If someone truly understands AI,
understands the technology,
understands government,
and also knows how policy is made,
then, naturally, AI companies highly want this kind of talent.
The question is:
What happens if this person, who just yesterday was designing AI policy for the government, moves straight to one of the world’s largest AI companies?
The UK is currently facing this exact issue.
The main figure is Matt Clifford.
He is no ordinary corporate advisor.
Over the past few years, he has been at the very center of the UK’s AI policy landscape.
But in September 2026, he joined Anthropic.
More sensitive still, he originally planned to keep his role as chair of ARIA, a publicly funded UK research agency.
The UK Parliament’s Science, Innovation and Technology Committee directly stated:
These two roles represent a “clear conflict of interest.”
Five days later,
Clifford announced he would resign from ARIA.
On the surface, it seemed solved.
But the real issue is just beginning.
How Central Was Matt Clifford to UK AI Policy?
Let’s look at his background.
In 2023, the UK hosted its first AI Safety Summit.
Clifford represented the Prime Minister and helped organize this global AI safety event.
Later, the UK established the AI Safety Institute (now the AI Security Institute), which he was also involved in.
Moving to 2024,
the UK government tasked him with another major project:
Designing the AI Opportunities Action Plan.
This was not an ordinary industry report.
It answered questions like:
How should the UK develop AI?
How to encourage businesses to use AI?
What data, talent, and infrastructure are needed?
How should the government itself use AI?
The plan was officially published in January 2025.
It contained 50 recommendations.
The UK government announced it would adopt all 50.
Subsequently, Clifford was appointed the Prime Minister’s AI Opportunities Adviser, providing AI policy guidance directly to the Prime Minister and technology departments.
Simply put,
if UK AI policy were a building under construction,
Clifford was not just a visitor,
he helped draw the blueprints.
Then, He Joined Anthropic
In early September 2026, news broke:
Clifford joined Anthropic as Managing Director of International Affairs.
Anthropic is the AI company behind Claude.
Clifford’s new role is not about developing models or researching algorithms.
Instead, his job overlaps highly with his previous experience:
Engaging with governments worldwide.
Anthropic stated the company needed someone who truly understands how governments work because AI now must collaborate with governments, policies, and social systems globally.
This is quite reasonable.
For Anthropic,
a person who has shaped national AI policy, understands government operations, knows what officials care about, and has extensive policy contacts
is obviously very valuable.
The question isn’t why Anthropic wants to hire him.
The real question is:
How should governments handle this kind of talent movement?
The Controversy Sparked Because He Initially Planned to Keep His ARIA Chairmanship
If Clifford had fully left government roles before joining Anthropic,
the controversy might have been minimal.
But at that time, he still held another role:
Chair of ARIA.
ARIA stands for Advanced Research and Invention Agency.
It’s a UK government-backed agency investing public funds into high-risk, high-potential research projects.
Many of its investments could relate to AI or AI-driven technologies.
The problem became:
On one side, a publicly funded agency deciding the direction of cutting-edge tech research on behalf of taxpayers.
On the other, one of the world’s leading AI companies.
And the same person was preparing to serve both sides simultaneously.
Chi Onwurah, Chair of Parliament’s Science, Innovation and Technology Committee, publicly stated:
This constituted a “clear conflict of interest.”
The initial solution was to put in place isolation measures.
For example, Clifford would have to recuse himself from any ARIA business related to Anthropic.
Anthropic also said he would not access commercial discussions between ARIA and Anthropic.
In theory, this could separate the two roles.
But the political issue remained unresolved.
Because Conflict of Interest Isn’t Just About “Cheating”
It’s important to understand this concept:
A conflict of interest doesn’t necessarily mean:
Someone has done something wrong.
Nor does it mean Clifford changed any UK policy favoring Anthropic.
To date, there’s no evidence he used his ARIA position to benefit Anthropic.
The real issue is this:
Would a reasonable outside observer suspect your public decisions might be influenced by private interests?
This is why many public institutions can’t simply address conflict of interest with the assurance:
"I will be fair."
Because what public systems truly need to protect isn’t just decisions themselves,
but also:
The public’s trust in those decisions.
September 7: Clifford Decides to Resign
Days after the controversy broke,
on September 7, 2026, Clifford announced he would step down as ARIA chair.
He stated he didn’t want his new Anthropic role to interfere with ARIA’s work.
However, he won’t leave immediately.
At the UK government’s request, he will remain until November 6 to allow time for a new chair to take over, continuing to apply conflict-of-interest safeguards during this period.
Committee Chair Onwurah welcomed the decision.
But she also emphasized:
The issue is far from resolved.
Parliament still demands government explanations on:
How was such an arrangement approved initially?
What conflict assessments were performed?
What systems are in place to prevent future incidents?
This is the truly important part of the whole matter.
Because Governments Will Need “People Who Understand AI” More and More
AI poses a very tricky policy problem.
If governments don’t engage industry experts,
they may never keep pace with the technology.
Those who build models know their limits.
Those who build AI agents understand how systems might fail.
Those who run data centers know the constraints of compute, energy, and chips.
Those managing AI companies also know what the industry needs.
Governments obviously need these people.
But on the other hand,
if all the top AI experts come from AI companies,
then the rules may end up as:
Those being regulated are also the most knowledgeable about how to design regulations.
This is the dilemma.
It’s Like Having the Best Bankers Write Banking Regulations
Imagine this.
The government is creating banking regulation policy.
It obviously needs people who truly understand banking,
otherwise rules may be unrealistic.
But if the person designing banking policy for the government
joins a major bank the next day,
handling regulatory communications with the government,
society will ask:
What knowledge did they take with them?
Only professional expertise?
Or also insights about internal government thinking, who’s responsible for what, where policies are heading, and what arguments sway decision makers?
These things are hard to capture in a simple "conflict of interest" statement.
But for companies,
they are all valuable.
The same applies to AI.
AI Companies Are Now Competing for “Government Savvy” Talent, Not Just Engineers
In past years, when people talked about the AI talent war,
they usually thought of:
Researchers.
Model engineers.
Chip designers.
Product managers.
Now, there is one more critical role:
Policy experts.
The reason is simple.
Whether AI companies can launch new models,
access data,
enter government markets,
assume legal responsibilities,
require pre-market model evaluations,
or whether a country restricts certain AI types —
these questions are increasingly about policy, not just technology.
This means “understanding government” is becoming a crucial competitive advantage for AI companies.
Anthropic even has a Chief Global Affairs Officer responsible for global policy, international relations, and government affairs.
So Clifford’s move is not isolated.
It reflects that:
AI competition has shifted from models to policy and institutional competition.
But Don’t Misinterpret: Moving from Government to Industry Isn’t Wrong in Itself
This issue should not simply become:
“Government employees are banned from joining tech companies.”
That wouldn’t be reasonable.
Governments need industry experience.
Industry needs people who understand public policy.
Talent flowing between both sectors can even be beneficial.
The real issue is about:
When can people change jobs?
What must be avoided?
How long after leaving can they be involved in previous policies?
Which government information is off-limits?
What interests must be disclosed?
Who decides if conflicts are properly managed?
If these rules are vague,
the problem is not any one individual’s fault.
It’s that systems fail to keep up with the speed of talent moving within the AI industry.
The Bigger Question Is: Who Is Influencing AI Rules?
More AI regulations are inevitable, covering:
Safety, copyright, data, employment, healthcare, defense, education, government procurement, model accountability.
Each area carries massive commercial interests.
So we cannot only ask:
What does the AI policy say?
We must also ask:
Who helped write it?
Where have they worked before?
Where might they go next?
Which industries get close to decision makers?
Who doesn’t?
These might sound mundane,
but they could be more important than any single AI regulation itself.
Because mature AI governance
requires not only finding the smartest AI experts,
but also establishing systems
to ensure their expertise serves the public interest,
while maintaining clear boundaries between government and business.
Clifford’s final decision to resign as ARIA chair
resolved the most obvious role conflict.
But the UK Parliament’s probing questions
reflect a challenge every country making AI policy must soon address:
When the top AI policy experts are also the most sought-after by AI companies, how can governments retain expertise while preserving independence?
Today, take a small step forward with AI.
Learn one AI skill daily.
Save a little time every day.
Improve your abilities bit by bit.
SasaDaily, growing with you.
Recommended Reading
When AI Requires Government Review Before Launch, It’s No Longer Just a Tech Product
Why Is Parliament Quoting AI-Generated Research in Official Investigations?
Is Open AI More Dangerous or Is Concentration in Few Companies Riskier?
