Governance

Kenya seeks $20.8m for AI-powered social media monitoring system

https://www.mexc.com/en-GB/news/1111153

President William Ruto’s government is seeking $20.8mn (KES 2.7bn) for an artificial intelligence system that will monitor online discourse and to establish a National Communications Centre for government to distribute information. Ruto’s United Democratic Alliance party argues that the system and Centre will combat misinformation and improve civic outreach. A cynical interpretation is that he aims to quash digital dissent before it can manifest into in-the-streets protests — as happened during the youth-led protests from June-July 2025.

Then, Kenyan youth mobilised in retaliation to tax increases and the murder of schoolteacher Albert Omondi Ojwang in policy custody after being arrested on charges of “false publication” after sharing a Twitter post that criticised the Deputy Inspector General of Police. The protests were organised almost entirely over social media, marking the first time such demonstrations have been organised outside of political or tribal structures. During the protests, state-sponsored individual trolls and coordinated groups launched online smear and intimidation campaigns against organisers, often deploying Islamophobic and sexist language to harness resentment towards these groups.

Kenya’s digital repression is significant for its coordination with the private sector and Chinese hardware. The Safe City Project, initiated by government in 2014 to combat crime and terrorism in Nairobi and Mombasa, is a joint project between Safaricom and Huawei. Safaricom is Kenya’s largest mobile-service provider and the most profitable company in east and central Africa. The Project uses Huawei CCTV cameras and facial recognition algorithms and is financed by soft loans from Chinese banks. Its systems were deployed to track and target dissidents in the 2025 protests. Safaricom itself has provided police with the locations of digital dissidents without any court order or disclosure of this information sharing.

Clearly, this requested funding follows a troubling precedent of increasing digital repression. It still awaits final review and approval by Parliament's Budget and Appropriations Committee. If granted, it’ll mark an unprecedented leap in the Kenyan government’s powers to stifle online dissent and target its sources.

Business and markets

DeepSeek V4 Pro tops global bang-for-buck ranking after 75% price cut

https://www.scmp.com/tech/tech-trends/article/3354668/deepseek-v4-pro-tops-global-bang-buck-ranking-after-75-price-cut

Since its release in January 2025, Chinese AI model DeepSeek’s unique selling point has been its blend of cheapness and performance that wiped one trillion dollars in market value from US tech stocks in a single day. Now, its cheapness has been consolidated by a permanent 75% price cut that makes it ~95% cheaper than the models of rival companies like OpenAI and Anthropic.

But it isn’t clear whether these price cuts will stimulate demand for its services. DeepSeek already has 89% domestic market share and its international expansion faces severe headwinds — not least by being partially or fully banned by Asian countries including India and South Korea over security concerns, obstructing it from the world’s fastest-growing chatbot market in the Asia-Pacific. Contrastingly, ChatGPT and Claude are booming in the APAC, with India and Japan as Claude’s leading headcounts.

OpenAI and Anthropic have sufficient capital to heavily subsidise users’ token consumption — particularly as they approach world-beating IPO’s and court significant foreign investment — and are moving to per-token pricing structures (as I wrote last week…). With such subsidies and localised pricing plans for APAC nations, their relative expensiveness over DeepSeek is unlikely to diminish demand for them. And, they face far fewer regulatory barriers.

Last year, I made a similar argument about Chinese electric vehicle manufacturer BYD’s aggressive price cuts. BYD had also saturated domestic demand and faced slews of regulatory probes into its attempts to expand into Europe which still continue. However, domestic sales accounted for 80% of BYD’s revenue whereas at most 44% of DeepSeek activity is from China. DeepSeek has a better international footing, but it’s far from certain that these 75% price cuts will overcome the significant obstacles to its overseas adoption.

Regulation

Palantir hits back at Sadiq Khan after £50m contract with Met police blocked

https://www.theguardian.com/technology/2026/may/22/palantir-hits-back-sadiq-khan-contract-met-police-blocked

After London Mayor Sadiq Khan blocked a £50mn contract between Palantir and the Metropolitan Police, advocates of the American defence-tech firm argue that Khan is “putting politics above public safety”. They highlight Khan’s previous comment that Londoners want public money paid to companies that “share the values of our city”.

In fact, Khan’s rejection of the contract that would’ve automated some aspects of police investigations is about procurement, not politics. The Met had breached procurement rules by failing to present its procurement strategy or detailed cost-benefit analysis to the Mayor’s Office for Policing and Crime (MOPAC) before seeking to award the contract. Moreover, the Met didn’t engage with any other potential suppliers. This willingness to throw £50mn at Palantir without a formal and diversified tender process is sufficient grounds to reject the contract.

It isn’t the first time Palantir’s benefitted from clandestine contracts. Under MOPAC’s Scheme of Delegation and Consent, spending over £500k requires formal approval by the Deputy Mayor for Policing and Crime. In February, the Met awarded Palantir an almost £10k contract for a pilot project that was then extended with another contract worth almost £490,000 — coming just under the £500k threshold for formal approval.

Scotland Yard warns that Khan’s decision will force it to cut officer numbers. But the Met’s £125mn funding shortfall is a consequence of political decisions, beginning with then-Home Secretary Theresa May cutting policing budgets by 18% in 2010, not technological inefficiency. Without even providing an argument for how outsourcing policing casework to AI systems will result in significantly improved margins and officer deployment, the Met’s behaviour is both unprofessional and ill-informed.

Culture

How Saudi Arabia is using AI to transform the Hajj experience

https://english.alarabiya.net/News/saudi-arabia/2026/05/23/how-saudi-arabia-is-using-ai-to-transform-the-hajj-experience

You may have heard Pope Leo XIV’s dire warnings about AI’s economic and political disruption. Interestingly, the previous Pope Leo, Leo XIII, warned about the exploitation of workers in the Industrial Revolution in his 1891 encyclical Rerum Novarum (“Of New Things”). Only 135 years separates both Popes’ similar warnings in unimaginably different contexts. But here’s another example of AI meeting religion.

As an estimated 1.5mn Muslims make the Hajj pilgrimage to Mecca to commemorate this year’s Eid al-Adha, Saudi Arabia is again deploying sophisticated AI techniques to coordinate their movements. Led by the Saudi Data and Artificial Intelligence Authority (SDAIA) and the Ministry of Interior, the kingdom is using computer vision, thermal imaging, and AI-driven analytics from live video feeds to predict and influence crowd density and movement patterns in real time. Crowd control has become a central concern after a tunnel stampede killed 1,462 pilgrims in 1990.

In this recent Hajj, however, the leading cause of deaths has become extreme temperatures. The 2024 Hajj saw over 1,300 heat-related deaths. The following year saw a 90% reduction in such deaths as authorities took measures like updating fitness requirements and reducing wait times for prayers. This year, the General Authority of Civil Aviation has given first-of-its-kind permission to Terra Drone Arabia to use its drones to deliver medicine and water among temperatures up to 50 degrees celsius. 80% of heat-related deaths were unrelated pilgrims, and authorities have already used their camera technology to arrest multiple pilgrims entering through unauthorised routes.

Given Hajj’s significance amid the five pillars of Islam and the world’s religious festivals, the adoption of AI to meet today’s climate-related exigencies is a striking example of old meets new.

Governance

GPs and hospitals in England to be required to share data to create single patient records

https://www.theguardian.com/society/2026/may/10/gps-and-hospitals-in-england-to-be-required-to-share-data-to-create-single-patient-records?utm_campaign=The%20Week%20in%20Data%20TWID&utm_medium=email&utm_content=418930885&utm_source=hs_email

Overlooked amid Starmer’s death throes and the local elections, last week’s Kings Speech announced the NHS Modernisation Bill. One of its central proposals is a legal requirement for all NHS service providers, including GPs and hospitals, to share patient data in a Single Patient Record. Currently, some emergency information like ongoing medication and known allergies is shared. But GPs often have to wait for emails from hospitals describing their patients’ last visits, which can take weeks to arrive. Proponents of the Bill argue that this centralisation will streamline healthcare provision by eliminating these delays.

Clearly, different service providers’ patient data must be accurate before centralisation into a SPR. That’s not the case. Research commissioned by Healthwatch, an independent body that monitors NHS services’ effectiveness, found that 23% of respondents had noticed errors or inconsistencies in their medical records. 16% of these errors were about their previous medication history. A separate audit of over 30,000 surgical patient records found coding errors in 51% (!?) of admissions.

Given the fragmentation of healthcare services and their varying legal distances from the Department of Health and Social Care which will be the legal data controller and custodian under the Bill, establishing accountability for incorrect data in SPR’s will be extremely difficult. Finger-pointing and buck-passing will abound. On these grounds, the British Medical Association has argued that “while we welcome future discussions on the SPR… GPs themselves must remain the data controller of the GP record”.

The SPR proposals come at an inauspicious time for patient data security.
In February, the government granted researchers from UK Biobank, a database for the long-term study of 500,000 citizens, access to coded GP patient data. Just two months later, this data was found for sale on three separate listings on Chinese website Alibaba. Despite assurances that it was anonymised, researchers were able to identify a patient from just their date of birth and details of a major surgery.

Speaking of accountability, Former Health Secretary Wes Streeting resigned on Thursday — the day after the Bill was announced. I can’t help but notice the timing…

Business and markets

Samsung granted court injunction against imminent strike action in last-minute reprieve — talks resume as unions barred from occupying or locking facilities, obstructing workers

https://www.tomshardware.com/tech-industry/samsung-granted-court-injunction-against-imminent-strike-action-in-last-minute-reprieve-talks-resume-as-unions-barred-from-occupying-or-locking-facilities-obstructing-workers

As disputes between Samsung and union representatives of a threatened 45,000-person strike at its memory chip plants reached breaking point, Suwon District Court on Monday partially accepted Samsung’s injunction request that would limit two unions’ planned 18-day walkouts from Thursday. The injunction bars strikers from from obstructing or locking facilities and mandates sufficient personnel to ensure normal safety operations. Strikers will be fined $74,000 per day if this is breached.

One way to interpret this injunction is as the triumph of capital over labour.
It’s significant that the strikes are at memory chip plants. Memory stocks are rallying during a critical memory hardware shortage amid exploding demand. AI stocks have so far driven 80% of S&P 500 gains, and 4/5 of the best-performing US stocks specialise in memory and storage chip production. This memory over-performance is starkest in South Korea, where memory manufacturers like SK Hynix and Samsung have accounted for 70% of its KOSPI index’s 45% year-on-year growth — the fastest globally this year. Samsung’s own share price has more than doubled this year.
No wonder Suwon’s District Court stepped in.

The dispute is far from resolved. Unions have rejected Samsung’s one-time offer of a one-time bonus payout of $340,000 per employee, somewhat overshadowed by SK Hynix’s bonuses amounting of $447,000 per employee this year and $900,000 next year. But the court’s injunction against obstructing facilities and risking manufacturing equipment clearly delineate the boundaries within which strikes are permitted.

Regulation

Anthropic to share Mythos cyber flaw findings with global finance watchdog

https://www.theguardian.com/technology/2026/may/18/anthropic-ai-claude-mythos-cyber-financial-stability-board-fsb

Anthropic is to hold a closed-doors meeting with the Financial Stability Board to discuss the implications of its new model, Mythos, that it claims can unprecedentedly detect and exploit cybersecurity vulnerabilities. The meeting will convene senior economists and central bank officials from across the world, and be chaired by Bank of England Governor Andrew Bailey.

Since Anthropic previewed Mythos on April 7th and warned of its ability to find “ground-zero” vulnerabilities in banking systems and other critical infrastructure, there’s been intense speculation about its capabilities and the risk it poses to worldwide digital security. In the two following weeks, cybersecurity stocks lost billions in share prices as markets feared that the offensive capabilities of programmes like Mythos made them redundant.

But third-party analysis of the methodology behind Anthropic’s assessment of Mythos supports a far more nuanced picture than headlines suggest. A centrepiece of Anthropic’s demonstration was Mythos achieving a 72.4% code execution rate against the open-source web browser Firefox 147. Yet, Mythos didn’t actually attack Firefox 147 but rather a JavaScript shell of it which used a testing configuration stripped of important defence mitigations. The 72.4% success rate relied upon two specific and highly-exploitable bugs without which Mythos’s success rate plummeted to 4.4% which is the same as earlier models. Finally, Anthropic’s human contractors only manually reviewed 198 vulnerability reports generated by the model and then extrapolated this to the “thousands of zero-days” claim.

Ultimately, Mythos must be contextualised. It was announced on the same day that Bloomberg first reported Anthropic’s plan to list on public markets. Securing partnerships with names like Microsoft and JPMorgan Chase make it appear a critical part of global cybersecurity infrastructure. But much of the scare Mythos created rests upon extrapolated success rates, unrealistic and weakened test environments, and pre-patched bugs. Until its successful in realistic testing environments and at realistic computing costs (given the rapidly increasing token costs my last newsletter discussed), any concrete decisions made by the FSB will be unfounded.

Culture

Chelsea flower show garden designers clash over use of AI

https://www.theguardian.com/lifeandstyle/2026/may/13/chelsea-flower-show-garden-designers-clash-over-ai?utm_campaign=The%20Week%20in%20Data%20TWID&utm_medium=email&utm_content=418930885&utm_source=hs_email

Big Tech meets small town. Alarmed parakeets scatter from Chelsea’s private gardens as uproar resounds over award-winning garden designer Matt Keightly using AI to design his garden at the Chelsea Flower Show. His soon-to-launch new app, Spacelift, can reportedly create garden designs from scratch, alarming horticulturists who fear competition from such software. Other horticulturists call Chelsea’s permitting of the AI a “betrayal”.

Sure, there’s an aesthetic discussion to be had about whether using AI replaces or complements the designer’s own creativity, etc. etc. But amid all the more serious discussion of AI, I just find this a refreshingly petty squabble that reads like satire. Perhaps the next season of Bake Off will be marred by AI-enhanced recipes or touched-up pictures of cakes. Amid all the doomsday wails and ecological and societal fuckery wrought by AI, there’s some perverse comfort to be had in how people still stubbornly argue about the small stuff.

Governance


UK immigration officer among two men guilty of working for Chinese intelligence

https://www.bbc.co.uk/news/articles/c0m2wjlkzplo

In the first prosecution of Chinese spies in British history, two men of dual British and Hong Kong nationality were found guilty on Thursday under the National Security Act of assisting a foreign intelligence service. One of them, Chi Leung "Peter" Wai, was working for the UK Border Force and used his access to government immigration databases to track and even attempt to arrest Hong Kong dissidents, feeding this information to Hong Kong authorities.

Wai’s misuse of such databases since 2013 represents not only gross negligence by the Home Office and Metropolitan Police, but the culmination of the gradual erosion of restrictions on the government’s handling of sensitive data. In its pursuit of economic growth and digital IDs, the government has introduced successive legislation — most notably the recent Data Protection (Use and Access) Act 2025 — that relaxes critical compliance mandates and introduced new “legitimate interests” under which sensitive data can be accessed while circumventing the checks-and-balances mandated by the GDPR.

This weakening exposes citizens’ and political refugees’ data not only to misuse by the domestic government, but also opens the door wide to the transnational repression exemplified in this story. Securing this data while enabling necessary access to it is possible, but only with sufficient and lasting political will.

Business and markets

Perspective: AI demand is inflated, and only Anthropic is being realistic

https://www.cnbc.com/2026/04/17/ai-tokens-anthropic-openai-nvidia.html

Tokens, the words and characters that constitute user inputs and model outputs, are the basic unit of AI usage and are used to model demand. Chatting with a LLM consumes a few hundred tokens per paragraph. But the increasingly popular agentic AIs that write code, trawl the web, and execute multi-step workflows, consume 50-100x more tokens. Given this disparity, extracting the actual demand underlying record levels of token consumption is critical for AI companies to judge infrastructural build-out.

Anthropic’s latest pricing model directly responds to this increased token consumption. While it hasn’t changed the price it charges per token, the latest Opus 4.7 model increases the amount of tokens used for processing input categories like code and structured datasets. It also increases the amount used for generating outputs, and output tokens cost 5x more than input! Staff have stated that the new model can use up to 35% more tokens than before. Anthropic has also cut off third-party agentic tools that were large token consumers, alongside tools like OpenClaw through which users would route their commands to bypass their plans’ limits.

By increasing token count and removing these third-party tools, Anthropic can simultaneously increase its token revenue and better estimate demand. As OpenAI pursues its aggressive scaling strategy at a cost of $1.4tn, Anthropic has in contrast invested just $50bn in infrastructure. But it’s projected to break even by 2028, while OpenAI projects losses into the 2030’s. Correctly estimating and monetising token consumption as a proxy of demand is critical for profitability. As Anthropic CEO Dario Amodei said in a recent podcast, “If you’re off by a couple years, that can be ruinous”.

Regulation

What the EU AI Omnibus Deal Changes for the AI Act and What Lies Ahead

https://www.techpolicy.press/what-the-eu-ai-omnibus-deal-changes-for-the-ai-act-and-what-lies-ahead/

One of the most contentious and far-reaching components of the Omnibus is Article 4(a). It will operate as lex specialis alongside existing GDPR to allow the use of sensitive data for bias detection and correction in narrowly defined cases for high-risk AI systems. These systems can be private or public sector and include those use for biometrics and identification, employment and worker management, and migration and border control. Article 4(a)’s application is contingent upon strict assessments of necessity and proportionality, security measures, access controls, and deletion requirements, and doesn’t mandate such bias detection and correction.

I welcome this change. From facial recognition CCTV systems to welfare assessments, there’s extensive evidence that algorithms do bias against ethnicity, health data, and religious beliefs. Processing such sensitive data is necessary to assess for bias, but is prohibited under Article 9 of GDPR. Article 4(a) finally creates a legal route for governments and researchers to do so.

Opponents to Article 4(a) warn of privacy breaches, mission creep, and data leaks, but sensitive data can be anonymised and processed by governmental and intergovernmental bodies like the EU AI Office in narrowly defined contexts and with tightly regulated access/retention durations. I think the biggest risk is data leaks, as swathes of recent examples like the Biobank leaks attest. But there are still various technical methods of storing and processing this data in maximally secure digital environments.

Algorithmic bias is rife and algorithmic governance is necessary. Only governmental bodies can possess the legal mandates and accountability to do it.

Research

LLMorphism: When humans come to see themselves as language models

https://arxiv.org/abs/2605.05419

Dr. Capraro presents a lucid exploration into how the characteristics of LLMs and the vocabulary used to describe them may be projected onto humans. I think some of the most compelling cases are in psychiatry.

LLMs are already being rapidly incorporated into psychiatry, from extracting underlying themes from unstructured data like assessment transcripts to monitoring patients’ progress through analysing their interactions with chatbots. All instances of such use involve making people’s self-descriptions machine-readable by distilling them into statistical distributions of classifiable variables etc. etc. etc. It might seem that the question is how to ensure that psychiatrists still see their patients as more than such data while working with it.

But it isn’t the individual psychiatrist’s perspective that should be considered. Once you place LLM use in psychiatry within an administrative context and recognise the accompanying pressures towards cost-efficiency, scaling, and interoperability, it seems unavoidable that patients must be represented in increasingly abstract and formalised variables to enable them.

Now, the question shifts from what a psychiatrist thinks to how psychiatrists collectively function. In a future digitised healthcare system that so strongly tends towards this interoperability, what room is left for them to even conduct psychiatry in ways whose purposes aren’t to make patients machine-readable? And what of the tragedy of those psychiatrists who do see their patients as more than this, but who can’t act on it?