The reputational baggage of AI providers

About half of AI users operate (for want of a better term) at least one of the main large language models - ChatGPT, Gemini, Claude, or Copilot. Like many products, these tools carry reputational heft as a result of who is providing them. For organisations deploying these models, that weight is worth thinking about.

A question of values

In February this year Anthropic - the company behind Claude - turn down a United States government contract linked to citizen surveillance and weapons development. OpenAI, the company behind ChatGPT, took the contract on.

While there was an immediate public backlash and a sharp spike in people uninstalling ChatGPT, we have no reliable data for total users who permanently left the platform[1]. It remains the case that ChatGPT has far more users than Claude (hundreds of millions compared to tens of millions).

But the episode is a useful reminder about AI providers. They make choices. And the risk to organisations that use their tools is that they become associated with those choices.

The impact of reputational association

What is the cost of that association?

While we have lots of evidence that AI incidents – in particular those involving privacy breaches - can cause reputation damage for the adopting organisation, we’re on much shakier ground when it comes to the ripple effect of the actions of AI firms specifically[2].

Data centres and community backlash

Perhaps the strongest clues can be found in relation to AI infrastructure - rapidly becoming contested ground.

We’ve seen protests against AI data centres in many countries, including the UK, Ireland, Mexico, the Netherlands, Spain, Portugal, and the United States, often driven by water and energy concerns.

In parallel, fears about AI’s environmental impact appear to be increasing, with polling in the US, for instance, showing broad public worry[3].

This may well be a correlative rather than causal relationship. And the question remains as to whether environmental protests would have an effect on the organisations using AI.

My prediction – and my predictions are often wrong – is that the campaigns against data centres will, over time, contribute to a widely held belief that AI is environmentally harmful. As I set out in ‘AI in PR’, I think this in turn will, all things being equal, make it difficult for some organisations to square (among stakeholders) their AI use with environmental commitments.

The geopolitical dimension

There is a structural risk that, I think, deserves attention, particularly from countries, like the UK, that have yet to develop ‘sovereign’ LLMs. The dominant AI providers are US-based, operating under US law and (as we saw in February) subject to US government influence. At a moment when the transatlantic relationship is under unusual strain, that creates a vulnerability for UK organisations. It’s not too difficult to imagine that if relations deteriorated further, the consequences might include higher price points for non-US users, throttling of outputs, changes to data handling, or - in an extreme scenario - interruption to service altogether. Organisations that have committed to specific tools or providers may find those commitments difficult to honour, and both operationally and reputationally difficult to unpick.

Additional considerations

There are further dimensions of provider reputations that organisations deploying AI might want to consider in their risk assessments, including:

• Poor labour practices – as Kate Crawford maps out (sorry) in her ‘Atlas of AI’, major providers have relied on low-paid workers in the Global South for content moderation and model training[4].

  
  

• Governance instability - the OpenAI board crisis of 2023, in which CEO Sam Altman was removed then quickly re-instated, demonstrated that major providers can be subject to sudden, unpredictable internal ruptures. As with the geopolitical challenge, that is both a reputational and an operational risk.

  
  

• Financial fragility - some providers are burning cash at extraordinary rates, raising questions about long-term viability.

• 
  

A change in behaviour?

It doesn’t necessarily follow that reputational harms caused by AI provider ‘activities’ would lead to a change in the behaviours of organisations suffering those harms. They may decide that productivity gains outweigh stakeholder criticism, for instance. There may be little to choose, reputationally, between the big LLM providers and ditching them all might be too hard a choice.

Reputation does not always translate into immediate commercial damage, especially for firms selling products or services considered necessary, where customers have few alternatives and/or the product is habitual. All might be considered to apply to the general revolutionary technology that is AI. In other words, adoption may continue at pace. Heavy baggage doesn’t always slow things down.  

A version of this blog was published on my 'AI and Reputation' Substack on 6 May 2026.

[1] TechCrunch (2026) ‘ChatGPT uninstalls surged by 295% after DoD deal’, TechCrunch, 2 March

[2] Holweg, M., Younger, R. and Wen, Y. (2022) ‘The reputational risks of AI’, California Management Review, 64(3), pp. 68–88

[3] AP-NORC and EPIC (2025) ‘Public attitudes toward climate policy, technology, and environmental impacts of AI’, AP-NORC Center for Public Affairs Research, 22 October

[4] Cawford, K. (2021) Atlas of AI: Power, Politics, and the Planetary Costs of Artificial Intelligence. New Haven: Yale University Press.