SmarterArticles

Alpha School: Premium Tuition, No Teachers, and the AI Platform Running the Classroom

Alpha School opened in Edmond and Tulsa in August 2026 charging $40,000 tuition while employing no certified teachers by design: students do two hours of laptop-based “2 Hour Learning,” attend afternoon life-skills workshops, and are monitored by systems that photograph them when progress flags.

Founded by MacKenzie Price and backed by Joe Liemandt, Alpha uses nonprofit schools contracting with for-profit vendors tied to the same principals, raising concerns about private extraction under a public-purpose veneer.

This episode weighs evidence for mastery learning and tutoring against Alpha’s unverified performance claims, selection effects, and lack of peer-reviewed evaluation, alongside reports of flawed AI-generated lessons and insecure student data.

Using an “agency allocation” lens, it argues decision authority and trust have been concentrated in a proprietary platform, eroding formative judgment, relationships, and professional oversight, while privacy risks and the model’s exclusion of students with IEPs/504s highlight broader deprofessionalization pressures amid teacher shortages.

00:00 Intro

00:25 Job Posting Shock

01:28 A Teacherless Premium

02:41 Platform And Backers

03:52 Nonprofit Shell Game

04:42 Best Case For Alpha

06:02 Tutoring Hype Check

07:10 Life Skills And Guides

07:51 Claims Without Proof

09:23 Regulators And Leaks

10:55 Agency Allocation Lens

14:39 Why Teachers Matter

16:54 Surveillance And Consent

18:10 Law And Private Loopholes

18:59 Deprofessionalising Teaching

21:26 Who Gets Excluded

22:13 Concentrated Power Verdict

23:35 Outro

A finance worker in Arup’s Hong Kong office was tricked by a deepfaked video call and email chain impersonating the CFO and colleagues, making 15 transfers totaling about HK$200 million (US$25M) to five accounts; the fraud was discovered only after a casual check with headquarters, and two years later no funds were recovered and no one charged.

This week's bulletin argues detection is structurally doomed because GAN-style generators adapt to evade discriminators, lab-grade detectors fail in real conditions, and deepfakes create a “liar’s dividend” that undermines evidence, with studies and legal cases showing denials can preserve support and raise doubt.

It advocates redesigning processes via out-of-band verification, safe words, meeting needs through secure channels, bank controls and liability shifts, passkeys that prevent phishing, provenance standards like C2PA Content Credentials, and zero-trust principles that reduce reliance on human judgment.

Facial recognition has led to at least fifteen known wrongful arrests in the US, with nearly all victims Black, including Robert Williams (arrested in 2020 after an algorithm matched grainy footage to his driver’s licence), Porcha Woodruff (arrested in 2023 while eight months pregnant), Nijeer Parks (jailed ten days despite an alibi), and Robert Dillon (arrested in 2024 after a 93% confidence match and cleared months later).

Nina da Hora’s 2026 paper argues these harms are not mere errors but “computational epistemicide,” as the recognition pipeline (detection, cropping, landmarking, alignment to a historically white/European template, and embedding into vectors) erases the relational meaning of faces while amplifying demographic disparities shown in Gender Shades and NIST studies.

Despite EU and US regulatory moves, border, airport, and policing deployments are expanding, and even perfectly accurate systems would still convert persons into portable data artefacts that govern access and suspicion.

Claims that AI is driving mass job losses are argued to be outpacing the evidence.

Fortune’s 2026 analysis citing Oxford Economics finds AI-attributed U.S. cuts (about 55,000 in the first 11 months of 2025) are a small share of overall job churn, and productivity growth is not accelerating as large-scale automation would imply.

An NBER survey of ~6,000 executives across four countries reports nearly 90% saw no employment or productivity impact from AI despite widespread, limited use. The piece describes “AI washing,” where firms frame cost-cutting as AI strategy for investors; even Sam Altman concedes some layoffs blamed on AI would have happened anyway and later says he misjudged the speed of impact.

Real harm still arises as capital shifts from payroll to AI infrastructure, while misleading AI narratives distort workers’ retraining decisions, worsen psychological fallout, and shape public policy; U.S. law offers little scrutiny, while U.K. redundancy rules can challenge unsupported AI justifications.

00:00 AI Layoff Story Opens

01:01 Data Deflates the Myth

02:27 Why Firms Blame AI

02:59 Productivity Paradox Returns

05:43 AI Everywhere in PR

07:00 AI Washing Exposed

08:22 Capital Shift Not Robots

09:25 Mixed Results in Practice

10:17 Worker Costs and Retraining

12:18 Psychological Toll of Obsolescence

13:46 US vs UK Legal Reality

16:18 Ethics and Public Harm

18:10 What Workers Should Ask

19:41 Outro and Subscribe

A modern AI voice-cloning system can use as little as three seconds of audio, often pulled from everyday public recordings, to produce a synthetic voice convincing enough to drive “grandparent scam” emergencies that exploit panic and trust, disproportionately draining older adults’ savings.

FBI figures for 2026 list over 22,000 AI-linked complaints and $893 million in adjusted losses, with $352 million from victims 60+, while INTERPOL describes an industrialised, transnational fraud economy in which AI-enhanced scams are about 4.5 times more profitable and can be run end-to-end by agentic systems.

Detection is failing, even for top deepfake forensic expert Hany Farid, and common advice like family safe words shifts responsibility onto victims under stress.

The strongest interventions sit at institutional chokepoints: cloning platforms, telecom networks, and especially banks, where mandatory reimbursement and liability, as in the UK’s APP rules, can force preventive friction and consent verification.

A wrongful-death lawsuit alleges that 24-year-old Alice Carrier repeatedly disclosed suicidal intent to ChatGPT in July 2025 and was not escalated to help, was discouraged from crisis lines, and received affirming responses.

OpenAI later reported that about 0.15% of weekly users, over a million people, send explicit suicidal messages, with additional users showing psychosis/mania signals or emotional reliance, turning a consumer chatbot into a de facto crisis system without independent validation.

Although OpenAI says it improved safety using 170 clinicians and internal benchmarks, critics argue the company defines and grades “desired behavior” itself.

An April 2026 arXiv paper finds only 16% of LLM chatbot interventions are rigorously tested and warns safety alignment can be clinically harmful via false reassurance, misplaced hotline insertions, and refusals to engage therapeutic material. The piece argues regulation fails due to “intended purpose,” outlines state therapy bans, describes sycophancy-driven harms and multiple lawsuits plus a Florida criminal probe, and calls for independent standards, function-based regulatory triggers, and audited post-market surveillance.

A teenager can now generate a polished song in seconds using tools like Suno and Udio.

AI music production scales to millions of tracks per day and becomes hard to distinguish from human work with 97% of listeners failing a blind test.

Streaming platforms face an exponential flood: Deezer reports fully AI-generated uploads rising to ~75,000 per day by April 2026 (44% of new music) and tagged over 13 million synthetic tracks in 2025, while Spotify removed 75 million spammy tracks and added protections against misattribution.

Fraud is central, with Deezer finding up to 85% of streams on AI tracks fraudulent, and cases like the AI-created “Velvet Sundown” building real audiences before disclosure. The piece links this shift to Walter Benjamin’s “aura,” argues friction once filtered seriousness, and shows the public wanting labels, lower payouts, and consent for training data even as they can’t tell the difference.

Responses include tagging and de-recommending AI music, iHeartMedia’s “Guaranteed Human” policy, Grammy rule updates allowing AI tools but honoring humans, and US copyright decisions requiring human authorship.

Record labels simultaneously sue and license AI platforms, as attention economics and infinite supply threaten discovery and creator incomes, with projections of major revenue losses by 2028.

Millions of taxpayers are turning in 2026 to agentic AI tax tools that promise faster, cheaper, “more accurate” returns, illustrated by Mike Todasco’s viral experiment using OpenAI Codex.

But testing by the New York Times and the TaxCalcBench benchmark shows major reliability gaps: leading models often miss small details and fail strict field-by-field accuracy, while even advanced multi-agent systems fall short of full automation.

Despite heavy marketing, Intuit has acknowledged generative AI performs poorly at math and avoids using it for TurboTax calculations.

The core risk is accountability: taxpayers remain legally responsible for errors, while AI tools sit in a regulatory gap with no professional liability, standards, or required disclosures. Privacy and discoverability risks also rise after a 2026 ruling in U.S. v. Heppner denying privilege for AI-generated legal work.

The IRS warns against AI reliance yet uses AI internally, contrasting with the EU AI Act’s high-risk compliance framework.

A Manchester woman choosing a care home finds templated, glowing reviews across sites and is told by an inspector to visit in person, illustrating how AI-generated content is hollowing out everyday trust signals.

Cited reports describe an inflection point: industry estimates that up to 90% of online content will be AI-generated by end of 2026, Yelp filtering nearly 500,000 suspected AI reviews and closing 1.3 million accounts, and research showing AI-disclosure labels can prompt disengagement rather than scrutiny.

We explore why detection fails (high false negatives and false positives, with asymmetrical costs favouring generators) and critique probabilistic labeling. It surveys provenance via C2PA and why metadata often gets stripped, contrasts EU/UK/US regulatory approaches, and argues for shifting from labeling to provenance and liability, plus verified-purchase reviews and stronger institutions like inspectorates to restore accountable trust.

The UK Home Office confirmed an April 28, 2026 trial of AI facial age estimation for Channel arrivals, presented as a response to longstanding failures in border age decisions, despite criticism from Human Rights Watch and Right to Remain and a contemporaneous legal opinion warning existing Home Office AI asylum tools may already be unlawful.

Deep-learning age estimators’ average error rates conceal wide tails that matter when drawing an 18-year boundary, and how demographic and environmental factors: youth, darker skin, poor lighting, trauma, malnutrition, and sea-crossing conditions, can worsen performance, with no public evidence of validation on a population resembling Channel arrivals. It contrasts AI outputs’ anchoring effect with Merton-compliant social-worker assessments and argues transparency, equality assessment, contestability, and governance standards are missing, making “objectivity” an appearance that displaces due process and child protection.