When AI is the crisis: reputation and credibility when the machines make the mistake

I opened my session at the PWorld Crisis Communications Bootcamp in Toronto with a disclosure. I told the room I use AI in my work and I showed a slide on how I used and verified the information.

Then I asked two questions.

How do you feel right now about my presentation?

Does my disclosure impact how you will receive my insights?

The reaction was mixed. Some people said my disclosure improved their trust. Others said it made them immediately skeptical. Either way, I knew the pressure to not sound like an LLM had just gone way up.

It was a deliberate point, and one that is increasingly impacting trust and reputation. What happens when your audiences learn AI was involved in a decision, a message, a recommendation or a mistake?

Disclosure matters and it matters more when something goes wrong.

A familiar scenario

A board gets an update. The team presents AI-assisted market analysis that projects a significant drop in demand. The approved response involves cuts and layoffs.

Weeks later, on an earnings call, someone asks a basic question about the assumptions behind the forecast. The answers are hard to find. After digging around, the team discovers the AI tool produced incorrect information, including two reports that didn’t exist.

The first responses are predictable.

• Experimental tool
• Third-party technology
• We are reviewing the system

This scenario is fiction, but it points to a pattern we’re seeing in crisis response related to AI.

The early cyber years are a warning

In the early years of major cyber incidents, many organizations responded similarly. Sophisticated attack, bad actor, vendor issue, or we take this seriously. The organization positioned itself as the victim, blamed the technology, or the vendors and avoided accountability.

The result was a loss in trust, and eventually regulations around disclosure and communications.

We’re seeing this with early AI incidents or mistakes. Blame the tool. Downplay the impact. We’re abandoning the basic principles of crisis communications: own it, acknowledge I and fix it..

Trust is already uncertain

Most organizations are moving fast on AI adoption. Governance not so much. In fact, the best way to put people to sleep is to start talking AI governance. I’ve started leaning into verification to get people to pay attention.

But while we’re all using AI more, trust is declining. A recent KPMG report indicates most people use AI regularly (66%), but trust is much lower (46%) and there is growing support for regulation (70%). The same report shows that 66% of respondents rely on AI output without evaluating accuracy, even while 56% admit they have made mistakes at work because of AI.

Inside organizations, most leaders say ethical AI matters. Most teams are using AI. And yet the 2025 CPRS State of PR in Canada study showed less than a third of communications and public relations professionals reported having a well-established AI policy.

What makes it more interesting is that another recent academic study indicated that prior disclosure on AI use reduces trust, but exposure, being found to have used AI after the fact has an even greater impact on trust.

I feel the disclosure dilemma is relatively short lived. As AI becomes more prevalent, disclosure will be expected. Those who disclose now may experience a slight dip in in trust but will rebound quickly and avoid the larger drops that come with being found out.

But that’s not what’s happening today. Organizations avoid disclosure until they are forced into it and then it comes with ‘disappointment’ in the technology and a lack of acknowledgement of the impact. Quick tip – your AI chatbot should be as locked down and accountable as your human spokespeople. 

At the end of the day, do you want disclosure to happen on your terms or through screenshots, speculation and mockery.

AI crisis communications is still crisis communications.

There’s no secret to AI crisis communications. Here is the framework I shared in Toronto.

Identify it. Map out where AI is being used. Not just the final output. Look at prompts, datasets, tools, review steps and any decisions influenced by the output.

Assess it. Assess impact before you draft language. What changed because of the AI output? Who was impacted? What decisions were shaped? What public-facing content, financial analysis, safety guidance, or policy decisions were influenced?

Own it. Accountability sits with the organization. “The tool did it” is not good enough. “We are investigating” is not a substitute for responsibility. Ownership means clear responsibility for the decision and the next steps.

Fix it. Correct the output and deal with the consequences of acting on it. If AI influenced layoffs, pricing, safety guidance, or services, the fix must address those consequences. And if you’re ‘standing by the decision’, you’d better have credible data to back that up.

Prevent it. Build verification and escalation into your workflow. Decide who signs off, when AI use is disclosed, and how. Practice an AI crisis scenario under time pressure, the same way many organizations now practice cyber simulations.

In almost every AI incident I see discussed, the same friction points show up:

• AI is used across the organization often without leadership visibility. IBM reports that close to a quarter of Canadian employees using AI at work are doing it on non-enterprise tools. This means your company information is likely sitting in the free version of ChatGPT, Claude or Gemini.
• Verification is treated as optional and everyone assumes someone else has done it
• Legal, comms and IT are all talking about it, but there’s no shared plan, and likely no risk mapping of AI use across all functions.
• Leaders default to vendor language, or anthropomorphize the tool saying, ‘it surprised me’ or ‘it let me down’, eroding accountability.

If you are a CEO, a board member or the head of communications, you need to understand where AI is being used, what verification exists and who owns what. Here are five questions to get you started:

1. Where is AI being used in your organization?
   Can you name the functions where AI is influencing public-facing content, analysis, or decisions?
   
2. What verification do you have in place?
   What guardrails exist before AI-assisted output can shape public communication, financial analysis, safety guidance or policy decisions?
   
3. Who is accountable if AI gets it wrong?
   Hint: it is not your vendor.
   
4. What is your policy on disclosure?
   When do you proactively disclose AI use?
   
5. Have you practiced an AI crisis scenario?
   Have you run an AI-specific crisis simulation that tests verification, escalation, and messaging under time pressure?

If you can’t answer these quickly, you’re likely at risk. The point is to see it clearly and take steps to reduce that risk and be ready when, not if, the machines make a mistake.