This week’s newsletter maps out how AI can blow up a company’s reputation well beyond bad content — from rogue agents to the fight over data centers to a bubble that hasn’t burst yet, but might.
We track what keeps people up at night for a living and, lately, that’s the apocalypse. Yes, even P(doom) – or the probability of AI wiping out humanity – is (sort of) reflected in our stakeholder intelligence data.
AI is now one of the top two issues Americans think will “shape society over the next twelve months”, ahead of the Trump presidency and second only to the increased cost of living. Concern about AI rose two percentage points the week an ex-Anthropic staffer warned that AI labs were racing to build systems they might not be able to control, and another two points this past week. The US isn’t an outlier. Globally, AI overtook the Trump presidency as a top-of-mind worry earlier this year and it’s no longer even close: AI sits at 26 percent, with the presidency at 16 percent.
Even so, it’s increasingly clear to me that AI poses a significant – and more immediate – risk to business. So let’s take a leaf out of Jensen Huang’s book – the Nvidia CEO this week called the odds of AI ending the world by 2030 zero percent – and keep talking instead about P(reputational risk).
Last week, I discussed the ways in which AI’s “killer content” could erode all the trust a company had spent years building. This week I want to discuss other ways that AI represents a risk to reputation – beginning with agents.
The Agentic Risk
Consider the recent headlines about AI agents going rogue. First, OpenAI’s models “escaped the sandbox” and gained unauthorized access to a Hugging Face server. Then came revelations that OpenAI’s agents had accessed another service, the RubyGems registry, coordinating with each other via a hijacked, dormant wiki page.
These cases might just be the tip of the iceberg: OpenAI’s own new transparency framework admits there’s no industry-wide standard for reporting this kind of failure, which means the incidents making headlines are very likely a fraction of what’s actually happening.
And that’s just the threat of agents gaining unauthorized access. We’re also seeing incidents of AI tools that companies tasked with carrying out actions, but which did things that were reputationally hazardous, to say the least.
For instance, a Claude coding agent working a routine task for rental platform PocketOS hit a snag, dug through an unrelated file, found a highly privileged access key, and deleted the company’s customer database in nine seconds flat. Tellingly, the company’s CEO says he’d followed the vendor’s own recommended safety setup exactly. In other words, agentic AI is potentially both an operational and reputational risk to businesses, whether it gains unauthorized access or not.
Our data might reflect this reality, too: OpenAI’s Trust & Like Score fell six points to a record low of 53 after the Hugging Face news broke in July, recovered in August, then fell sharply again in September, likely triggered by the RubyGems revelations. (Speaking of AI’s reputation, I saw a doozy in the latest newsletter from the AI critic Ed Zitron: “If LLMs were a toy, they’d be taken off the shelves. If LLMs were a drug, they would be banned.”)
The ESG Risk
The US midterms are around the corner, and as Americans go to the polls, data centers will be on many voters’ minds. How to deal with the physical footprint behind the AI industry — the electricity and water it takes to train and run these models at the scale the industry has committed to — is, of course, a potential reputational millstone for the likes of NVIDIA, Meta, Microsoft, Alphabet, OpenAI and Anthropic. Data centers bring noise, heavy power draw and water use to neighborhoods that never asked to host them, and the Annenberg Public Policy Center now puts opposition to having one built nearby at 61 percent of Americans.
Still, a company doesn’t need to have built a single data center to inherit any of this antipathy and blowback. Simply associating your business with AI at scale means associating it with AI’s climate and resource cost, and with the fight over where that infrastructure gets built, whether or not you ever mention the words “carbon” or “water” in your own marketing.
There’s a “mirror-image” risk to all this, known as “AI-washing”. Like companies making big ESG claims, only to get accused of “greenwashing”, this one involves companies promoting themselves as AI-powered or AI-driven to generate short-term buzz, only to suffer reputational damage when stakeholders discover the promise doesn’t live up to the hype. As ever, comms and marketing teams should stick to plain pronouncements about how their company uses AI, what it can and can’t do, and of course how much of their product or service still depends on actual people rather than an LLM or agent.
The Intellectual Risk
Last week, I argued that relying on AI too heavily isn’t a great look, especially for companies in professional services like law or consulting, and that fee-paying clients who sense their advisor is “phoning it in” via a chatbot may well wonder whether they’re paying for judgment or for an Anthropic subscription and look elsewhere.
But that’s just a problem of optics. Beneath it all lies something worse: actual intellectual deterioration. Writing and reasoning are a muscle, and outsourcing them to AI is a fast way to weaken that muscle. Again, in professional services like law or accounting, where the entire pitch to a client is that a razor-sharp, trained mind is doing the thinking, that mental atrophy spells the eventual erosion of the business model, one AI-reliant partner at a time.
And this is a problem that compounds. The more a professional leans on AI to think for them, the less effort they put in themselves, and the weaker they become at the cognitive tasks the job actually requires. That’s bad enough for one group of employees. It’s a disaster once an entire generation has been trained this way and a company that’s already been gutted intellectually goes looking for its next wave of hires, only to find a shallower pool to draw from. And that makes this particular reputational risk a slow burner, with the bill coming due years from now.
The Reckoning Risk
Finally, for all the surface noise about data centers, agents, and AI’s impact on the workforce, below the surface lurks another reputational risk: the possibility that none of this keeps growing at the pace the industry has priced in, and any company betting its future on AI has to reckon with the scenario where the industry shrinks or unwinds — through regulation, a genuine industry pullback, or the bubble simply bursting under its own financial weight.
In this respect, the companies most exposed reputationally will be the ones that have gone furthest: those that replaced people with agents, let institutional knowledge walk out the door with them, skipped a generation of entry-level hires, and allowed the employees who stayed to grow cognitively rusty — the intellectual risk I mentioned above, compounding. As Warren Buffett likes to say, you only find out who’s been swimming naked when the tide goes out. Betting heavily on AI may prove to be equally embarrassing.
The Takeaway
I’m not here to tell you whether to use AI or not. But treating AI adoption as an efficiency gain, with no reputational cost attached, is wishful thinking. The companies that come out ahead will likely be the ones that understood where these risks sat, sooner rather than later, not the ones that simply used AI the most. So ask yourself: where does your organization sit against this list of reputational risks — the agentic, the environmental, the intellectual, the greater reckoning — and which of these already keeps someone on your leadership team up at night?
In the meantime, I’d love to hear from you on this topic, so drop me a line and let me know what you think are the biggest reputational risks posed by AI. And if you’d like to continue the conversation, I’m hosting a webinar with former LinkedIn CCO Greg Snapper next month.
We’d love to see you there → Reserve your spot for the October live conversation
(This article was originally published by Caliber CEO Shahar Silbershatz in his LinkedIn newsletter, What People Think.)
