The Dark Side of AI — Statistics Index
The data layer behind AI misuse: deepfake fraud, AI-enabled cybercrime, synthetic abuse, and social-engineering — tracked year over year, with the trend and the forecast. Every figure is primary-sourced and dated.
Thank you for reading this post, don't forget to subscribe!
Looking for the narrative write-ups? See the Dark Side of AI report series.
documented AI incidents in 2025 (+55% YoY).
Stanford HAI →US consumer fraud losses in 2024 (+25% YoY).
FTC →of phishing emails used AI in 2025 (+53.5% YoY).
2025 industry report →projected US gen-AI fraud by 2027, from $12.3B in 2023.
Deloitte →Year over year: the core series
Four independent datasets, each measuring a different slice of AI misuse. Read trackers of detected incidents (e.g. deepfake fraud) as a blend of real growth and improving detection — see the methodology note.
1 · Documented AI incidents
| Year | Incidents | Change |
|---|---|---|
| 2012 | <10 | baseline |
| 2024 | 233 | — |
| 2025 | 362 | ▲ +55% |
Source: AI Incident Database, via Stanford HAI AI Index 2026 (Apr 2026).
2 · Deepfake fraud incidents (detected)
| Period | Incidents | Change |
|---|---|---|
| 2017–2022 (cumulative) | 22 | baseline |
| 2023 | 42 | ▲ ~2× |
| 2024 | 150 | ▲ ~4× |
| Q1 2025 | 179 | ▲ >100% of 2024 in one quarter |
Source: Sumsub / Security.org deepfake incident tracking (2024–2025). Counts reflect detected incidents.
3 · US consumer fraud losses (all methods)
| Year | Total fraud losses | Imposter scams |
|---|---|---|
| 2022 | $8.8B | $2.6B |
| 2023 | ~$10.0B | $2.7B |
| 2024 | $12.5B | $2.95B |
Source: FTC Consumer Sentinel Network Data Books (2023–2025). Total fraud rose +25% in 2024. Not all fraud is AI-enabled — this is the backdrop AI is accelerating.
4 · AI in social engineering
| Metric | Figure | Window |
|---|---|---|
| Phishing emails using AI | 82.6% | 2025 (+53.5% YoY) |
| Phishing attacks using AI | 67.4% | 2024 |
| Gen-AI-linked phishing growth | +1,265% | since 2023 |
| Voice phishing (vishing) | +442% | 2023→2024 |
| Deepfake phishing | +310% | 2023→2025 |
Sources: industry phishing-threat reports (2025–2026), incl. Security Magazine, VIPRE/KnowBe4-class telemetry. Methodologies vary by vendor.
5 · Synthetic sexual abuse (NCII)
| Metric | Figure | Window |
|---|---|---|
| Deepfakes that are nonconsensual sexual imagery | ~96% | 2018, ~steady to 2023 |
| Visits to 16 “nudify” sites | 200M | H1 2024 |
| Nudify-app users in a single month | 24M+ | 2024 |
Sources: Sensity AI; Institute for Strategic Dialogue; Graphika. Full write-up in the NCII report.
The trend
Every series above points the same direction: up, and accelerating. Documented AI incidents rose 55% in a single year. Detected deepfake-fraud attempts roughly doubled in 2023, quadrupled in 2024, and cleared all of 2024’s total within the first quarter of 2025. AI is now the default tool in phishing — from a minority of attacks in 2024 to 82.6% of phishing emails in 2025. The common thread is a collapsing cost of production: deepfakes and machine-written lures that once needed skill and time are now cheap, fast, and abundant.
One honest caveat runs through the incident counts: they measure detected events, so improving detection inflates the curve alongside genuine growth. The direction is not in doubt; the precise slope is.
The forecast
To 2030: no single primary source publishes a credible point estimate for total AI-enabled crime in 2030. If Deloitte’s 32% CAGR held past 2027, US gen-AI fraud would roughly double again by 2030 — but we present that as an extrapolation, not a sourced figure, and flag it accordingly.
- Primary-source only. Each figure is attributed to the originating study, agency, or filing, with its year.
- Detected ≠ total. Incident trackers (AI Incident Database, Sumsub) count what’s found; true volumes are higher and the growth rate blends real increase with better detection.
- Mixed methodologies. Vendor phishing telemetry differs in how it labels “AI use”; treat cross-vendor comparisons as directional.
- Forecasts are labeled. Projections (Deloitte, Cybersecurity Ventures) are cited; our own extrapolations are marked as such and never presented as data.
- Freshness: Active — last reviewed June 2026.
Go deeper: the full series
These numbers feed the narrative Dark Side of AI report series — 16 sourced reports covering deepfake fraud, AI cybercrime, algorithmic bias, surveillance, copyright, market manipulation, frontier risk, and the LLM & agent failure modes behind them:
- The Dark Side of AI in Numbers
- Deepfake Fraud: The $25-Million Heist Era
- AI Deepfakes & Fraud Statistics 2026
- AI in Cybercrime: WormGPT to Phishing
- When Chatbots Fail: Liability
- Synthetic Deception: Elections & Voice Scams
- AI-Generated Sexual Abuse (NCII)
- AI Companions & Mental Health
- Algorithmic Bias at Scale
- AI Slop: The Synthetic Flood
- Copyright and AI: Who Owns the Rights?
- AI Surveillance & the Panopticon
- AI Market Manipulation
- Frontier Risks: AI & Bioweapons
- The Dark Side of LLMs & AI Agents
- Daniel Kokotajlo: AI Risk Predictions
Sources
- Stanford HAI — AI Index 2026 (AI Incident Database), April 2026
- FTC — Consumer Sentinel Network Data Books, 2022–2024 (released 2023–2025)
- Deloitte Center for Financial Services — generative-AI fraud projection, 2024
- Sumsub / Security.org — deepfake incident tracking, 2024–2025
- Security Magazine and industry phishing-threat reports, 2025–2026
- Sensity AI; Institute for Strategic Dialogue; Cybersecurity Ventures