not declare the content authentic or fake.

Return:
1. preliminary risk level
2. specific concerns
3. evidence still required
4. recommended verification steps
5. mandatory escalation triggers

Source Corroboration

Compare the submitted claim against the approved records provided.

Create a table containing:
- claim
- supporting evidence
- conflicting evidence
- missing evidence
- confidence level
- source location

Clearly distinguish verified facts from assumptions and unresolved questions.

Executive Incident Brief

Prepare a concise executive brief for a suspected synthetic-media incident.

Include:
- what was received
- where it originated
- what has been verified
- what remains uncertain
- likely operational or reputational impact
- immediate containment actions
- decision required from leadership

Do not overstate certainty.

4) Guardrails

• Never use a single detector as proof of authenticity or manipulation.

• Preserve original files before compression, conversion or editing.

• Separate machine-generated risk signals from final human conclusions.

• Require two independent forms of corroboration for consequential decisions.

• Restrict sensitive evidence using least-privilege access.

• Record uncertainty instead of forcing a binary verdict.

• Define legal, HR, communications and security escalation paths in advance.

5) Pilot Rollout — 3 hours

1. Choose one high-risk content channel, such as executive requests, media submissions or identity documents.

2. Create a controlled intake form that captures the original file, source and claimed context.

3. Add automated metadata extraction and a structured AI triage step.

4. Connect the workflow to one trusted internal knowledge source for corroboration.

5. Configure a Teams approval queue with named reviewers and escalation thresholds.

6. Test the system against five genuine and five manipulated examples, then record every failure.

6) Metrics

• Percentage of submissions with complete provenance

• Average verification turnaround time

• False-positive and false-negative rates

• Percentage of cases independently corroborated

• High-risk escalation volume

• Reviewer agreement rate

• Audit-trail completeness

• Incidents intercepted before distribution

Pro Tip: Stop asking whether AI can detect a fake. Ask whether your organisation can prove why it trusted something.

🎯 The Arsenal — Tools & Platforms

• ExifTool · extracts metadata from images, video and documents · Link

• Microsoft Entra ID · controls access to sensitive evidence and workflows · Link

• Azure AI Search · retrieves corroborating evidence from approved organisational sources · Link

• C2PA · open technical standard for content provenance and authenticity · Link

• Microsoft Teams Approvals · routes consequential verification decisions to accountable humans · Link

Copy-paste prompt block:

You are designing a digital provenance and synthetic-media verification workflow for my organisation.

Context:
- High-risk content types: [LIST]
- Current intake channels: [LIST]
- Trusted internal records: [LIST]
- Relevant regulators or policies: [LIST]
- Available Microsoft systems: [LIST]

Design a system that:
- preserves original evidence
- records source and chain of custody
- extracts and validates metadata
- cross-checks claims against trusted records
- uses AI only for triage and anomaly detection
- requires human approval for consequential decisions
- maintains role-based access and a complete audit trail

Return:
1. architecture
2. intake schema
3. verification workflow
4. escalation matrix
5. permission model
6. test plan
7. operational metrics

💡 Free Office Hours

Synthetic media, sovereign infrastructure and concentrated AI power are quickly becoming board-level concerns. The practical move is to establish what your organisation trusts, how it verifies it and who remains accountable when certainty runs out.

Book here: https://calendly.com

Speak naturally. Send without fixing.

Wispr Flow turns your voice into clean, professional text you can send the moment you stop talking. Not rough transcription you have to clean up. Actual polished text — ready for email, Slack, or any app.

Speak the way you think. Go on tangents. Change your mind mid-sentence. Flow strips the filler, fixes the grammar, and gives you text that reads like you spent five minutes writing it.

89% of messages sent with zero edits. Millions of professionals use Flow daily, including teams at OpenAI, Vercel, and Clay. Works on Mac, Windows, and iPhone.

Start flowing free

🕹️ Game Over

Soon, “I saw it with my own eyes” may carry roughly the same evidentiary weight as “a bloke on Facebook told me.”

Build the trust layer now.

— Aaron Automating the boring. Amplifying the brilliant.

Subscribe: link