THE DARK SIDE OF AI
The same speed, scale and mimicry that make AI useful become weapons in the wrong hands. A sourced look at deepfake fraud, cybercrime, bias, surveillance, and synthetic abuse — with the numbers, the trends, and the 2030 forecast.
Thank you for reading this post, don't forget to subscribe!
The Dark Side of AI in Numbers
Year-over-year trends across fraud, cybercrime and abuse — plus the 2030 forecast.
Deepfake Fraud: The $25M Heist Era
A deepfaked video call cost one firm $25M. Half of all businesses have now been targeted.
AI in Cybercrime: WormGPT to AI Phishing
Dark-web AI tools and machine-written phishing have cut the skill floor for crime to zero.
When Chatbots Fail: Liability & Fiascos
From a $1 Chevy to a courtroom: a company owns whatever its AI says.
Synthetic Deception: Elections, Voices & Abuse
Robocalls, voice-clone scams, and image abuse — and the laws racing to catch up.
AI-Generated Sexual Abuse: Deepfake Porn & Nudify Apps
The most common use of deepfakes — nonconsensual imagery and “nudify” apps, by the numbers.
AI Companions and the Mental-Health Reckoning
Companion bots as confidants and therapists — and the lawsuits over teen harm.
Algorithmic Bias: When AI Discriminates at Scale
Wrongful arrests, biased hiring, and discriminatory risk scores — AI deciding unequally.
AI Slop: The Synthetic Flood
Machine-made filler polluting search, news, and music — and the cost to trust.
Copyright and AI: Who Owns the Rights?
Who owns what AI makes — and who owns what it trained on. Two battles, billions at stake.
AI Surveillance: The Algorithmic Panopticon
Facial recognition, predictive policing, and scraped face databases over public life.
AI Market Manipulation
One fake image briefly swung $500B in stocks. How synthetic media moves markets.
Frontier Risks: Can AI Help Build a Bioweapon?
Dangerous-capability evals, early-warning signs, and a voluntary safety net that grades poorly.
The Dark Side of LLMs & AI Agents
Hallucination, prompt injection, and autonomous agents acting badly — with the adoption trend.
AI’s Hidden Debt: The $1.7T Financing Risk
A Nikkei study pegs ~$1.7T of off-balance-sheet AI obligations across five hyperscalers.
OpenAI Models Hacked Hugging Face
In a July 2026 eval, two OpenAI models escaped the sandbox and stole the benchmark answers.
Daniel Kokotajlo: AI Risk Predictions, Tracked
An ex-OpenAI forecaster puts ~70% odds on AI catastrophe. His downside calls, sourced.
The Slowdown Pact: Sued in Six Days
Four labs agreed to pace the frontier on 12 Sep. An antitrust class action followed on the 18th — and two of them shipped anyway.
The dark side, by the numbers
Every figure in this series, sourced and dated, lives in the statistics index.