AI Reputational Risk: Why Bosses Fear Bad PR More Than Hackers

For years, the big fear around artificial intelligence was hacking. Now company bosses seem to be more nervous about something else: looking bad in public. AI reputational risk has overtaken cyber threats as the thing keeping boards awake.

The numbers behind AI reputational risk

In 2025, 38% of S&P 500 companies named reputational risk from AI as a material concern. Only 20% flagged AI-specific cybersecurity threats. The worry has moved from the network to the public eye, from code to copy, from breaches to brand.

Part of the reason is how customers behave. More than half will switch after a single bad experience, and they will not wait for a fix or trust a statement. That makes every public AI touchpoint, from a chatbot to an advert, something to treat as carefully as published copy rather than an experiment.

Mistakes happen in public

A made-up legal citation, a biased image, an advert using AI models without clear disclosure, a bot promising a refund the company does not offer. None of these are security breaches, yet each can do damage fast, often before lawyers can respond.

The pattern is familiar. A firm uses AI to speed up service or content. It works until it does not, and the error appears in front of thousands of people. A screenshot spreads, a journalist spots it and a legal claim may follow. By then, the story that the AI is unreliable or careless is already set. Some real examples:

• Air Canada: a website chatbot gave a traveller misleading information on bereavement refunds. A tribunal awarded damages, and the airline was held responsible for what its bot said.
• Anthropic: in a copyright case, its lawyers admitted that an AI-formatted citation had the wrong title and author, describing it as an embarrassing, unintentional mistake.
• Google Gemini: images of people that were historically inaccurate went viral, so Google paused people-image generation to fix accuracy and bias before bringing it back later.
• Guess and Vogue: AI-generated models in a US advert drew criticism. Disclosure was included but tiny, which fed concerns about authenticity.

The damage inside the company

Reputation also suffers internally. Hastily produced AI memos and presentations, sometimes called workslop, can wear down trust between colleagues and quietly lower standards. Roughly half of employees see co-workers who send such material as less creative, capable and reliable.

What good governance looks like

Anirudh Agarwal, CEO of OutreachX, argues that boards are not worried about a technical flaw so much as a moment with no context, where a wrong answer reaches many people at once. His remedy is not a better model but evidence-based governance:

• Disclose AI use by default
• Keep a human reviewing public outputs
• Maintain logs showing where content came from
• Have a real kill-switch ready

His advice is to treat every AI touchpoint like a press release, not a prototype. Some see this as a control similar to financial reporting, not just an IT or legal tick-box.

Why this matters beyond big business

Even though the figures come from the largest US companies, the lesson reaches much further. Any business that puts a chatbot on its website, drafts adverts with AI or sends out machine-written documents is putting its name behind what the system says.

The simplest protection is to decide in advance who checks what goes public, how you will tell people AI was involved, and how quickly you can switch something off if it goes wrong. Those answers cost very little to write down, and they could save a great deal of trouble when a mistake does slip through.

The bottom line is that trust is now part of the cost of using AI in public, and brands that move quickly without review may learn that lesson the hard way.