Summary of "Trust in AI Is Falling...What Happens Next?"

Summary of main arguments and coverage

Trust is becoming the central issue as AI spreads widely. Deepak Seth argues that people are increasingly unable to distinguish “reality, fact from fiction,” which risks “vitiating” the value of AI adoption across individuals, enterprises, and governments.
“AI-free” content is emerging as a cultural and market shift. Seth frames this as an “AI free trust revolution” analogous to familiar consumer labels (e.g., BPA-free, gluten-free, organic). He suggests formal AI-free certifications may appear in the next 1–3 years, citing a BBC story about rising demand for AI-free labeling.
Early market behavior already reflects this demand.
- Creators (artists, authors, musicians) are self-labeling work as AI-free.
- Some audiences are boycotting creators who use AI.
- Seth notes that third-party or organizational certifiers could emerge, and that trust in a label will likely depend on trust in the certifying body.
AI-free may command a premium in some domains. Seth compares it to the value of human-made originals (e.g., a Van Gogh original vs. a machine copy). He suggests some content categories—printed works, movies, etc.—may support premium pricing for “human touch” or AI-free outputs.
Organizations can’t simply stop using AI—so governance becomes the balancing mechanism.
- Seth emphasizes a graded approach: not all tasks should require the same level of AI involvement.
- Enterprises must decide, task-by-task, where 0% AI (human only) is required, where 100% AI is acceptable, and where some mix works.
Strong AI governance must include continuous monitoring and adaptation.
- Governance is described as more than enforcement/compliance; it involves engagement, enablement, and enforcement.
- Ongoing observability is critical because models can change behavior over time (e.g., becoming biased or hallucinating when they previously did not).
- Governance needs people/process/technology: committees/councils, controls and guidelines, plus governance and trust/risk/security tool stacks.
- He uses metaphors (air traffic control, traffic lights) to argue governance may lag technology, but not “too far behind.”
High-profile incidents are treated as warning signs that partial trust is not enough.
- Examples cited:
  - OpenAI’s Sora: prompts generated outputs with concerns like bias/hallucinations and copyright-related issues; Disney allegedly walked away from a deal.
  - Hatchet/book publisher: reportedly recalled a novel because it contained too much AI-generated content.
- The takeaway: organizations must calibrate AI usage based on audience needs and also legal liability.
Liability and “misplaced trust” are explicit risks for producers, not just consumers.
- Seth describes an incident where a consulting firm in Australia allegedly faced a major fine after AI-generated hallucinations appeared in a government report.
- He argues that creators may assume AI outputs are accurate and publish them—until the problem is discovered, creating real consequences.
“Context is king” is positioned as the solution layer connecting trust and governance.
- Seth reiterates a Gartner theme: data provides “what,” context provides “how.”
- For AI-free or mixed AI scenarios, context helps define guardrails for how the AI should respond and what level of AI involvement is acceptable.
- Without context, AI can be analogous to dangerous tools used without proper framing (his monkey/sword analogy; also a “text without context is trouble” quote).
Action guidance for CIOs/AI leaders: “Build trust.”
- Seth’s direct advice: two words—build trust.
- Trust is built through:
  - verified content
  - good provenance
  - outputs that are relevant/valid for intended value
- He expands trust to agentic workflows, where AI must “trust” other components in a chain (AI-to-AI trust), not only humans trusting AI.
Advice to vendors: ensure human takeover/interoperability is seamless.
- Using an aviation metaphor: if a human must take over, it should work smoothly.
- Removing humans from the loop without designing for handoffs can cause disasters.
- He recommends designing for human-plus-AI realities rather than “human out-of-the-loop” assumptions.
Future operating model: move from “human in the loop” to “guardian agents” and escalation.
- Humans can’t verify every AI action continuously (human time is limited).
- Seth references a Gartner concept: guardian/validating agents that review AI outputs and escalate true exceptions to humans.
- He expects humans to remain crucial for judgment, especially for complex or high-stakes tasks, while acknowledging some tasks may eventually allow more automation.
Closing reference: “Catch-22” as a lens for AI governance paradoxes.
- Seth recommends rereading Joseph Heller’s “Catch-22.”
- He uses it to illustrate how AI proliferation creates Catch-22-style contradictions (e.g., needing AI benefits but avoiding it for high-stakes decisions due to hallucination risk).