Ethics Insight
Menu

The Integrity Gap

Political risk, click-based chaos, pensioners writing AI regs, and another performative policy

Political exposure, streaming fraud, AI misuse and the gap between a published code of conduct and useful governance.

First published in The Integrity Gap on 2026-09-24. Read the original on Substack.

Welcome to this month’s long-form edition of The Integrity Gap. For new subscribers, the general format is:

What We’re Seeing - Political risk metastasises

Political risk is very much back (and has been for a few years now), but it’s no longer about “places you can’t drink the tap water”; so-called mature markets (and institutions) are as unstable and unreliable as I’ve ever seen them.

  1. Political exposure: Saima Wazed, the World Health Organisation’s Southeast Asia head and daughter of ousted Bangladeshi Prime Minister Sheikh Hasina, resigned on Wednesday, according ‌to her resignation letter (source: Reuters), following fraud investigations by the U.N. agency and Bangladeshi authorities. It’s suggested she jumped before being pushed, following an investigation into allegations of fraud and forgery related to travel to China. Lesson: fiddling with your travel expenses tends not to attract strict censure from unaccountable transnational bodies (in my experience). So, expect to see more politicisation of multilateral rule-setters and opinion shapers.

  2. Donations, donations: As the UK press devours crypto donations to the Reform party, two prominent schools of debate emerge, both equally depressing. Response 1: It’s not so bad; all the other parties have received fat donations from shadowy, bullying donors, which then shape policy. The evidence presented here points to Reform’s generally infantile policy agenda, except for its crypto stance, which is, ahem, remarkably well articulated. Response 2: If donations remain as low as they are in the UK, opposition parties can typically muster only two or three aides for each shadow minister, which means that when they come to power, policy formation is as you’d expect (basic and amateurish). So, raise the cap and allow the kind of industrial lobbying that has worked so well for big Ag, Tech, and Pharma elsewhere 😬. Lesson: Both perspectives are right and wrong, but the subtext is the more worrying part: our political system is (generally) failing to attract talented people with experience in the ministries over which they would preside. So, having worked in Singapore for over a decade, I’d suggest upping the salaries of MPs to make it a job that attracts genuine talent, not the C-grade dross of CV fabricators, con men, and general decepticons who predominate.

  3. Let them eat cake: As tax agencies and other bureaucratic behemoths continue to make life miserable (running a small business in the UK, from an administrative angle, is a study in masochism), they will continue to break their own rules. HMRC (which collects UK tax) hasn’t passed its own audit in two decades. Sit with that for a minute. Lesson: Stop assuming regulators know what they’re doing or should be trusted. Verify EVERYTHING.

Overall lesson: If you’re in business, find ways to succeed despite the political and bureaucratic climate, not because of it. If you’re investing in businesses, ALWAYS back those that don’t rely on political connections or systems.

Case Study - Beware businesses valued on clicks

The music industry has long been irked (understandably) by streaming fraud.

How does it currently work (with thanks to the analysts at Finimize):

  1. Music streaming services collect subscription and ad revenue, set aside a portion to pay royalties to rights holders, and then divide it according to each owner’s share of total streams.

  2. Scammers have been uploading AI-made tracks (cheap) and unleashing bots to play them on repeat. Platforms can treat those plays and the “artists” as real, paying them royalties and leaving less for legit artists with real listeners.

  3. Fraud may account for up to 10% of streams. So the big labels’ global trade body is mustering support for tougher checks on who’s uploading what. The likes of Universal, Warner, and Sony are supporting the initiative (unsurprising, given that they each own distributors). But without streaming platforms’ buy-in, impact could be limited.

Music is just the tip of the iceberg

Streaming fraud and bot-made tunes are already devaluing music (and other art). The share prices tell the story: Universal and Warner are both down more than 30% over the last five years.

That grim picture will continue until the industry can curb fraud and counter AI. The issue doesn’t stop with music (or broader art). I have seen endemic and systemic fraud in industries including review sites, product listings, influencer accounts, and apps.

Lesson: Buyer beware. Beyond the obvious woes of industries directly threatened by AI, tread with extreme caution when investing in (or considering working for) industries that rely on clicks, reviews, stars, or streams to justify their existence.

The Reality Check - Are we too late on AI?

As some of the AI behemoths consider cooperation around loose regulation, it’s time to look at why they might suddenly be having this slightly unconvincing ‘come to Jesus’ moment:

  1. Anthropic’s latest threat report revealed that Claude had been used by Iran-linked operations to target US naval assets, by a Yemeni insurgent group to guide rockets and missiles, by a China-linked entity to hunt Uighurs in Syria, and by a super-surveillance project in Mali. They pat themselves on the back for Claude’s refusal to research a biohazard virus, but played down the suggestion to use a rival model with weaker controls. Why care: OpenAI and Anthropic, while huge, do need US government and popular support, for now.

  2. Chinese labs are doing what we might fear: stealing. Alibaba, DeepSeek, Moonshot, and Xiaomi have all been trying to pinch Anthropic know-how using fraudulent accounts.

  3. The rest: Some of the other attempts (e.g., 4700 dating app personas, rebuilding malware to evade antivirus detection, and cloning an activist’s writing style to target their contacts) suggest that organised criminals, security agencies, and general scammers will continue to use AI to brew up bad things.

I’m not altogether convinced regulation will help. I’m not saying we shouldn’t try. I am pointing out that the average age in the Senate is 64, and in the House of Representatives, it is 58. The adoption rates in those two age brackets range from 20% to 40%. In other words, for any regulation to work, it must NOT be written by the people who usually oversee regulation.

Secondly, even if the US leads (with or without Europe), China and the rest don’t seem to care. As the Online Safety Act here in the UK, and other attempts elsewhere, have shown, getting around steps to capture someone’s ID is relatively easy with a VPN and a few YouTube walkthroughs. So, I don’t fancy our chances of mandating a ban on rogue models.

Instead, and maybe I would say this, I feel that those of us involved in fighting these sorts of wrongs (fraud, criminality, and general unpleasantness) should start to train and arm the model(s) that do enact sensible guardrails to beef them up against the corrupt models.

What’s Next? Oh, great, another fcuking policy

What I don’t think will work is another wretched policy. Microsoft have released an AI code of conduct.

It’s pure PR sh!t masquerading as governance. Why?

  1. It’s not even used: “we are not using it to train our models today.” Instead, it’s a “north star”; I have some ideas about where they might find that north star 🍑💩. And even that north star is utterly fungible: “It is not a guarantee of present-day performance.”

  2. It applies to nothing meaningful: It applies to “MAI Models,” when most of Copilot is powered by OpenAI or Anthropic models with a Microsoft skin. So, it’s a bit like H&M’s Conscious Choice” line (which was less than 10% of volume vs the usual fast fashion dross), or Apple’s Zero-Carbon Manufacturing Charter, which only focused on an Apple Watch line (then less than 20% of sales).

  3. It’s ungoverned: There are no consequences for violations, no monitoring or auditing, and no independent oversight; it “does not substitute for internal or external safety, legal, or governance processes.” In contrast, try ending your contract with Microsoft early to understand how vociferously they can enforce consequences.

  4. It’s waffle: The Code is peppered with phrases I have come to recognise as the distinct scent of bullshit, including “meaningful human oversight,” “proportionate,” “reasonably asked,” and “human flourishing” (whatever the fcuk that means).

If I read a policy like this during due diligence, I’d have a few follow-up questions, initially marking it as “lacking any substance, performative, and self-assessed.” Lesson: building on the point above, even if AI behemoths do commit to self-regulate, we should watch out for the holy trinity of bollocks:

  1. Bollock 1: no teeth (no consequences, which usually means a lack of independent oversight).

  2. Bollock 2: more back doors than Buckingham Palace (~1,500), where the regs give every possible opt-out, carve-out, or exemption to ensure that, ultimately, nothing is really captured.

  3. Bollock 3: protectionist (anything that focuses on creating barriers to entry for new market entrants).

Quote

“Show me the incentive, and I’ll show you the outcome.” Charlie Munger

More Ethics Insight writing