Michael Webb and Lawrie Phipps.
decide that it shouldn’t have said that, and it will disappear saying, ‘okay, I don’t want to go there’. It’s easy for government to interfere at that level, rather than with the model, which is way more complex.”
Neutrality
“The last one I think this is the most inter- esting one,” Michael said: “It is how they try to determine the neutrality of their model.” He points to a research paper by Open AI (
https://openai.com/index/defin- ing-and-evaluating-political-bias-in-llms/), which explains why neutrality is important: “People use ChatGPT as a tool to learn and explore ideas. That only works if they trust ChatGPT to be objective.” But Michael says the research is telling because it shows how OpenAI evaluates neutrality: “They took about 500 state- ments based on political issues in the US. “They graded them on a US-centric scale of liberal/conservative, extreme to not, with the middle-ground being in-between, and when it presented what they consid- ered a balance between both sides, they said it was neutral. And that’s quite clearly a very specific framing of this affair.”
Behaviour
Lawrie Phipps highlights the cultural and behavioural impact, saying: “That’s not actually to do with the models. It is to do with the marketing around the models and the way that some of the narratives around AI are pushing us into certain directions.
April-May 2026
“One of the things that I think has become really strong is the ‘inevitability narrative’ around ‘this is going to happen’ and ‘you’ve got to catch up’… because that’s not a well-researched, well-thought through technology perspective. It’s actu- ally a really effective marketing strategy that the AI companies use.” He also described what he called the “absence of the AI”: a shift in behaviour where people change how they write so they do not sound machine-generated. He suggested this may be because large lan- guage models tend to produce statistically average language, meaning writing that sits close to those norms can be seen as artificial, prompting some people to adopt less predictable styles to signal that they are human. One example is how some people have stopped using em dashes, because it is a common punctuation in machine generated outputs.
Layers of power
In a closer look at the AI power ecosystem Rebecca Flook explored what the disa- greement between Anthropic and the US Dept of Defence had revealed. Anthropic refused to remove guardrails within their contract in relation to domestic surveil- lance.
It was then blacklisted by the US gov- ernment in what looked like a straightfor- ward cost to sticking to its principles. But Rebecca said: “In the same week we saw Open AI sign a deal with the Department of War, which led to users asking what
they had compromised on. As a result, we saw a 295 per cent surge in ‘uninstalls’ for the ChatGPT mobile app while Anthropic moved to the top spot – an interesting overview of where power sits in every layer of AI ecosystem, from governments to tech vendors to users.”
Michael agreed, saying UK government pressured Grok over sexualised images showing that “the government is not powerless,” but their power is not in the model. For example, when US govern- ment officials were asked how they knew if Claude would follow Anthropic’s world view or the US government’s when it comes across a complex moral system, Michael said: “It was almost the wrong question because it absolutely will make moral choices based on Anthropic’s values. So, the reality is without transpar- ency, a huge amount of power sits with a small number of people who define the values of the AI companies.”
Research concern “My concern right now,” Lawrie Phipps says, “is mostly around government inter- ference in models (and tools). Most of the models are located in one country and the geopolitical stability of that country is not what we want it to be.” He said: “Bringing that back to education – what happens when the US Secretary of State for Health has said things about vaccines that perhaps in this country, we might take a different view of? What happens when pressure is exerted on companies
INFORMATION PROFESSIONAL 31
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60
