Burnham ally ‘used AI’ to make case for tax rises

An ally of Greater Manchester mayor Andy Burnham is facing scrutiny after claims they relied on an artificial intelligence tool to help draft arguments in favour of tax rises. The episode has ignited a wider debate in Westminster over how AI is creeping into policy work—and what transparency the public should expect when algorithms are used to shape decisions that hit their wallets.

Details are still emerging and the identity of the individual has not been confirmed publicly. At the time of writing, neither the mayor’s office nor the alleged ally has set out a full account of what was generated, how it was used, or whether any AI-produced material was presented as authoritative “evidence.” What is clear is that the mere suggestion of AI-assisted justification for higher taxes is proving politically explosive.

What we know—and what we don’t

• Reports circulating in Westminster suggest an AI tool was used to help assemble or draft talking points in support of raising taxes. It is not yet clear whether this involved modelling likely impacts, summarising external research, or producing a narrative case.
• No official documentation has been released to show how the model’s outputs were verified, what sources it drew from, or whether human checks caught potential errors or bias.
• There is, so far, no evidence that AI systems were used to make the final decision—only to inform or frame the case for it. Nonetheless, critics say undisclosed AI assistance risks misleading stakeholders if outputs are presented as independent analysis.

The tech angle: Why AI can skew a policy case

Generative AI can be a powerful drafting assistant, but its strengths—speed, fluency, and confidence—can also mask weaknesses that matter in public finance:

• Hallucinations and overreach: Language models are prone to inventing citations or misrepresenting studies, especially on niche fiscal questions.
• Outdated or opaque sources: Many models are trained on data that may be out of date or poorly documented, making it hard to trace where claims originate.
• Optimism bias: AI often produces polished, definitive prose that can downplay uncertainty or distributional trade-offs in tax policy.
• Cherry-picking: When prompted carelessly, models can surface only supportive evidence, entrenching confirmation bias.

Used responsibly, AI can still help: it can summarise lengthy consultations, compare international approaches, and draft plain-English explanations of complex rules. But this requires rigorous human oversight, clear provenance for claims, and explicit disclosure of when and how AI assisted.

Ethics and transparency: The bar for public-sector AI

While organisations differ in their rules, there is broad agreement on a few guardrails when AI informs public policy or communications about taxation:

• Disclose assistance: If AI helped draft briefing notes or consultation documents, say so. Hidden use undermines trust.
• Maintain human accountability: Elected or appointed officials must take responsibility for the content—AI is a tool, not a decision-maker.
• Show your workings: Provide citations to underlying data and studies, not just an AI’s summary, and keep an auditable record of prompts and outputs.
• Validate claims: Independently fact-check model outputs, stress-test assumptions, and model impacts on different income groups.
• Protect data: Avoid feeding confidential or sensitive material into third-party tools without proper safeguards.

In the UK, government guidance already encourages transparency when automated systems assist in public services and decision support. Applying that spirit to political communications about tax would help avoid future rows.

Reactions and next steps

The allegations have prompted calls for clarity on the role AI played, if any, in shaping arguments for higher taxes. Opponents say undisclosed AI use risks manufacturing consent; supporters counter that digital tools are now a routine part of modern policy work, akin to using spreadsheets or search engines, provided humans ultimately verify the content.

We have asked the mayor’s office and individuals linked to the claims for comment about what tools were used, whether AI outputs were disclosed, and what validation steps—if any—were applied. This article will be updated if substantive responses are provided.

The bigger picture: AI is already in the room

Whether or not AI was central in this case, the controversy underscores a reality: generative models are quickly becoming standard-issue in political research, campaigning, and government communications. The question is no longer whether they will be used, but how to set the rules so they support better decisions without smuggling in errors, bias, or fake certainty.

What to watch

• Disclosure standards: Will parties and public bodies adopt formal labels or audit trails when AI assists in drafting fiscal documents?
• Verification pipelines: Do teams put model outputs through independent analysis, reproducible code, and expert review?
• Procurement and security: Are approved tools being used with proper data protections and impact assessments?
• Public engagement: Will AI help explain tax trade-offs to citizens—or be used to spin them?

For now, the lesson is straightforward: if AI touches sensitive policy arguments—especially on taxation—say so, show your sources, and make the human reasoning visible. Anything less invites exactly the kind of skepticism this row has unleashed.